mnist_aip.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "mnist-aip.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "L4pvKb8AQisu",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 68
        },
        "outputId": "c88abb46-0812-4b17-a1a7-827bfe02633f"
      },
      "source": [
        "from keras.datasets import mnist\n",
        "(x_train_o, y_train), (x_test_o, y_test) = mnist.load_data()\n",
        "x_train_o=x_train_o.astype(float)\n",
        "x_test_o=x_test_o.astype(float)"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://s3.amazonaws.com/img-datasets/mnist.npz\n",
            "11493376/11490434 [==============================] - 1s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LJtqPdiDi9aq",
        "colab_type": "code",
        "outputId": "119b1c20-6309-46ae-cf3e-7fc78868bf02",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        }
      },
      "source": [
        "print(x_train_o.shape)\n",
        "print(y_train.shape)\n",
        "print(x_test_o.shape)\n",
        "print(y_test.shape)"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(60000, 28, 28)\n",
            "(60000,)\n",
            "(10000, 28, 28)\n",
            "(10000,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1jx_iCwdg-nY",
        "colab_type": "code",
        "outputId": "3412ce65-dfaf-491d-d2ab-f258c041467b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "# print(np.sum(y_train==3))\n",
        "# l_tr=np.where(y_train==3)\n",
        "# l2_tr=np.where(y_train!=3)\n",
        "# print(y_train[l_tr][1:5])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "6131\n",
            "[3 3 3 3]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SJBs_QFXSJdO",
        "colab_type": "code",
        "outputId": "167e24a4-6a4e-46c4-a637-42c0f797a712",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 68
        }
      },
      "source": [
        "import numpy as np\n",
        "print(np.shape(x_train_o))\n",
        "print(np.shape(y_train))\n",
        "x_train = np.reshape(x_train_o, (60000, 784))\n",
        "print(np.shape(x_train))"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(60000, 28, 28)\n",
            "(60000,)\n",
            "(60000, 784)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Y6-TKfQOVydv",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from sklearn.utils import shuffle\n",
        "x_train, y_train = shuffle(x_train, y_train)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9he45UJDUyTH",
        "colab_type": "code",
        "outputId": "675782a1-5556-406f-c106-de5fbe64ee92",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 68
        }
      },
      "source": [
        "ns = 500\n",
        "l_tr=np.where(y_train==3)\n",
        "l2_tr=np.where(y_train!=3)\n",
        "x_train_d=np.array(x_train)\n",
        "y_train_d=np.array(y_train)\n",
        "print(y_train_d.shape)\n",
        "print(x_train_d.shape)\n",
        "x_train[0:ns//2][:] = x_train_d[l_tr[0][0:ns//2]][:]\n",
        "y_train[0:ns//2]= y_train_d[l_tr[0][0:ns//2]]\n",
        "x_train[ns//2:ns][:] = x_train_d[l2_tr[0][0:ns//2]][:]\n",
        "y_train[ns//2:ns]= y_train_d[l2_tr[0][0:ns//2]][:]\n",
        "print(np.sum(y_train==3))"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(60000,)\n",
            "(60000, 784)\n",
            "6321\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uCYPe8Xtqd1G",
        "colab_type": "code",
        "outputId": "1415908a-45e9-4de3-9dc1-2b85e4092014",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 272
        }
      },
      "source": [
        "y_train=y_train[0:ns]\n",
        "x_train=x_train[0:ns][:]\n",
        "\n",
        "y_train[0:ns//2]=1\n",
        "y_train[ns//2:ns]=0\n",
        "print(y_train)\n",
        "print(x_train.shape)"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
            " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
            " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
            " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
            " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
            " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1\n",
            " 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0\n",
            " 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n",
            " 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n",
            " 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n",
            " 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n",
            " 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n",
            " 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n",
            " 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]\n",
            "(500, 784)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "r-uJP0LAU-vq",
        "colab_type": "code",
        "outputId": "24dc064c-17d5-4f81-b546-b8952c62ed8a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "for x in np.unique(y_train):\n",
        "  print(np.shape(y_train[y_train==x])[0]/ns)"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.5\n",
            "0.5\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GDX5Jyv_0ctm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "x_train, y_train = shuffle(x_train, y_train)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "I47jiSqXRc-2",
        "colab_type": "code",
        "outputId": "15e3903e-5096-41cc-a77f-2af6186fc672",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        }
      },
      "source": [
        "from sklearn.svm import SVC\n",
        "from sklearn.model_selection import GridSearchCV, StratifiedKFold\n",
        "\n",
        "kfold = StratifiedKFold(n_splits=10)\n",
        "\n",
        "svclassifier = SVC()\n",
        "svc_param_grid = {'kernel': ['linear'],\n",
        "                  'C': [0.01, 0.1, 1, 10, 50, 100, 200, 300, 1000]}\n",
        "gsSVM = GridSearchCV(svclassifier,param_grid = svc_param_grid, cv=kfold,scoring=\"accuracy\", n_jobs= -1, verbose = 1)\n",
        "gsSVM.fit(x_train, y_train)\n",
        "\n",
        "SVM_best = gsSVM.best_estimator_\n",
        "gsSVM.best_score_"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  46 tasks      | elapsed:    3.7s\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    6.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.8619999999999999"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 100
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xwZZyeS8gV_Q",
        "colab_type": "code",
        "outputId": "53a4eb53-1b41-47a2-ed94-5c55300d9c48",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "train_data=np.sum(SVM_best.predict(x_train) == y_train)/ns\n",
        "print(train_data)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4nQVACDiXtfG",
        "colab_type": "code",
        "outputId": "8a272e08-201c-40da-a325-ac6b52d46acf",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        }
      },
      "source": [
        "\n",
        "print(np.shape(y_test))\n",
        "print(np.sum(y_test==3))\n",
        "x_test = np.reshape(x_test_o, (10000, 784))\n",
        "print(np.shape(x_test))\n",
        "y_test.shape"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(10000,)\n",
            "1010\n",
            "(10000, 784)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(10000,)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bLi1ojnWYI6E",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "nt = 2000\n",
        "l_tr=np.where(y_test==3)\n",
        "l2_tr=np.where(y_test!=3)\n",
        "x_test_d=np.array(x_test)\n",
        "y_test_d=np.array(y_test)\n",
        "x_test[0:nt//2][:] = x_test_d[l_tr[0][0:nt//2]][:]\n",
        "y_test[0:nt//2]=1\n",
        "x_test[nt//2:nt][:] = x_test_d[l2_tr[0][0:nt//2]][:]\n",
        "y_test[nt//2:nt]=0\n",
        "x_test = x_test[0:nt][:]\n",
        "y_test = y_test[0:nt]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dk17iGDG0gqZ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "x_test, y_test = shuffle(x_test, y_test)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mS0vQDhN2y1o",
        "colab_type": "code",
        "outputId": "02c830af-5ce8-4d5e-e49a-3ce219a5b8ed",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        }
      },
      "source": [
        "print(x_test.shape)\n",
        "print(y_test.shape)\n",
        "print(np.sum(y_test==0))\n",
        "print(np.sum(y_test==1))"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(2000, 784)\n",
            "(2000,)\n",
            "1000\n",
            "1000\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7h41m1eLW6Vw",
        "colab_type": "code",
        "outputId": "6ccbd46f-fd1f-4279-bce8-1145fa86371d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "test_data=np.sum(SVM_best.predict(x_test) == y_test)/nt\n",
        "print(test_data)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.871\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0Ldjota8ZeLz",
        "colab_type": "code",
        "outputId": "f786866c-d70c-4206-d3c7-2f2bcf4e8930",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "n = 784\n",
        "m = 200\n",
        "A = np.random.randn(m,n)\n",
        "B=np.random.binomial(1,0.5,(m,n))\n",
        "B[np.where(B==0)]=-1\n",
        "# x_train_c = np.matmul(x_train, A.T)/m**0.5\n",
        "# x_test_c = np.matmul(x_test, A.T)/m**0.5\n",
        "x_train_g = np.matmul(x_train, A.T)/m**0.5\n",
        "x_test_g = np.matmul(x_test, A.T)/m**0.5\n",
        "x_train_b = np.matmul(x_train, B.T)/m**0.5\n",
        "x_test_b = np.matmul(x_test, B.T)/m**0.5\n",
        "print(np.shape(x_train_b))\n",
        "print(np.shape(x_train_g))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(500, 200)\n",
            "(500, 200)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gxa0Qf0VbIcp",
        "colab_type": "code",
        "outputId": "3418831b-9d73-481a-afd8-734af619af77",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 153
        }
      },
      "source": [
        "from sklearn.svm import SVC\n",
        "from sklearn.model_selection import GridSearchCV, StratifiedKFold\n",
        "\n",
        "kfold = StratifiedKFold(n_splits=10)\n",
        "\n",
        "svclassifier = SVC()\n",
        "svc_param_grid = {'kernel': ['linear'],\n",
        "                  'C': [0.01, 0.1, 1, 10, 50, 100, 200, 300, 1000]}\n",
        "gsSVM_g = GridSearchCV(svclassifier,param_grid = svc_param_grid, cv=kfold, scoring=\"accuracy\", n_jobs= -1, verbose = 1)\n",
        "gsSVM_g.fit(x_train_g, y_train)\n",
        "\n",
        "SVM_best_g = gsSVM_g.best_estimator_\n",
        "print(gsSVM_g.best_score_)\n",
        "\n",
        "gsSVM_b = GridSearchCV(svclassifier,param_grid = svc_param_grid, cv=kfold, scoring=\"accuracy\", n_jobs= -1, verbose = 1)\n",
        "gsSVM_b.fit(x_train_b, y_train)\n",
        "\n",
        "SVM_best_b = gsSVM_b.best_estimator_\n",
        "print(gsSVM_b.best_score_)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8580000000000002\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n",
            "0.868\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.5s finished\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "x4sp5o6DguFU",
        "colab_type": "code",
        "outputId": "f89a3853-2e2f-4f64-d460-b01f1cd28434",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "train_measurement_gaussian=np.sum(SVM_best_g.predict(x_train_g) == y_train)/ns\n",
        "print(train_measurement_gaussian)\n",
        "train_measurement_bernoulli=np.sum(SVM_best_b.predict(x_train_b) == y_train)/ns\n",
        "print(train_measurement_bernoulli)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1.0\n",
            "1.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_0fsW8BUbZWF",
        "colab_type": "code",
        "outputId": "64ac0211-ab22-4e49-9b9e-7c4da2b9df43",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "test_measurement_gaussian=np.sum(SVM_best_g.predict(x_test_g) == y_test)/nt\n",
        "print(test_measurement_gaussian)\n",
        "test_measurement_bernoulli=np.sum(SVM_best_b.predict(x_test_b) == y_test)/nt\n",
        "print(test_measurement_bernoulli)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.852\n",
            "0.8475\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1E3cD0ZXdvZD",
        "colab_type": "code",
        "outputId": "0cb49771-678a-4b68-f967-b1e49e43acdb",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 119
        }
      },
      "source": [
        "print(\"train_data\", train_data)\n",
        "print(\"test_data\", test_data)\n",
        "print(\"train_measurement_gaussian\", train_measurement_gaussian)\n",
        "print(\"test_measurement_gaussian\", test_measurement_gaussian)\n",
        "print(\"train_measurement_bernoulli\", train_measurement_bernoulli)\n",
        "print(\"test_measurement_bernoulli\", test_measurement_bernoulli)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "train_data 1.0\n",
            "test_data 0.8745\n",
            "train_measurement_gaussian 1.0\n",
            "test_measurement_gaussian 0.852\n",
            "train_measurement_bernoulli 1.0\n",
            "test_measurement_bernoulli 0.8475\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3WUv75WFhAZU",
        "colab_type": "code",
        "outputId": "ce5c15c9-1988-46f7-91d5-b148855825de",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 255
        }
      },
      "source": [
        "print(y_train)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0 1 1 1 0 0 1 0 0 0 0 0 0 1 1 0 0 1 1 1 1 0 0 1 0 1 0 1 1 1 1 1 1 0 1 1 1\n",
            " 1 0 0 0 0 1 0 0 1 0 0 0 0 0 1 0 0 1 1 0 0 1 0 1 0 1 1 0 1 0 1 0 1 0 0 1 1\n",
            " 1 0 0 1 1 1 0 0 1 0 1 0 0 0 1 0 1 0 1 1 0 0 0 0 1 1 1 0 1 0 0 1 0 1 1 0 1\n",
            " 1 1 1 1 1 0 0 0 0 0 1 1 1 0 1 1 0 1 0 1 1 1 1 1 1 1 1 0 0 0 1 0 0 1 0 0 0\n",
            " 1 0 1 1 0 1 1 1 0 1 1 0 1 1 1 0 1 0 0 1 0 1 1 0 0 1 0 1 0 1 1 0 0 1 0 0 1\n",
            " 0 0 0 0 0 0 1 1 1 1 1 0 1 1 0 0 0 0 1 0 1 1 0 1 1 1 0 1 0 0 1 1 0 0 1 0 1\n",
            " 1 1 0 0 1 0 0 0 0 1 0 1 0 0 0 0 1 1 1 0 0 1 0 0 1 1 1 0 0 0 1 0 1 0 1 1 0\n",
            " 0 0 0 1 0 0 0 1 1 1 0 0 1 1 0 1 0 1 1 1 1 0 1 0 1 1 1 0 1 1 0 0 0 0 1 0 1\n",
            " 0 0 0 1 0 1 1 1 0 0 1 0 1 0 1 0 0 0 0 1 1 0 0 1 1 0 1 1 0 0 0 0 0 1 1 0 1\n",
            " 0 0 0 0 1 1 1 1 1 1 1 1 1 0 0 1 1 0 0 1 0 1 1 0 1 0 1 0 0 0 1 0 1 1 1 1 0\n",
            " 0 1 1 1 1 0 1 0 0 1 0 1 0 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 1 1 1 0 0 0 0 1 1\n",
            " 1 0 1 1 1 0 1 1 1 0 1 1 0 1 1 0 0 0 0 1 1 1 0 0 0 1 0 1 0 0 0 0 0 1 1 0 0\n",
            " 0 1 1 0 1 1 1 0 1 0 1 1 1 1 0 1 1 0 1 0 0 0 0 0 0 1 0 0 0 1 0 1 1 1 1 1 0\n",
            " 1 1 0 0 0 1 0 1 1 1 1 1 0 0 1 0 1 1 0]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8clMMHx9K2Ln",
        "colab_type": "code",
        "outputId": "065bd661-aa9c-4895-c058-fb8f61f6efc2",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "from sklearn.svm import SVC\n",
        "from sklearn.model_selection import GridSearchCV, StratifiedKFold\n",
        "n = 784\n",
        "test_acc_g=[]\n",
        "train_acc_g=[]\n",
        "test_acc_b=[]\n",
        "train_acc_b=[]\n",
        "meas=[]\n",
        "\n",
        "m = 200\n",
        "for m in list(range(800,5,-20)):\n",
        "  A = np.random.randn(m,n)\n",
        "  B=np.random.binomial(1,0.5,(m,n))\n",
        "  B[np.where(B==0)]=-1\n",
        "  # x_train_c = np.matmul(x_train, A.T)/m**0.5\n",
        "  # x_test_c = np.matmul(x_test, A.T)/m**0.5\n",
        "  x_train_g = np.matmul(x_train, A.T)/m**0.5\n",
        "  x_test_g = np.matmul(x_test, A.T)/m**0.5\n",
        "  x_train_b = np.matmul(x_train, B.T)/m**0.5\n",
        "  x_test_b = np.matmul(x_test, B.T)/m**0.5\n",
        "  print(np.shape(x_train_b))\n",
        "  print(np.shape(x_train_g))\n",
        "  kfold = StratifiedKFold(n_splits=10)\n",
        "\n",
        "  svclassifier = SVC()\n",
        "  svc_param_grid = {'kernel': ['linear'],\n",
        "                    'C': [0.01, 0.1, 1, 10, 50, 100, 200, 300, 1000]}\n",
        "  gsSVM_g = GridSearchCV(svclassifier,param_grid = svc_param_grid, cv=kfold, scoring=\"accuracy\", n_jobs= -1, verbose = 1)\n",
        "  gsSVM_g.fit(x_train_g, y_train)\n",
        "\n",
        "  SVM_best_g = gsSVM_g.best_estimator_\n",
        "  print(gsSVM_g.best_score_)\n",
        "\n",
        "  gsSVM_b = GridSearchCV(svclassifier,param_grid = svc_param_grid, cv=kfold, scoring=\"accuracy\", n_jobs= -1, verbose = 1)\n",
        "  gsSVM_b.fit(x_train_b, y_train)\n",
        "\n",
        "  SVM_best_b = gsSVM_b.best_estimator_\n",
        "  print(gsSVM_b.best_score_)\n",
        "  print(m)\n",
        "  train_acc_g.append(np.sum(SVM_best_g.predict(x_train_g) == y_train)/ns)\n",
        "  test_acc_g.append(np.sum(SVM_best_g.predict(x_test_g) == y_test)/nt)\n",
        "  train_acc_b.append(np.sum(SVM_best_b.predict(x_train_b) == y_train)/ns)\n",
        "  test_acc_b.append(np.sum(SVM_best_b.predict(x_test_b) == y_test)/nt)\n",
        "  meas.append(m)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(500, 800)\n",
            "(500, 800)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    4.1s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.9s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9020000000000001\n",
            "800\n",
            "(500, 780)\n",
            "(500, 780)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.8s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8959999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.8s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9119999999999999\n",
            "780\n",
            "(500, 760)\n",
            "(500, 760)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.8s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.906\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.8s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8959999999999999\n",
            "760\n",
            "(500, 740)\n",
            "(500, 740)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  87 out of  90 | elapsed:    3.7s remaining:    0.1s\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.8s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9099999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.7s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9179999999999999\n",
            "740\n",
            "(500, 720)\n",
            "(500, 720)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  87 out of  90 | elapsed:    3.5s remaining:    0.1s\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.5s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8939999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.6s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.906\n",
            "720\n",
            "(500, 700)\n",
            "(500, 700)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.908\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.7s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9039999999999999\n",
            "700\n",
            "(500, 680)\n",
            "(500, 680)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9019999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.5s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9040000000000001\n",
            "680\n",
            "(500, 660)\n",
            "(500, 660)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.4s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9119999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9019999999999999\n",
            "660\n",
            "(500, 640)\n",
            "(500, 640)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.906\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.3s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.908\n",
            "640\n",
            "(500, 620)\n",
            "(500, 620)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.2s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.906\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.906\n",
            "620\n",
            "(500, 600)\n",
            "(500, 600)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.2s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.906\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.898\n",
            "600\n",
            "(500, 580)\n",
            "(500, 580)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.0s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.898\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.908\n",
            "580\n",
            "(500, 560)\n",
            "(500, 560)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    3.1s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9019999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.9s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9059999999999999\n",
            "560\n",
            "(500, 540)\n",
            "(500, 540)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.908\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.8s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9019999999999999\n",
            "540\n",
            "(500, 520)\n",
            "(500, 520)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8879999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.7s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9099999999999999\n",
            "520\n",
            "(500, 500)\n",
            "(500, 500)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.8s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.906\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.7s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.908\n",
            "500\n",
            "(500, 480)\n",
            "(500, 480)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.898\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.7s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9099999999999999\n",
            "480\n",
            "(500, 460)\n",
            "(500, 460)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.908\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.5s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9159999999999998\n",
            "460\n",
            "(500, 440)\n",
            "(500, 440)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.4s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9039999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8979999999999999\n",
            "440\n",
            "(500, 420)\n",
            "(500, 420)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8960000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8939999999999999\n",
            "420\n",
            "(500, 400)\n",
            "(500, 400)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.2s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9059999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8859999999999999\n",
            "400\n",
            "(500, 380)\n",
            "(500, 380)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.1s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9039999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.0s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8959999999999999\n",
            "380\n",
            "(500, 360)\n",
            "(500, 360)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.0s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9100000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9019999999999999\n",
            "360\n",
            "(500, 340)\n",
            "(500, 340)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.9s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.898\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.0s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8939999999999999\n",
            "340\n",
            "(500, 320)\n",
            "(500, 320)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.0s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8880000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.9s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.916\n",
            "320\n",
            "(500, 300)\n",
            "(500, 300)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.9s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8939999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.0s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8859999999999999\n",
            "300\n",
            "(500, 280)\n",
            "(500, 280)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8880000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.7s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9099999999999999\n",
            "280\n",
            "(500, 260)\n",
            "(500, 260)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9039999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.7s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8720000000000001\n",
            "260\n",
            "(500, 240)\n",
            "(500, 240)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8960000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.6s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.89\n",
            "240\n",
            "(500, 220)\n",
            "(500, 220)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8959999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.5s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.906\n",
            "220\n",
            "(500, 200)\n",
            "(500, 200)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.6s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.89\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.5s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.89\n",
            "200\n",
            "(500, 180)\n",
            "(500, 180)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.892\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8940000000000001\n",
            "180\n",
            "(500, 160)\n",
            "(500, 160)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8959999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8880000000000001\n",
            "160\n",
            "(500, 140)\n",
            "(500, 140)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8580000000000002\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8720000000000001\n",
            "140\n",
            "(500, 120)\n",
            "(500, 120)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.866\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.6s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8620000000000001\n",
            "120\n",
            "(500, 100)\n",
            "(500, 100)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8460000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8699999999999999\n",
            "100\n",
            "(500, 80)\n",
            "(500, 80)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HMgj1-1ET2n1",
        "colab_type": "code",
        "outputId": "96dee42c-67c9-4229-86a3-7b743b7050f8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 660
        }
      },
      "source": [
        "print(\"Stats for MNIST dataset\")\n",
        "print(\"No. of Training Samples = \",ns)\n",
        "print(\"No. of Testing Samples = \",nt)\n",
        "print(\"Dimensions of images 28*28\")\n",
        "print(\"Number of features in data domain (originally) = 784\")\n",
        "print(\"Train accuracy in original domain = \", train_data)\n",
        "print(\"Test accuracy in original domain = \", test_data)\n",
        "from matplotlib import pyplot as plt\n",
        "plt.plot(meas,test_acc_g,label='Gaussian')\n",
        "plt.plot(meas,test_acc_b,label='Bernoulli')\n",
        "plt.xlabel('Measurements')\n",
        "plt.ylabel('Test Accuracy')\n",
        "plt.legend()\n",
        "plt.show()\n",
        "plt.plot(meas,train_acc_g,label='Gaussian')\n",
        "plt.plot(meas,train_acc_b,label='Bernoulli')\n",
        "plt.xlabel('Measurements')\n",
        "plt.ylabel('Train Accuracy')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Stats for MNIST dataset\n",
            "No. of Training Samples =  500\n",
            "No. of Testing Samples =  2000\n",
            "Dimensions of images 28*28\n",
            "Number of features in data domain (originally) = 784\n",
            "Train accuracy in original domain =  1.0\n",
            "Test accuracy in original domain =  0.882\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9FzaQFuic_VI",
        "colab_type": "code",
        "outputId": "5e031800-c457-4cc8-d7af-785431dcf6f8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 122
        }
      },
      "source": [
        "print(train_acc_g)\n",
        "print(train_acc_b)\n",
        "print(test_acc_g)\n",
        "print(test_acc_b)\n",
        "print(meas)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]\n",
            "[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]\n",
            "[0.883, 0.876, 0.8805, 0.881, 0.887, 0.878, 0.876, 0.877, 0.874, 0.8765, 0.87, 0.874, 0.886, 0.8855, 0.8775, 0.876, 0.869, 0.872, 0.877, 0.87, 0.8755, 0.8765, 0.871, 0.88, 0.8675, 0.8635, 0.879, 0.8535, 0.861, 0.8585, 0.8595, 0.8675, 0.8485, 0.8345, 0.8115, 0.8505]\n",
            "[0.891, 0.8815, 0.883, 0.88, 0.8845, 0.8695, 0.877, 0.8765, 0.8765, 0.8725, 0.8755, 0.877, 0.8755, 0.871, 0.8775, 0.878, 0.8845, 0.8815, 0.886, 0.88, 0.8635, 0.884, 0.8725, 0.868, 0.876, 0.8795, 0.8745, 0.873, 0.8625, 0.8745, 0.8635, 0.868, 0.8635, 0.842, 0.8555, 0.838]\n",
            "[800, 780, 760, 740, 720, 700, 680, 660, 640, 620, 600, 580, 560, 540, 520, 500, 480, 460, 440, 420, 400, 380, 360, 340, 320, 300, 280, 260, 240, 220, 200, 180, 160, 140, 120, 100]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IFiLNSuqOdfb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# from sklearn.preprocessing import StandardScaler\n",
        "# # e=np.identity(784)\n",
        "# # x_train=np.matmul(x_train,e)\n",
        "# # x_train_s = StandardScaler().fit_transform(x_train)\n",
        "# x_train_s = x_train\n",
        "# print(x_train_s.shape)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PHfvY6emQACN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# from sklearn.decomposition import PCA\n",
        "# pca = PCA(n_components=100)\n",
        "# p1 = pca.fit(x_train_s)\n",
        "# print(pca.explained_variance_ratio_)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yXWmOlSSQ2mO",
        "colab_type": "code",
        "outputId": "892b568f-bbd9-42ea-bf95-fcb94e0dba3a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 323
        }
      },
      "source": [
        "# pca2 = PCA(100)\n",
        "# pca2.fit(x_train_s)\n",
        "# print(pca2.explained_variance_ratio_)\n",
        "# print(np.sum(pca2.explained_variance_ratio_))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0.1187872  0.07602445 0.06104695 0.05761873 0.04419068 0.03726334\n",
            " 0.03101868 0.03000281 0.02844271 0.02411497 0.02143465 0.01972672\n",
            " 0.01903912 0.01706615 0.01564088 0.01525045 0.01465126 0.01345674\n",
            " 0.0127508  0.01180023 0.01139071 0.01082283 0.00931059 0.00913229\n",
            " 0.00862724 0.00840951 0.00788485 0.00763385 0.00701885 0.00689526\n",
            " 0.006626   0.00637509 0.00627922 0.00603032 0.00588726 0.00552375\n",
            " 0.00533307 0.00516881 0.00495445 0.0046778  0.0044384  0.00421031\n",
            " 0.00409538 0.0038901  0.00378071 0.00370007 0.00359166 0.00357035\n",
            " 0.0033973  0.0033085  0.00324312 0.00303653 0.00288717 0.0028718\n",
            " 0.00281096 0.00277219 0.00264129 0.00249072 0.00243559 0.00235564\n",
            " 0.00228506 0.00221647 0.00220177 0.00213945 0.00210565 0.00201262\n",
            " 0.00200479 0.00193307 0.00190787 0.00184419 0.0017829  0.00173486\n",
            " 0.00169417 0.00166654 0.00162604 0.00161814 0.00157097 0.00154003\n",
            " 0.00150493 0.00141698 0.00140607 0.00136533 0.00134852 0.00131496\n",
            " 0.00128084 0.00127241 0.00124901 0.00122073 0.00120059 0.00116977\n",
            " 0.00114454 0.00111074 0.00107388 0.0010556  0.00101676 0.00099729\n",
            " 0.0009776  0.00093873 0.00093168 0.00091045]\n",
            "0.9386290921772807\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TP_z5RXnQU7w",
        "colab_type": "code",
        "outputId": "873d87af-de1a-4479-8e98-4bb216e3d41c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 323
        }
      },
      "source": [
        "# print(pca2.explained_variance_ratio_)\n",
        "# print(pca2.explained_variance_ratio_.shape)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0.1187872  0.07602445 0.06104695 0.05761873 0.04419068 0.03726334\n",
            " 0.03101868 0.03000281 0.02844271 0.02411497 0.02143465 0.01972672\n",
            " 0.01903912 0.01706615 0.01564088 0.01525045 0.01465126 0.01345674\n",
            " 0.0127508  0.01180023 0.01139071 0.01082283 0.00931059 0.00913229\n",
            " 0.00862724 0.00840951 0.00788485 0.00763385 0.00701885 0.00689526\n",
            " 0.006626   0.00637509 0.00627922 0.00603032 0.00588726 0.00552375\n",
            " 0.00533307 0.00516881 0.00495445 0.0046778  0.0044384  0.00421031\n",
            " 0.00409538 0.0038901  0.00378071 0.00370007 0.00359166 0.00357035\n",
            " 0.0033973  0.0033085  0.00324312 0.00303653 0.00288717 0.0028718\n",
            " 0.00281096 0.00277219 0.00264129 0.00249072 0.00243559 0.00235564\n",
            " 0.00228506 0.00221647 0.00220177 0.00213945 0.00210565 0.00201262\n",
            " 0.00200479 0.00193307 0.00190787 0.00184419 0.0017829  0.00173486\n",
            " 0.00169417 0.00166654 0.00162604 0.00161814 0.00157097 0.00154003\n",
            " 0.00150493 0.00141698 0.00140607 0.00136533 0.00134852 0.00131496\n",
            " 0.00128084 0.00127241 0.00124901 0.00122073 0.00120059 0.00116977\n",
            " 0.00114454 0.00111074 0.00107388 0.0010556  0.00101676 0.00099729\n",
            " 0.0009776  0.00093873 0.00093168 0.00091045]\n",
            "(100,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dxfmj0hFRiIj",
        "colab_type": "code",
        "outputId": "2bb386da-af2a-4152-8aa9-ba681f93d2c5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 340
        }
      },
      "source": [
        "# # x_train_g_s = StandardScaler().fit_transform(x_train_g)\n",
        "# x_train_g_s = x_train_g\n",
        "# pca3 = PCA(100)\n",
        "# pca3.fit(x_train_g_s)\n",
        "# print(pca3.explained_variance_ratio_)\n",
        "# print(np.sum(pca3.explained_variance_ratio_))\n",
        "# print(pca3.explained_variance_ratio_.shape)\n",
        "# print(pca3.transform(x_train_g_s))\n",
        "# print(pca3.transform(x_train_g_s).shape)\n",
        "# p3=pca3.transform(x_train_g_s)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0.10136557 0.08884812 0.07133839 0.060348   0.0489301  0.041918\n",
            " 0.04049491 0.03210349 0.02946936 0.02528267 0.02274691 0.02173362\n",
            " 0.02010017 0.01784548 0.01749369 0.0163638  0.01445538 0.01313563\n",
            " 0.0130017  0.01290844 0.01181681 0.01129066 0.01066643 0.00992211\n",
            " 0.00939147 0.00878199 0.00842141 0.00789705 0.0073277  0.00723026\n",
            " 0.00693794 0.0065028  0.00639558 0.00627498 0.00590695 0.00577146\n",
            " 0.00532779 0.00517917 0.00495064 0.0045746  0.00448405 0.00427659\n",
            " 0.00407023 0.00396634 0.00388147 0.00365967 0.00351846 0.00340114\n",
            " 0.00330532 0.00307051 0.00301261 0.00289087 0.00279516 0.00271143\n",
            " 0.00257429 0.00242653 0.00238336 0.00226191 0.00222555 0.00217695\n",
            " 0.00202833 0.00199503 0.00196973 0.00192673 0.00188594 0.00186235\n",
            " 0.00184918 0.00180246 0.00170594 0.00163601 0.00159782 0.00155834\n",
            " 0.00153391 0.00150529 0.00145637 0.00139384 0.00135056 0.00134569\n",
            " 0.00128714 0.00124701 0.00120438 0.00118428 0.00113874 0.00112914\n",
            " 0.00108647 0.00107133 0.00103691 0.00100318 0.00095671 0.00094717\n",
            " 0.00091239 0.0008824  0.000841   0.00082065 0.00080068 0.00078522\n",
            " 0.00077276 0.00076253 0.00075778 0.0007357 ]\n",
            "0.9753107226318692\n",
            "(100,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vz3I9QcIS2os",
        "colab_type": "code",
        "outputId": "b5bd9f19-634b-48de-baa0-39922f468073",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 255
        }
      },
      "source": [
        "# print(pca2.transform(x_train_s))\n",
        "# p2=pca2.transform(x_train_s)\n",
        "# print(pca2.transform(x_train_s).shape)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[   93.77549469  -112.96252599  -694.95271395 ...    12.47921129\n",
            "    -71.97629731    26.81630453]\n",
            " [   48.00456725  -295.43262103   759.75228033 ...     7.09205976\n",
            "    191.37987836    99.93252753]\n",
            " [ -387.93404847  -375.40029327   463.89934244 ...   -51.29385901\n",
            "     62.15847244   117.75716271]\n",
            " ...\n",
            " [  352.45649736   189.5021722    479.92515627 ...  -153.3619427\n",
            "    -52.63169093  -115.56540577]\n",
            " [-1047.34799139   289.86889345   351.48319449 ...   -29.47659195\n",
            "     25.63104309   -55.3834367 ]\n",
            " [  731.38837885  -698.63698837   384.80718565 ...   -12.17517292\n",
            "     -9.78134263    -8.03557871]]\n",
            "(500, 100)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wmih-YbVT6yw",
        "colab_type": "code",
        "outputId": "94bee4f6-f6bd-47d3-a2e1-91721a0d27ad",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 255
        }
      },
      "source": [
        "# print(pca2.components_)\n",
        "# print(pca2.components_.shape)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[ 1.17120295e-17  6.17823562e-18 -1.80831046e-18 ...  0.00000000e+00\n",
            "   0.00000000e+00  0.00000000e+00]\n",
            " [-1.25816057e-16 -7.17142133e-17  1.70635117e-17 ... -0.00000000e+00\n",
            "  -0.00000000e+00 -0.00000000e+00]\n",
            " [-5.72977030e-17  7.49827322e-17 -2.81390298e-18 ...  0.00000000e+00\n",
            "   0.00000000e+00  0.00000000e+00]\n",
            " ...\n",
            " [ 5.14368503e-15 -5.00147208e-15  2.86671485e-15 ... -0.00000000e+00\n",
            "  -0.00000000e+00 -0.00000000e+00]\n",
            " [-1.47627528e-15  6.92090998e-16  2.95727361e-15 ... -0.00000000e+00\n",
            "  -0.00000000e+00 -0.00000000e+00]\n",
            " [ 6.57209443e-15  3.81986418e-16  5.89704168e-15 ...  0.00000000e+00\n",
            "   0.00000000e+00  0.00000000e+00]]\n",
            "(100, 784)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ER9-iAqvUnun",
        "colab_type": "code",
        "outputId": "f3f83891-9096-48a4-fa73-b5f6377977d5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 255
        }
      },
      "source": [
        "# print(pca3.components_)\n",
        "# print(pca3.components_.shape)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[ 0.01051753 -0.01433448  0.07061212 ...  0.04902194  0.00973866\n",
            "   0.0468803 ]\n",
            " [-0.01355811  0.01370842 -0.02664255 ... -0.01389386 -0.00152329\n",
            "  -0.02741747]\n",
            " [ 0.1222946   0.00218178  0.06322695 ... -0.06567156 -0.08421939\n",
            "  -0.07447478]\n",
            " ...\n",
            " [ 0.01541889  0.03329524  0.0481359  ...  0.04748982 -0.09662783\n",
            "   0.0333932 ]\n",
            " [-0.04274338  0.01112188 -0.02250073 ... -0.02027999 -0.05215625\n",
            "   0.00921714]\n",
            " [-0.08440445  0.11669689 -0.0234734  ...  0.02056057 -0.03041587\n",
            "  -0.02011582]]\n",
            "(100, 200)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1FQbJBo7U_GK",
        "colab_type": "code",
        "outputId": "9c91ef04-76b9-4a8d-b2d2-68e07c60ba8a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 255
        }
      },
      "source": [
        "# print(A.shape)\n",
        "# print(np.matmul(pca2.components_, A.T)/m**0.5)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(200, 784)\n",
            "[[-0.01345875 -0.02153387  0.05062122 ...  0.0578118   0.02765047\n",
            "   0.06406866]\n",
            " [-0.04845799  0.02559786 -0.02369851 ...  0.00974163  0.0077139\n",
            "  -0.00992022]\n",
            " [ 0.11148777 -0.01638759  0.07989597 ... -0.0600941  -0.09330948\n",
            "  -0.11067085]\n",
            " ...\n",
            " [-0.04981458  0.00569272 -0.15470102 ... -0.14737023  0.01685272\n",
            "   0.08455821]\n",
            " [-0.02512094  0.00434216  0.01717218 ... -0.01074811 -0.08361906\n",
            "  -0.00960689]\n",
            " [-0.07544536 -0.01026065  0.04494983 ...  0.03091999  0.05464291\n",
            "  -0.0567323 ]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3eU4lL4FYT-b",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u-N6rXQrVsR2",
        "colab_type": "code",
        "outputId": "20eb49d8-1259-4905-9b01-759824f403f9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "# exp=np.matmul(pca2.components_, A.T)/m**0.5\n",
        "# # exp=exp/np.linalg.norm(exp, ord=2, axis=1, keepdims=True)\n",
        "# print(exp/np.linalg.norm(exp, ord=2, axis=1, keepdims=True))\n",
        "# print(np.linalg.norm(exp, ord=2, axis=1, keepdims=True))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[-0.01511097 -0.0241774   0.05683556 ...  0.06490887  0.03104489\n",
            "   0.07193383]\n",
            " [-0.0465449   0.02458728 -0.02276291 ...  0.00935704  0.00740936\n",
            "  -0.00952858]\n",
            " [ 0.10791063 -0.01586179  0.07733246 ... -0.05816595 -0.0903156\n",
            "  -0.10711992]\n",
            " ...\n",
            " [-0.04756956  0.00543617 -0.14772903 ... -0.14072862  0.0160932\n",
            "   0.08074738]\n",
            " [-0.02427694  0.00419628  0.01659524 ... -0.010387   -0.08080966\n",
            "  -0.00928412]\n",
            " [-0.07523401 -0.01023191  0.04482392 ...  0.03083337  0.05448984\n",
            "  -0.05657337]]\n",
            "[[0.89066105]\n",
            " [1.04110196]\n",
            " [1.03314911]\n",
            " [0.95616669]\n",
            " [1.05842841]\n",
            " [0.9974882 ]\n",
            " [1.06347892]\n",
            " [1.0539605 ]\n",
            " [1.01405241]\n",
            " [0.92092341]\n",
            " [0.9631496 ]\n",
            " [0.94994901]\n",
            " [0.9596576 ]\n",
            " [0.96558902]\n",
            " [1.01608326]\n",
            " [1.02895515]\n",
            " [1.01430719]\n",
            " [1.01786808]\n",
            " [0.9797709 ]\n",
            " [1.0244365 ]\n",
            " [0.9670145 ]\n",
            " [0.93795286]\n",
            " [1.04037827]\n",
            " [0.91866948]\n",
            " [1.04167781]\n",
            " [1.00073001]\n",
            " [1.0721994 ]\n",
            " [1.04342717]\n",
            " [0.98916638]\n",
            " [0.99818479]\n",
            " [0.96461871]\n",
            " [1.03204839]\n",
            " [1.0423364 ]\n",
            " [0.93394599]\n",
            " [1.04266691]\n",
            " [1.01492423]\n",
            " [0.9853859 ]\n",
            " [1.05398094]\n",
            " [0.98771309]\n",
            " [1.00284707]\n",
            " [0.94631697]\n",
            " [0.99431612]\n",
            " [1.01888789]\n",
            " [1.03212863]\n",
            " [1.03947869]\n",
            " [1.05699164]\n",
            " [1.00812787]\n",
            " [1.05174212]\n",
            " [0.93716147]\n",
            " [0.93247187]\n",
            " [1.04729903]\n",
            " [1.0016021 ]\n",
            " [1.06521248]\n",
            " [1.02194362]\n",
            " [0.94661726]\n",
            " [0.85979046]\n",
            " [1.1027194 ]\n",
            " [1.04565134]\n",
            " [0.95704789]\n",
            " [0.87553278]\n",
            " [0.92188271]\n",
            " [1.03274321]\n",
            " [0.98356026]\n",
            " [0.96308478]\n",
            " [0.97511923]\n",
            " [0.99188704]\n",
            " [0.91987426]\n",
            " [1.01774358]\n",
            " [0.9552449 ]\n",
            " [0.94825168]\n",
            " [0.9582941 ]\n",
            " [1.00979451]\n",
            " [1.03453907]\n",
            " [0.97809902]\n",
            " [1.0968468 ]\n",
            " [1.01002907]\n",
            " [0.93391361]\n",
            " [0.98491417]\n",
            " [1.05447775]\n",
            " [0.91142263]\n",
            " [0.9707721 ]\n",
            " [1.01634586]\n",
            " [0.86578571]\n",
            " [1.03294612]\n",
            " [1.0187463 ]\n",
            " [1.05680737]\n",
            " [0.97852005]\n",
            " [0.99154647]\n",
            " [1.0080611 ]\n",
            " [0.95912828]\n",
            " [0.93642488]\n",
            " [1.07312913]\n",
            " [0.95912937]\n",
            " [1.02055513]\n",
            " [1.10690133]\n",
            " [0.98408752]\n",
            " [1.00690179]\n",
            " [1.04719447]\n",
            " [1.03476554]\n",
            " [1.00280915]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NV3LSGCaY4IA",
        "colab_type": "code",
        "outputId": "18cf0eee-ac0a-479f-ed04-f63e0dcbb43a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "# d=np.sqrt(np.sum(((exp - pca3.components_) ** 2),axis=1,keepdims=True)/np.sum(((pca3.components_) ** 2),axis=1,keepdims=True))\n",
        "# print(d.shape)\n",
        "# print(d)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(100, 1)\n",
            "[[0.34535823]\n",
            " [0.44159351]\n",
            " [0.53885665]\n",
            " [1.86192442]\n",
            " [1.95700426]\n",
            " [1.64145775]\n",
            " [0.9408333 ]\n",
            " [0.91136135]\n",
            " [1.78380266]\n",
            " [0.83014942]\n",
            " [1.52298266]\n",
            " [1.59562985]\n",
            " [1.20114571]\n",
            " [1.47951832]\n",
            " [1.56232117]\n",
            " [1.74121068]\n",
            " [1.47785174]\n",
            " [1.32436517]\n",
            " [1.22674169]\n",
            " [1.58625888]\n",
            " [1.26835963]\n",
            " [1.48760292]\n",
            " [1.10423861]\n",
            " [1.45397638]\n",
            " [1.51617427]\n",
            " [1.67760834]\n",
            " [1.43454186]\n",
            " [1.44878009]\n",
            " [1.65321201]\n",
            " [1.26220403]\n",
            " [1.43011179]\n",
            " [1.44668894]\n",
            " [1.56913959]\n",
            " [1.36883261]\n",
            " [1.51500926]\n",
            " [1.38650727]\n",
            " [1.67515939]\n",
            " [1.58810757]\n",
            " [1.65658987]\n",
            " [1.29895683]\n",
            " [1.40785653]\n",
            " [1.29624004]\n",
            " [1.49636938]\n",
            " [1.26006193]\n",
            " [1.3497903 ]\n",
            " [1.45677536]\n",
            " [1.39499108]\n",
            " [1.2058093 ]\n",
            " [1.5190142 ]\n",
            " [1.44944069]\n",
            " [1.35946226]\n",
            " [1.40487394]\n",
            " [1.50178539]\n",
            " [1.36083041]\n",
            " [1.63341816]\n",
            " [1.38529757]\n",
            " [1.51475026]\n",
            " [1.33061555]\n",
            " [1.30353721]\n",
            " [1.2419419 ]\n",
            " [1.46113135]\n",
            " [1.47456432]\n",
            " [1.42264636]\n",
            " [1.1451985 ]\n",
            " [1.39190877]\n",
            " [1.40227314]\n",
            " [1.25088813]\n",
            " [1.23576374]\n",
            " [1.30525957]\n",
            " [1.33241822]\n",
            " [1.1957398 ]\n",
            " [1.35993611]\n",
            " [1.37668723]\n",
            " [1.38373401]\n",
            " [1.57089249]\n",
            " [1.42174302]\n",
            " [1.33929265]\n",
            " [1.30059923]\n",
            " [1.36019275]\n",
            " [1.24188785]\n",
            " [1.23692147]\n",
            " [1.43981048]\n",
            " [1.3132175 ]\n",
            " [1.47888002]\n",
            " [1.38413176]\n",
            " [1.43954927]\n",
            " [1.39703239]\n",
            " [1.37205863]\n",
            " [1.42830203]\n",
            " [1.46504697]\n",
            " [1.41871863]\n",
            " [1.42711008]\n",
            " [1.32161276]\n",
            " [1.4694398 ]\n",
            " [1.57107837]\n",
            " [1.37498114]\n",
            " [1.55015674]\n",
            " [1.46089455]\n",
            " [1.40736608]\n",
            " [1.29269895]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xviRWdXbbKx_",
        "colab_type": "code",
        "outputId": "18d4e10a-02ff-4094-aee4-f2080eb62962",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 374
        }
      },
      "source": [
        "# print(np.matmul(x_train_g,x_train_g.T))\n",
        "# print()\n",
        "# print(np.matmul(x_train,x_train.T))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[4762719.97604734 2647520.14809971  935513.02561291 ... 1541642.89979657\n",
            "  2633514.99754573 1347046.60003925]\n",
            " [2647520.14809971 9788425.93711823 2772690.18882019 ... 2428364.53668376\n",
            "  4867962.39016634 2651072.49990036]\n",
            " [ 935513.02561291 2772690.18882019 3875996.16699548 ... 1069189.11023283\n",
            "  2623148.1857106  1337015.92699937]\n",
            " ...\n",
            " [1541642.89979657 2428364.53668376 1069189.11023283 ... 7387579.37591788\n",
            "  2057250.53372578 2994199.51053679]\n",
            " [2633514.99754573 4867962.39016634 2623148.1857106  ... 2057250.53372578\n",
            "  6596103.51977023 1562814.16331764]\n",
            " [1347046.60003925 2651072.49990036 1337015.92699937 ... 2994199.51053679\n",
            "  1562814.16331764 4613882.3777731 ]]\n",
            "\n",
            "[[4911926. 2964366. 1144113. ... 1584636. 2595095. 2239907.]\n",
            " [2964366. 9769138. 2914362. ... 2822156. 4985827. 3520947.]\n",
            " [1144113. 2914362. 3662527. ... 1163343. 2219314. 1503692.]\n",
            " ...\n",
            " [1584636. 2822156. 1163343. ... 6742444. 2128127. 2676895.]\n",
            " [2595095. 4985827. 2219314. ... 2128127. 6803373. 1594382.]\n",
            " [2239907. 3520947. 1503692. ... 2676895. 1594382. 5331041.]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6ZyWpHDwpxjo",
        "colab_type": "code",
        "outputId": "4654544f-ee0a-4853-caa1-9e6303388a2e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 187
        }
      },
      "source": [
        "# for i in list(range(0,10)):\n",
        "#   print(np.dot(exp[i,:],pca3.components_[i,:]) /(np.linalg.norm(exp[i,:]) * np.linalg.norm(pca3.components_[i,:])))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.9397541309163913\n",
            "0.9071582486685295\n",
            "0.8600068336743151\n",
            "-0.811839517645674\n",
            "-0.8076101098239998\n",
            "-0.3505809965687908\n",
            "0.585728633690942\n",
            "0.6073535085025124\n",
            "-0.5688313651896967\n",
            "0.629233577468679\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "C6ny1QpAslu0",
        "colab_type": "code",
        "outputId": "d3e26162-37e3-4152-9388-89c9e1d32806",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 187
        }
      },
      "source": [
        "# for i in list(range(0,10)):\n",
        "#   print(np.dot(p2[i,:],p3[i,:]) /(np.linalg.norm(p2[i,:]) * np.linalg.norm(p3[i,:])))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.07960057864787332\n",
            "-0.13478527358360667\n",
            "0.3546096386917526\n",
            "0.34177957795106206\n",
            "0.10966844560911572\n",
            "0.32151676794710987\n",
            "0.2050376623573918\n",
            "0.007009560314412689\n",
            "0.09439600807868054\n",
            "-0.05161731699520901\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BUlJEYJtt5z2",
        "colab_type": "code",
        "outputId": "72607be1-d38d-4f14-b96b-b91475bba5ed",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "# # print(p2.shape)\n",
        "# # print(p3.shape)\n",
        "# p2_test=pca2.transform(x_test)\n",
        "# p3_test=pca3.transform(x_test_g)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(500, 100)\n",
            "(500, 100)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gQGiPC3wueCV",
        "colab_type": "code",
        "outputId": "f2e83028-f7b9-4392-ed64-57f441e7a4eb",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 119
        }
      },
      "source": [
        "from sklearn.svm import SVC\n",
        "from sklearn.decomposition import PCA\n",
        "from sklearn.preprocessing import StandardScaler\n",
        "from sklearn.model_selection import GridSearchCV, StratifiedKFold\n",
        "\n",
        "kfold = StratifiedKFold(n_splits=10)\n",
        "x_train_s = x_train\n",
        "\n",
        "pca2 = PCA(100)\n",
        "pca2.fit(x_train_s)\n",
        "p2=pca2.transform(x_train_s)\n",
        "p2_test=pca2.transform(x_test)\n",
        "\n",
        "svclassifier = SVC()\n",
        "svc_param_grid = {'kernel': ['linear'],\n",
        "                    'C': [0.01, 0.1, 1, 10, 50, 100, 200, 300, 1000]}\n",
        "\n",
        "gsSVM_data = GridSearchCV(svclassifier,param_grid = svc_param_grid, cv=kfold, scoring=\"accuracy\", n_jobs= -1, verbose = 1)\n",
        "gsSVM_data.fit(p2, y_train)\n",
        "\n",
        "SVM_best_data = gsSVM_data.best_estimator_\n",
        "print(gsSVM_data.best_score_)\n",
        "# print(m)\n",
        "# print(\"measurement train \",np.sum(SVM_best_g.predict(p3) == y_train)/ns)\n",
        "# print(\"measurement test \",np.sum(SVM_best_g.predict(p3_test) == y_test)/nt)\n",
        "train_data=np.sum(SVM_best_data.predict(p2) == y_train)/ns\n",
        "test_data=np.sum(SVM_best_data.predict(p2_test) == y_test)/nt\n",
        "print(\"data train\",np.sum(SVM_best_data.predict(p2) == y_train)/ns)\n",
        "print(\"data test\",np.sum(SVM_best_data.predict(p2_test) == y_test)/nt)\n",
        "\n",
        "\n",
        "\n",
        "\n"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9039999999999999\n",
            "data train 1.0\n",
            "data test 0.875\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    2.1s finished\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NqGCrYUEiXcu",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "8a5f22e7-2e1a-402c-db29-84ce00fbefc3"
      },
      "source": [
        "n = 784\n",
        "test_acc_g=[]\n",
        "train_acc_g=[]\n",
        "test_acc_b=[]\n",
        "train_acc_b=[]\n",
        "meas=[]\n",
        "\n",
        "m = 200\n",
        "for m in list(range(800,100,-20)):\n",
        "  A = np.random.randn(m,n)\n",
        "  B=np.random.binomial(1,0.5,(m,n))\n",
        "  B[np.where(B==0)]=-1\n",
        "  # x_train_c = np.matmul(x_train, A.T)/m**0.5\n",
        "  # x_test_c = np.matmul(x_test, A.T)/m**0.5\n",
        "  x_train_g = np.matmul(x_train, A.T)/m**0.5\n",
        "  x_test_g = np.matmul(x_test, A.T)/m**0.5\n",
        "  x_train_b = np.matmul(x_train, B.T)/m**0.5\n",
        "  x_test_b = np.matmul(x_test, B.T)/m**0.5\n",
        "  # x_train_g_s = x_train_g\n",
        "  pca3 = PCA(100)\n",
        "  pca3.fit(x_train_g)\n",
        "  p3=pca3.transform(x_train_g)\n",
        "  p3_test=pca3.transform(x_test_g)\n",
        "  pca4 = PCA(100)\n",
        "  pca4.fit(x_train_b)\n",
        "  p4=pca4.transform(x_train_b)\n",
        "  p4_test=pca4.transform(x_test_b)\n",
        "  print(np.shape(x_train_b))\n",
        "  print(np.shape(x_train_g))\n",
        "  kfold = StratifiedKFold(n_splits=10)\n",
        "\n",
        "  svclassifier = SVC()\n",
        "  svc_param_grid = {'kernel': ['linear'],\n",
        "                    'C': [0.01, 0.1, 1, 10, 50, 100, 200, 300, 1000]}\n",
        "  gsSVM_g = GridSearchCV(svclassifier,param_grid = svc_param_grid, cv=kfold, scoring=\"accuracy\", n_jobs= -1, verbose = 1)\n",
        "  gsSVM_g.fit(p3, y_train)\n",
        "\n",
        "  SVM_best_g = gsSVM_g.best_estimator_\n",
        "  print(gsSVM_g.best_score_)\n",
        "\n",
        "  gsSVM_b = GridSearchCV(svclassifier,param_grid = svc_param_grid, cv=kfold, scoring=\"accuracy\", n_jobs= -1, verbose = 1)\n",
        "  gsSVM_b.fit(p4, y_train)\n",
        "\n",
        "  SVM_best_b = gsSVM_b.best_estimator_\n",
        "  print(gsSVM_b.best_score_)\n",
        "  print(m)\n",
        "  train_acc_g.append(np.sum(SVM_best_g.predict(p3) == y_train)/ns)\n",
        "  test_acc_g.append(np.sum(SVM_best_g.predict(p3_test) == y_test)/nt)\n",
        "  train_acc_b.append(np.sum(SVM_best_b.predict(p4) == y_train)/ns)\n",
        "  test_acc_b.append(np.sum(SVM_best_b.predict(p4_test) == y_test)/nt)\n",
        "  meas.append(m)"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(500, 800)\n",
            "(500, 800)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8939999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.884\n",
            "800\n",
            "(500, 780)\n",
            "(500, 780)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.884\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.882\n",
            "780\n",
            "(500, 760)\n",
            "(500, 760)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.898\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8880000000000001\n",
            "760\n",
            "(500, 740)\n",
            "(500, 740)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8800000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.884\n",
            "740\n",
            "(500, 720)\n",
            "(500, 720)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9120000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9039999999999999\n",
            "720\n",
            "(500, 700)\n",
            "(500, 700)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.892\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8800000000000001\n",
            "700\n",
            "(500, 680)\n",
            "(500, 680)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8939999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.892\n",
            "680\n",
            "(500, 660)\n",
            "(500, 660)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8999999999999998\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.892\n",
            "660\n",
            "(500, 640)\n",
            "(500, 640)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9120000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8960000000000001\n",
            "640\n",
            "(500, 620)\n",
            "(500, 620)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8780000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.882\n",
            "620\n",
            "(500, 600)\n",
            "(500, 600)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8800000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8699999999999999\n",
            "600\n",
            "(500, 580)\n",
            "(500, 580)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.0s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9039999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8779999999999999\n",
            "580\n",
            "(500, 560)\n",
            "(500, 560)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9\n",
            "560\n",
            "(500, 540)\n",
            "(500, 540)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.89\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.89\n",
            "540\n",
            "(500, 520)\n",
            "(500, 520)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.0s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8959999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8939999999999999\n",
            "520\n",
            "(500, 500)\n",
            "(500, 500)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8720000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.898\n",
            "500\n",
            "(500, 480)\n",
            "(500, 480)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.884\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.882\n",
            "480\n",
            "(500, 460)\n",
            "(500, 460)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.868\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8879999999999999\n",
            "460\n",
            "(500, 440)\n",
            "(500, 440)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8780000000000001\n",
            "440\n",
            "(500, 420)\n",
            "(500, 420)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8880000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.874\n",
            "420\n",
            "(500, 400)\n",
            "(500, 400)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.884\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8959999999999999\n",
            "400\n",
            "(500, 380)\n",
            "(500, 380)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8800000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8779999999999999\n",
            "380\n",
            "(500, 360)\n",
            "(500, 360)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8859999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.89\n",
            "360\n",
            "(500, 340)\n",
            "(500, 340)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.2s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8819999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.85\n",
            "340\n",
            "(500, 320)\n",
            "(500, 320)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.876\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9\n",
            "320\n",
            "(500, 300)\n",
            "(500, 300)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.868\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8780000000000001\n",
            "300\n",
            "(500, 280)\n",
            "(500, 280)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.0s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9019999999999999\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.882\n",
            "280\n",
            "(500, 260)\n",
            "(500, 260)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8540000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8960000000000001\n",
            "260\n",
            "(500, 240)\n",
            "(500, 240)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.9\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.1s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.876\n",
            "240\n",
            "(500, 220)\n",
            "(500, 220)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8860000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.7s finished\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8640000000000001\n",
            "220\n",
            "(500, 200)\n",
            "(500, 200)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n",
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.5s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.882\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.6s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8700000000000001\n",
            "200\n",
            "(500, 180)\n",
            "(500, 180)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.4s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8800000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8780000000000001\n",
            "180\n",
            "(500, 160)\n",
            "(500, 160)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.6s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8919999999999998\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.882\n",
            "160\n",
            "(500, 140)\n",
            "(500, 140)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8720000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.3s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8720000000000001\n",
            "140\n",
            "(500, 120)\n",
            "(500, 120)\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.7s finished\n",
            "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 2 concurrent workers.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.8720000000000001\n",
            "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n",
            "0.8720000000000001\n",
            "120\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:    1.5s finished\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SCJrijfphNyw",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 677
        },
        "outputId": "29fd9c01-3015-41a7-9851-a53fc0953fd0"
      },
      "source": [
        "print(\"Stats for MNIST dataset SVM using PCA with 100 components\")\n",
        "print(\"No. of Training Samples = \",ns)\n",
        "print(\"No. of Testing Samples = \",nt)\n",
        "print(\"Dimensions of images 28*28\")\n",
        "print(\"Number of features in data domain (originally) = 784\")\n",
        "print(\"Number of PCA components = 100 for both data domain and measurement domain \")\n",
        "print(\"Train accuracy in original domain = \", train_data)\n",
        "print(\"Test accuracy in original domain = \", test_data)\n",
        "from matplotlib import pyplot as plt\n",
        "plt.plot(meas,test_acc_g,label='Gaussian')\n",
        "plt.plot(meas,test_acc_b,label='Bernoulli')\n",
        "plt.xlabel('Measurements')\n",
        "plt.ylabel('Test Accuracy')\n",
        "plt.legend()\n",
        "plt.show()\n",
        "plt.plot(meas,train_acc_g,label='Gaussian')\n",
        "plt.plot(meas,train_acc_b,label='Bernoulli')\n",
        "plt.xlabel('Measurements')\n",
        "plt.ylabel('Train Accuracy')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Stats for MNIST dataset SVM using PCA with 100 components\n",
            "No. of Training Samples =  500\n",
            "No. of Testing Samples =  2000\n",
            "Dimensions of images 28*28\n",
            "Number of features in data domain (originally) = 784\n",
            "Number of PCA components = 100 for both data domain and measurement domain \n",
            "Train accuracy in original domain =  1.0\n",
            "Test accuracy in original domain =  0.875\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eFK0XANTxsph",
        "colab_type": "code",
        "outputId": "23e54a75-5a53-4b65-8561-4695c184bb46",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        }
      },
      "source": [
        "print(p3.shape)\n",
        "print(p2.shape)\n",
        "print(p2_test.shape)\n",
        "print(p3_test.shape)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(500, 100)\n",
            "(500, 100)\n",
            "(2000, 100)\n",
            "(2000, 100)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "prOLtSXHyfzJ",
        "colab_type": "code",
        "outputId": "9309fb01-26f8-474e-cc22-c515e1c102ca",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 793
        }
      },
      "source": [
        "t2=np.matmul(x_train_s,pca2.components_.T)\n",
        "print(np.matmul(x_train_s,pca2.components_.T))\n",
        "f1=np.matmul(x_train_s,A.T)\n",
        "print(f1.shape)\n",
        "print(A.shape)\n",
        "g=np.matmul(f1,A)\n",
        "print(np.matmul(g,pca2.components_.T)/m)\n",
        "print()\n",
        "# print(pca2.transform(x_train_s))\n",
        "t3=np.matmul(x_train_g_s,pca3.components_.T)\n",
        "print(np.matmul(x_train_g_s,pca3.components_.T))\n",
        "# print(np.matmul(pca2.components_,pca2.components_.T))\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[ 8.21712376e+02  6.91342522e+02 -4.76140271e+02 ... -3.83742661e-01\n",
            "  -7.13217161e+01  3.54302759e+01]\n",
            " [ 7.75941449e+02  5.08872427e+02  9.78564723e+02 ... -5.77089419e+00\n",
            "   1.92034460e+02  1.08546499e+02]\n",
            " [ 3.40002833e+02  4.28904755e+02  6.82711785e+02 ... -6.41568130e+01\n",
            "   6.28130537e+01  1.26371134e+02]\n",
            " ...\n",
            " [ 1.08039338e+03  9.93807221e+02  6.98737599e+02 ... -1.66224897e+02\n",
            "  -5.19771097e+01 -1.06951434e+02]\n",
            " [-3.19411110e+02  1.09417394e+03  5.70295637e+02 ... -4.23395459e+01\n",
            "   2.62856243e+01 -4.67694654e+01]\n",
            " [ 1.45932526e+03  1.05668060e+02  6.03619628e+02 ... -2.50381269e+01\n",
            "  -9.12676138e+00  5.78392625e-01]]\n",
            "(500, 200)\n",
            "(200, 784)\n",
            "[[ 5.08211506e+02  8.97108634e+02 -5.90601305e+02 ...  9.41187099e+00\n",
            "  -2.74944179e+02 -1.17099634e+00]\n",
            " [ 4.27547177e+02  8.23324170e+02  1.02280522e+03 ... -3.79571044e+02\n",
            "   4.60171703e+02  3.18604032e+02]\n",
            " [ 8.78519924e+01  8.62657208e+02  7.14995033e+02 ... -1.90120899e+02\n",
            "   6.17678635e+01  2.89261392e+02]\n",
            " ...\n",
            " [ 1.05764844e+03  1.03117519e+03  9.14455098e+02 ... -5.32157553e+02\n",
            "  -2.73547538e+02 -8.59458320e+00]\n",
            " [-2.12117261e+02  1.39309708e+03  5.83135749e+02 ... -1.68618900e+02\n",
            "  -1.26141419e+02  3.01556738e+01]\n",
            " [ 1.05946699e+03  3.85527814e+02  8.56055389e+02 ... -2.62634997e+02\n",
            "  -2.29848936e+02  1.76606780e+02]]\n",
            "\n",
            "[[ 5.89942081e+02  5.90646881e+02 -9.60642667e+02 ...  3.11110506e+01\n",
            "  -6.27413284e+01  4.30575169e+01]\n",
            " [ 8.21308941e+02  1.04293109e+03  8.29090183e+02 ... -5.74901872e+01\n",
            "  -2.18580262e+00  8.53356283e+01]\n",
            " [ 3.85682123e+02  9.05629964e+02  3.72836170e+02 ...  4.88666582e+01\n",
            "   5.23885543e+01  3.71468502e+01]\n",
            " ...\n",
            " [ 1.49414196e+03  4.28633018e+02  4.72819696e-01 ...  2.64558359e+01\n",
            "  -1.17074355e+02  6.01363778e+01]\n",
            " [ 2.07434376e+02  1.47844478e+03  1.66625607e+02 ...  4.48008186e+00\n",
            "   6.06657723e+01  2.74853348e+01]\n",
            " [ 1.34917899e+03  2.87527814e+01  3.71780124e+02 ... -5.85917799e+01\n",
            "   1.28757348e+02 -4.03650364e+01]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9ac9TsXE1pvl",
        "colab_type": "code",
        "outputId": "4f2ce43f-9caf-4444-db84-644a17a8323c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 187
        }
      },
      "source": [
        "for i in list(range(0,10)):\n",
        "  print(np.dot(t2[i,:],t3[i,:]) /(np.linalg.norm(t2[i,:]) * np.linalg.norm(t3[i,:])))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "-0.002070577751351654\n",
            "-0.09326591928620177\n",
            "0.3273862287060057\n",
            "0.5438094024611159\n",
            "0.373085088809044\n",
            "0.40128263326989605\n",
            "0.2795994498725493\n",
            "0.24515628642539605\n",
            "0.30755402444328717\n",
            "0.02385367195660565\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}