Invertable Neural net.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Populating the interactive namespace from numpy and matplotlib\n"
     ]
    }
   ],
   "source": [
    "%pylab inline\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import tensorflow.keras as keras\n",
    "from tensorflow.keras.datasets import mnist\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "import torch.nn.functional as F\n",
    "import tqdm as tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [],
   "source": [
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
    "num_classes = 10\n",
    "x_train = x_train.astype('float32') /255.0\n",
    "x_test = x_test.astype('float32')/255.0\n",
    "y_train = keras.utils.to_categorical(y_train, num_classes)\n",
    "y_test = keras.utils.to_categorical(y_test, num_classes)\n",
    "\n",
    "x_train = torch.tensor(x_train.reshape((-1,784)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [],
   "source": [
    "modules = []\n",
    "for i in range(1):\n",
    "    modules.append(nn.Linear(784,784,bias=True))\n",
    "    modules.append(nn.ELU())\n",
    "modules.append(nn.Linear(784,784,bias=True))\n",
    "model = nn.Sequential(*modules)\n",
    "#xt = torch.tensor(x_train[0:10].reshape((-1,784)))\n",
    "#model(xt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 600/600 [00:09<00:00, 62.64it/s]\n"
     ]
    }
   ],
   "source": [
    "#trainer\n",
    "optimizer = optim.Adam(model.parameters())\n",
    "model.train()\n",
    "bs=100\n",
    "for i in tqdm.tqdm(range(0,x_train.shape[0],bs)):\n",
    "    optimizer.zero_grad()\n",
    "    xx = x_train[i:i+bs]\n",
    "    yy = model(xx)\n",
    "    loss = F.mse_loss(xx,-yy)\n",
    "    loss.backward()\n",
    "    optimizer.step()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [],
   "source": [
    "def elu(x):\n",
    "    x = np.copy(x)\n",
    "    negs = x < 0\n",
    "    x[negs] = np.exp(x[negs])-1\n",
    "    return x\n",
    "\n",
    "def invelu(x):\n",
    "    x = np.copy(x)\n",
    "    negs = x < 0\n",
    "    x[negs] = np.log(x[negs]+1)\n",
    "    return x\n",
    "\n",
    "\n",
    "def forward(x):\n",
    "    for m in list(model.children()):\n",
    "        if type(m) == nn.Linear:\n",
    "            A = m.weight.detach().numpy()\n",
    "            x =np.dot(x,np.transpose(A))\n",
    "            x+= m.bias.detach().numpy()\n",
    "        elif type(m) == nn.ELU:\n",
    "            x = elu(x)\n",
    "    return x\n",
    "\n",
    "def inverse(x):\n",
    "    for m in list(model.children())[::-1]:\n",
    "        if type(m) == nn.Linear:\n",
    "            x-= m.bias.detach().numpy()\n",
    "            A = m.weight.detach().numpy()\n",
    "            x =np.dot(x,np.linalg.inv(np.transpose(A)))\n",
    "        elif type(m) == nn.ELU:\n",
    "            x = invelu(x)\n",
    "    return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fcdb8900da0>"
      ]
     },
     "execution_count": 82,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cc = random.randint(0,x_train.shape[0])\n",
    "xx = x_train[cc:cc+1]\n",
    "yy = forward(xx.detach().numpy()[0])\n",
    "imshow(yy.reshape(28,28))\n",
    "figure()\n",
    "imshow(xx.reshape(28,28))\n",
    "figure()\n",
    "zz = inverse(yy)\n",
    "imshow(zz.reshape(28,28))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [],
   "source": [
    "#cc = random.randint(0,x_train.shape[0])\n",
    "#xx = x_train[cc:cc+1]\n",
    "#imshow(xx.reshape(28,28))\n",
    "#figure()\n",
    "#yy = model(xx).detach().numpy()\n",
    "#imshow(yy.reshape(28,28))\n",
    "#figure()\n",
    "#zz = model(torch.tensor(yy)).detach().numpy()\n",
    "#imshow(yy.reshape(28,28))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}