VGG19_encoding_model.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "VGG19 encoding model of fMRI data.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "EdHVQrqyDB50",
        "FgVfBlvZDEja",
        "d6zmir64CXlo",
        "0fNcF4-4Cd-R",
        "2rzGLI22CkfB",
        "B0U08ONGCzkA",
        "9_ivWo4xDCPq",
        "scnuhATSDJX9",
        "jJeWCZ3ODQD7"
      ],
      "toc_visible": true,
      "machine_shape": "hm"
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qbLBFKq3YZpB"
      },
      "source": [
        "# Stay connected \n",
        "```\n",
        "var startClickConnect = function startClickConnect(){\n",
        "    var clickConnect = function clickConnect(){\n",
        "        console.log(\"Connnect Clicked - Start\");\n",
        "        document.querySelector(\"#top-toolbar > colab-connect-button\").shadowRoot.querySelector(\"#connect\").click();\n",
        "        console.log(\"Connnect Clicked - End\"); \n",
        "    };\n",
        "\n",
        "    var intervalId = setInterval(clickConnect, 60000);\n",
        "\n",
        "    var stopClickConnectHandler = function stopClickConnect() {\n",
        "        console.log(\"Connnect Clicked Stopped - Start\");\n",
        "        clearInterval(intervalId);\n",
        "        console.log(\"Connnect Clicked Stopped - End\");\n",
        "    };\n",
        "\n",
        "    return stopClickConnectHandler;\n",
        "};\n",
        "\n",
        "var stopClickConnect = startClickConnect();\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EdHVQrqyDB50"
      },
      "source": [
        "# Memory Information"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "E58HaeA6CsM9",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "f101a092-2ae2-463d-c141-cb1711c77244"
      },
      "source": [
        "import psutil\n",
        "def get_size(bytes, suffix=\"B\"):\n",
        "    factor = 1024\n",
        "    for unit in [\"\", \"K\", \"M\", \"G\", \"T\", \"P\"]:\n",
        "        if bytes < factor:\n",
        "            return f\"{bytes:.2f}{unit}{suffix}\"\n",
        "        bytes /= factor\n",
        "print(\"=\"*40, \"Memory Information\", \"=\"*40)\n",
        "svmem = psutil.virtual_memory()\n",
        "print(f\"Total: {get_size(svmem.total)}\") ; print(f\"Available: {get_size(svmem.available)}\")\n",
        "print(f\"Used: {get_size(svmem.used)}\") ; print(f\"Percentage: {svmem.percent}%\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "======================================== Memory Information ========================================\n",
            "Total: 25.46GB\n",
            "Available: 24.60GB\n",
            "Used: 526.25MB\n",
            "Percentage: 3.4%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FgVfBlvZDEja"
      },
      "source": [
        "# GPU Information"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rxlkxvkrCyin",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e0c8a991-bd1d-4dd5-b785-a8ff004d6f52"
      },
      "source": [
        "! nvidia-smi"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Thu Feb 25 14:01:26 2021       \n",
            "+-----------------------------------------------------------------------------+\n",
            "| NVIDIA-SMI 460.39       Driver Version: 460.32.03    CUDA Version: 11.2     |\n",
            "|-------------------------------+----------------------+----------------------+\n",
            "| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n",
            "| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\n",
            "|                               |                      |               MIG M. |\n",
            "|===============================+======================+======================|\n",
            "|   0  Tesla P100-PCIE...  Off  | 00000000:00:04.0 Off |                    0 |\n",
            "| N/A   34C    P0    26W / 250W |      0MiB / 16280MiB |      0%      Default |\n",
            "|                               |                      |                  N/A |\n",
            "+-------------------------------+----------------------+----------------------+\n",
            "                                                                               \n",
            "+-----------------------------------------------------------------------------+\n",
            "| Processes:                                                                  |\n",
            "|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |\n",
            "|        ID   ID                                                   Usage      |\n",
            "|=============================================================================|\n",
            "|  No running processes found                                                 |\n",
            "+-----------------------------------------------------------------------------+\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "d6zmir64CXlo"
      },
      "source": [
        "# **Setup**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jyDODoaWC6KI"
      },
      "source": [
        "# install libraries not in Colab\n",
        "%%capture\n",
        "!pip install decord"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KVkBbPrwCZy2"
      },
      "source": [
        "# standard sci libraries\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "# machine learning libraries\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import torchvision.models as models\n",
        "import torch.hub as hub\n",
        "from typing import Union, List, Dict, Any, cast\n",
        "\n",
        "from torchvision import transforms as trn\n",
        "from torch.autograd import Variable as V\n",
        "from sklearn.decomposition import PCA\n",
        "from sklearn.preprocessing import StandardScaler\n",
        "from sklearn.linear_model import LinearRegression\n",
        "from sklearn.multioutput import MultiOutputRegressor\n",
        "\n",
        "# image & video libraries\n",
        "from decord import VideoReader  \n",
        "from PIL import Image    \n",
        "\n",
        "# others\n",
        "import time\n",
        "import glob\n",
        "import os\n",
        "import pickle"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1smtTLMmCbwG"
      },
      "source": [
        "# random number generators for reproducibility\n",
        "seed = 24\n",
        "\n",
        "torch.manual_seed(seed)\n",
        "np.random.seed(seed)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0fNcF4-4Cd-R"
      },
      "source": [
        "# **Data loading**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "htBGNx62Cdph",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c0571a9f-7525-495c-af4a-0a1414e77c45"
      },
      "source": [
        "# download dataset from dropbox\n",
        "dropbox_link = 'https://www.dropbox.com/s/agxyxntrbwko7t1/participants_data.zip?dl=0'\n",
        "os.environ['download_link'] = dropbox_link  \n",
        "!echo $download_link\n",
        "!wget -O participants_data.zip -c $download_link\n",
        "!wget -c https://raw.githubusercontent.com/Neural-Dynamics-of-Visual-Cognition-FUB/Algonauts2021_devkit/main/class_names_ImageNet.txt"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "https://www.dropbox.com/s/agxyxntrbwko7t1/participants_data.zip?dl=0\n",
            "--2021-08-08 12:58:39--  https://www.dropbox.com/s/agxyxntrbwko7t1/participants_data.zip?dl=0\n",
            "Resolving www.dropbox.com (www.dropbox.com)... 162.125.5.18, 2620:100:601f:18::a27d:912\n",
            "Connecting to www.dropbox.com (www.dropbox.com)|162.125.5.18|:443... connected.\n",
            "HTTP request sent, awaiting response... 301 Moved Permanently\n",
            "Location: /s/raw/agxyxntrbwko7t1/participants_data.zip [following]\n",
            "--2021-08-08 12:58:40--  https://www.dropbox.com/s/raw/agxyxntrbwko7t1/participants_data.zip\n",
            "Reusing existing connection to www.dropbox.com:443.\n",
            "HTTP request sent, awaiting response... 302 Found\n",
            "Location: https://uc649bf93b7bb08afeb287c313d0.dl.dropboxusercontent.com/cd/0/inline/BT2auM6oXoCdGEZR-DE30oxoTHacOLS96ZSWD2uG-yJaUadF-e3Y2a3XoolgPmTQPjQoGhoijMVAeuEgmBHTqMW8TQOexBIiWHSY9Lhp43TIJHhv59uT_XViN2Sb_Ud_g4XdQvaOdjCauM52RDjxPRU8/file# [following]\n",
            "--2021-08-08 12:58:40--  https://uc649bf93b7bb08afeb287c313d0.dl.dropboxusercontent.com/cd/0/inline/BT2auM6oXoCdGEZR-DE30oxoTHacOLS96ZSWD2uG-yJaUadF-e3Y2a3XoolgPmTQPjQoGhoijMVAeuEgmBHTqMW8TQOexBIiWHSY9Lhp43TIJHhv59uT_XViN2Sb_Ud_g4XdQvaOdjCauM52RDjxPRU8/file\n",
            "Resolving uc649bf93b7bb08afeb287c313d0.dl.dropboxusercontent.com (uc649bf93b7bb08afeb287c313d0.dl.dropboxusercontent.com)... 162.125.83.15, 2620:100:601f:15::a27d:90f\n",
            "Connecting to uc649bf93b7bb08afeb287c313d0.dl.dropboxusercontent.com (uc649bf93b7bb08afeb287c313d0.dl.dropboxusercontent.com)|162.125.83.15|:443... connected.\n",
            "HTTP request sent, awaiting response... 302 Found\n",
            "Location: /cd/0/inline2/BT3OC9XNDo2KFh5BYxl6NYh280OyJFJG0Jii2D_vgIlB_KezXz_-4mFvC0rKB4G_cO5W5yCOOFIUtpUyC2cg6mipAKsb4b2zXTqOibozorf6yp1bXT7GhViYyzyzKXYz0tlxTwgfPZ-ozcRUkPr4CYg66Ueek4MFJcBOvzinNbqx2BhcQ78CcNVmvKFJgkbsOs8pzJ_wyMXNvx3rFnYRnc_9LIIMnGEjdwsASmX9vnTTKImo3zCFNnI1fOajsRoA9qygNFYSz5bP7SLOCXYAciRwq9BbD9ClvSWuSu9PduriZUVpgpHE4Jeh-ldkFRmcDMqvDJ7EBQMFTPuNh3UOlQt9i3XN8yR0tllCemEa6t9TmDGJGO8FL1XGfIqeej3t3n8/file [following]\n",
            "--2021-08-08 12:58:41--  https://uc649bf93b7bb08afeb287c313d0.dl.dropboxusercontent.com/cd/0/inline2/BT3OC9XNDo2KFh5BYxl6NYh280OyJFJG0Jii2D_vgIlB_KezXz_-4mFvC0rKB4G_cO5W5yCOOFIUtpUyC2cg6mipAKsb4b2zXTqOibozorf6yp1bXT7GhViYyzyzKXYz0tlxTwgfPZ-ozcRUkPr4CYg66Ueek4MFJcBOvzinNbqx2BhcQ78CcNVmvKFJgkbsOs8pzJ_wyMXNvx3rFnYRnc_9LIIMnGEjdwsASmX9vnTTKImo3zCFNnI1fOajsRoA9qygNFYSz5bP7SLOCXYAciRwq9BbD9ClvSWuSu9PduriZUVpgpHE4Jeh-ldkFRmcDMqvDJ7EBQMFTPuNh3UOlQt9i3XN8yR0tllCemEa6t9TmDGJGO8FL1XGfIqeej3t3n8/file\n",
            "Reusing existing connection to uc649bf93b7bb08afeb287c313d0.dl.dropboxusercontent.com:443.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 4591675883 (4.3G) [application/zip]\n",
            "Saving to: ‘participants_data.zip’\n",
            "\n",
            "participants_data.z 100%[===================>]   4.28G  12.1MB/s    in 6m 12s  \n",
            "\n",
            "2021-08-08 13:04:54 (11.8 MB/s) - ‘participants_data.zip’ saved [4591675883/4591675883]\n",
            "\n",
            "--2021-08-08 13:04:54--  https://raw.githubusercontent.com/Neural-Dynamics-of-Visual-Cognition-FUB/Algonauts2021_devkit/main/class_names_ImageNet.txt\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.109.133, 185.199.110.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 416 Range Not Satisfiable\n",
            "\n",
            "    The file is already fully retrieved; nothing to do.\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Bc-xzICNCh8w"
      },
      "source": [
        "# unzip data file (%%capture suppresses the output)\n",
        "%%capture\n",
        "!unzip -o participants_data.zip"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2rzGLI22CkfB"
      },
      "source": [
        "# **Video preprocessing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qaB2fqzIClnr",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "19faef00-2b3c-4046-f0b4-0aefde91a41a"
      },
      "source": [
        "# Colab directory containing videos\n",
        "video_dir = '/content/AlgonautsVideos268_All_30fpsmax'\n",
        "\n",
        "# get path of all videos\n",
        "video_list = glob.glob(video_dir + '/*.mp4')\n",
        "\n",
        "# sort videos in ascending order\n",
        "video_list.sort()\n",
        "\n",
        "print('Total number of videos: ', len(video_list))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Total number of videos:  1102\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WCTF6gkSCtTt"
      },
      "source": [
        "def sample_video_from_mp4(file, num_frames = 16):\n",
        "    \"\"\"This function takes a mp4 video file as input and returns\n",
        "    an array of uniformly sampled frames.\n",
        "    \n",
        "    Args\n",
        "    ----------\n",
        "    file : str\n",
        "        path to mp4 video file\n",
        "    num_frames : int\n",
        "        number of frames to select with uniform frame sampling\n",
        "    \n",
        "    Returns\n",
        "    -------\n",
        "    frames: list of frames as PIL images\n",
        "    num_frames: number of sampled frames\n",
        "\n",
        "    \"\"\"\n",
        "    \n",
        "    # read video file\n",
        "    video = VideoReader(file)\n",
        "\n",
        "    # get total number of video frames\n",
        "    total_frames = len(video)\n",
        "\n",
        "    # create frame indices \n",
        "    frame_indices = np.linspace(0, total_frames-1, num_frames, dtype = np.int) \n",
        "\n",
        "    video_frames = []\n",
        "\n",
        "    # fill list of video frames as PIL images\n",
        "    for i in frame_indices:\n",
        "      video_frames.append(Image.fromarray(video[i].asnumpy()))\n",
        "\n",
        "    return video_frames, num_frames\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QuVF-wdTCu4z",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "7449f27c-d22e-47bf-9bea-960171ad58b7"
      },
      "source": [
        "# example of using the previous function\n",
        "example_frames, num_frames = sample_video_from_mp4(video_list[8])\n",
        "len(example_frames)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "16"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 85
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HkZ9Ll0fCwgb"
      },
      "source": [
        "# **Neural network feature extraction**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "B0U08ONGCzkA"
      },
      "source": [
        "## Pretrained model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ff2Zm-BnCyqw"
      },
      "source": [
        "__all__ = ['VGG', 'vgg19_bn', 'vgg19']\n",
        "\n",
        "\n",
        "model_urls = {\n",
        "    'vgg19': 'https://download.pytorch.org/models/vgg19-dcbb9e9d.pth',\n",
        "    'vgg19_bn': 'https://download.pytorch.org/models/vgg19_bn-c79401a0.pth',\n",
        "    }\n",
        "\n",
        "\n",
        "class VGG19(nn.Module):\n",
        "\n",
        "    def __init__(self, num_classes = 1000):\n",
        "        super(VGG19, self).__init__()\n",
        "        \n",
        "        self.layers = nn.ModuleList([nn.Sequential(\n",
        "            nn.Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True),\n",
        "            nn.MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True),\n",
        "            nn.MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True),\n",
        "            nn.MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True),\n",
        "            nn.MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),\n",
        "            nn.ReLU(inplace=True),\n",
        "            nn.MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Linear(in_features=512 * 7 * 7, out_features=4096, bias=True),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Dropout(p=0.5, inplace=False),\n",
        "            nn.Linear(in_features=4096, out_features=4096, bias=True),\n",
        "            nn.ReLU(inplace=True)\n",
        "        ),\n",
        "        nn.Sequential(\n",
        "            nn.Dropout(p=0.5, inplace=False),\n",
        "            nn.Linear(in_features=4096, out_features=num_classes, bias=True)\n",
        "        )])\n",
        "\n",
        "\n",
        "    def forward(self, x):\n",
        "        \n",
        "        activations = []\n",
        "          \n",
        "        for i, layer in enumerate(self.layers): \n",
        "            if i == 0:\n",
        "                activations.append(layer(x))\n",
        "            elif i == 16:\n",
        "                out15_reshaped = activations[i-1].view(activations[i-1].size(0), 512 * 7 * 7)\n",
        "                activations.append(layer(out15_reshaped))\n",
        "            else:\n",
        "                activations.append(layer(activations[i-1]))  \n",
        "\n",
        "        return activations\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CgkWQ2_iC4ru"
      },
      "source": [
        "def load_model(model_url):\n",
        "    \"\"\"This function initializes a neural network (VGG19) and load\n",
        "    its weights from a pretrained model.\n",
        "    \n",
        "    Args:\n",
        "        arch: class\n",
        "            pytorch neural network.\n",
        "        model_url: str\n",
        "            pytorch model. \n",
        "\n",
        "    Returns:\n",
        "        model: class\n",
        "            pytorch model ready for inference.\n",
        "    \"\"\"\n",
        "\n",
        "    # instantiate model architecture\n",
        "    model = VGG19()\n",
        "    \n",
        "    # list of parameters name of the model\n",
        "    param_names = list(model.state_dict())  \n",
        "\n",
        "    # initialise dictionary of model parameters\n",
        "    model_dict = {k:None for k in param_names}\n",
        "\n",
        "    # load parameters of pretrained model\n",
        "    state_dict = hub.load_state_dict_from_url(model_url)\n",
        "    \n",
        "    i = 0\n",
        "    for v in state_dict.values():\n",
        "        model_dict[param_names[i]] = v\n",
        "        i += 1\n",
        "\n",
        "    model.load_state_dict(model_dict)\n",
        "    \n",
        "    # gpu availability\n",
        "    if torch.cuda.is_available():\n",
        "        model.cuda()\n",
        "\n",
        "    # set inference mode\n",
        "    model.eval()\n",
        "\n",
        "    return model"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dhpP7S5mDBK-"
      },
      "source": [
        "vgg19 = load_model(model_urls['vgg19'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9_ivWo4xDCPq"
      },
      "source": [
        "## Model activations"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mGgFybtQDEOX"
      },
      "source": [
        "def get_activations_and_save(model, video_list, save_dir, layer):\n",
        "    \"\"\"This function extracts the activations (features) of a specific layer of\n",
        "    a model to a set of videos and save them in a specified directory. Every\n",
        "    file is a list with a vector containing the activations of that layer to\n",
        "    a particular video. The activations are averaged over the frames of every\n",
        "    video. \n",
        "\n",
        "    Parameters\n",
        "    ----------\n",
        "    model :\n",
        "        pytorch model.\n",
        "    video_list : list\n",
        "        list containing path to all videos.\n",
        "    save_dir : str\n",
        "        save path for extracted activations.\n",
        "    layer : int\n",
        "        integer specifying layer number.\n",
        "    \"\"\"\n",
        "    \n",
        "    # define preprocessing function\n",
        "    preprocess = trn.Compose([\n",
        "            trn.Resize((224,224)),\n",
        "            trn.ToTensor(),\n",
        "            trn.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])\n",
        "\n",
        "    for i, video in enumerate(video_list):\n",
        "\n",
        "      # name video file\n",
        "      video_file_name = os.path.split(video)[-1].split(\".\")[0]\n",
        "      \n",
        "      # load video frames\n",
        "      video_frames, num_frames = sample_video_from_mp4(video)\n",
        "      \n",
        "      activations = []\n",
        "      for frame, image in enumerate(video_frames):\n",
        "\n",
        "        # preprocess video frame\n",
        "        input_image = V(preprocess(image).unsqueeze(0))\n",
        "        \n",
        "        # put image onto gpu if available\n",
        "        if torch.cuda.is_available():\n",
        "            input_image = input_image.cuda()\n",
        "        \n",
        "        # feed image through the model\n",
        "        layer_outputs = model.forward(input_image)\n",
        "\n",
        "        if frame==0:\n",
        "          # append and flatten layer activations\n",
        "          activations.append(layer_outputs[layer].ravel())\n",
        "        else:\n",
        "          # add activations over frames\n",
        "          activations[0] = activations[0] + layer_outputs[layer].ravel()\n",
        "        \n",
        "        # average layer activations across frames\n",
        "        avg_layer_activations = np.array([activations])/float(num_frames)\n",
        "        \n",
        "        # define saving directory\n",
        "        save_path = os.path.join(save_dir, video_file_name + \"_\" +\n",
        "                               \"layer\" + \"_\" + str(layer) + \".npy\")\n",
        "\n",
        "        # save activations for a particular video\n",
        "        np.save(save_path, avg_layer_activations)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vXsCAFgMDIFa"
      },
      "source": [
        "############################### change dir ###########################\n",
        "# create saving directory for activations\n",
        "activations_dir = \"/content/activations_vgg19\"\n",
        "if not os.path.exists(activations_dir):\n",
        "  os.makedirs(activations_dir)\n",
        "\n",
        "# get activations\n",
        "get_activations_and_save(vgg19, video_list, activations_dir, layer = 18)   ##### SPECIFY LAYER NUMBER #####"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "scnuhATSDJX9"
      },
      "source": [
        "## Perform PCA on activations"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6KJibw6_DKz-"
      },
      "source": [
        "def do_PCA_and_save(activations_dir, save_dir): \n",
        "    \"\"\"This function preprocesses a neural network's features using PCA and\n",
        "    save the results in a specified directory.\n",
        "\n",
        "    Parameters\n",
        "    ----------\n",
        "    activations_dir : str\n",
        "        save path for extracted activations.\n",
        "    save_dir : str\n",
        "        save path for extracted PCA features.\n",
        "    \"\"\"\n",
        "\n",
        "    # number of PCA components\n",
        "    n_components = 100     \n",
        "        \n",
        "    # get path of activations of a layer to all videos\n",
        "    activations_file_list = glob.glob(activations_dir + '/*.npy')\n",
        "    # sort activations in ascending order\n",
        "    activations_file_list.sort()\n",
        "    # load one activation to later get dimension\n",
        "    feature_dim = np.load(activations_file_list[0], allow_pickle=True)[0][0].shape[0]\n",
        "        \n",
        "    # initialise data matrix for PCA - #videos x layer dimension\n",
        "    x = np.zeros((len(activations_file_list), feature_dim))\n",
        "        \n",
        "    # fill matrix with activations of every video (for a given layer)\n",
        "    for i, video_activations in enumerate(activations_file_list):\n",
        "        temp = np.load(video_activations, allow_pickle=True)[0][0].detach().numpy()\n",
        "        x[i,:] = temp\n",
        "        \n",
        "    # train-test split\n",
        "    x_train = x[:1000,:]\n",
        "    x_test = x[1000:,:]\n",
        "\n",
        "    print(\"Scaling data\")\n",
        "    # scale training and test sets\n",
        "    x_train = StandardScaler().fit_transform(x_train)\n",
        "    x_test = StandardScaler().fit_transform(x_test)\n",
        "\n",
        "    print(\"Fitting PCA\")\n",
        "    # find principal components of training set\n",
        "    pca = PCA(n_components = n_components, random_state = seed)\n",
        "    pca.fit(x_train)\n",
        "\n",
        "    print(\"Projecting data onto components\")\n",
        "    # project data onto components - #videos x #components \n",
        "    x_train = pca.transform(x_train)\n",
        "    x_test = pca.transform(x_test)\n",
        "\n",
        "    print(\"Saving\")\n",
        "    # create saving directory\n",
        "    train_save_path = os.path.join(save_dir, \"train_layer_19\")  ############ SPECIFY LAYER NUMBER #######\n",
        "    test_save_path = os.path.join(save_dir, \"test_layer_19\")\n",
        "\n",
        "    # save results\n",
        "    np.save(train_save_path, x_train)\n",
        "    np.save(test_save_path, x_test)\n",
        "        "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KTG4hqsGDORl",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5edad82d-6652-42b8-8602-6b958400e9d3"
      },
      "source": [
        "# create saving directory for pca features\n",
        "pca_dir = '/content/pca_activations'\n",
        "if not os.path.exists(pca_dir):\n",
        "  os.makedirs(pca_dir)\n",
        "\n",
        "# call function to get dimensionality-reduced activations\n",
        "do_PCA_and_save(activations_dir, pca_dir)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Scaling data\n",
            "Fitting PCA\n",
            "Projecting data onto components\n",
            "Saving\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jJeWCZ3ODQD7"
      },
      "source": [
        "# **Regression on fMRI data**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "z2WkeShVDSW0",
        "cellView": "form"
      },
      "source": [
        "#@title Helper functions for data loading\n",
        "\n",
        "\n",
        "# loading function for pickle ROI file\n",
        "def load_dict(filename_):\n",
        "    # read file in bytes\n",
        "    with open(filename_, 'rb') as f:\n",
        "        # create pickle object\n",
        "        u = pickle._Unpickler(f)\n",
        "        # # change encoding appropriate for numpy array\n",
        "        u.encoding = 'latin1'\n",
        "        # load it\n",
        "        ret_di = u.load()\n",
        "        \n",
        "    return ret_di\n",
        "\n",
        "\n",
        "def get_fmri(fmri_dir, sub, ROI):\n",
        "    \"\"\"This function loads fMRI data into a numpy array for a given\n",
        "    participant and ROI.\n",
        "    \n",
        "    Parameters\n",
        "    ----------\n",
        "    fmri_dir : str\n",
        "        path to fMRI data.\n",
        "    sub : str\n",
        "        participant number path\n",
        "    ROI : str\n",
        "        name of ROI.\n",
        "\n",
        "    Returns\n",
        "    ---------\n",
        "    np.array\n",
        "        matrix of dimensions #train_vids x #repetitions x #voxels\n",
        "        containing fMRI responses to train videos of a given ROI.\n",
        "    \"\"\"\n",
        "\n",
        "    # Load ROI data\n",
        "    ROI_file = os.path.join(fmri_dir, sub, ROI + \".pkl\")\n",
        "    ROI_data = load_dict(ROI_file)\n",
        "\n",
        "    # averaging ROI data across repetitions\n",
        "    ROI_data_train = np.mean(ROI_data[\"train\"], axis = 1)\n",
        "\n",
        "    return ROI_data_train\n",
        "\n",
        "\n",
        "def get_activations(activations_dir, layer_name):\n",
        "    \"\"\"This function loads neural network features/activations (preprocessed\n",
        "    using PCA) into a numpy array according to a given layer.\n",
        "    \n",
        "    Parameters\n",
        "    ----------\n",
        "    activations_dir : str\n",
        "        path to PCA/processed neural network features\n",
        "    layer_name : str\n",
        "        which layer of the neural network to load\n",
        "    \n",
        "    Returns\n",
        "    ---------\n",
        "    train_activations : np.array\n",
        "        matrix of dimensions #train_vids x #pca_components\n",
        "        containing activations of train videos\n",
        "    \"\"\"\n",
        "\n",
        "    # load training and test file of a given layer\n",
        "    train_file = os.path.join(activations_dir,\"train_\" + layer_name + \".npy\")\n",
        "    #test_file = os.path.join(activations_dir,\"test_\" + layer_name + \".npy\")\n",
        "    train_activations = np.load(train_file)\n",
        "    #test_activations = np.load(test_file)\n",
        "\n",
        "    # scale training and test activations\n",
        "    scaler = StandardScaler()\n",
        "    train_activations = scaler.fit_transform(train_activations)\n",
        "    #test_activations = scaler.fit_transform(test_activations)\n",
        "\n",
        "    return train_activations"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UMQ4JZPHDWpg",
        "cellView": "form"
      },
      "source": [
        "#@title Correlation function for model validation\n",
        "\n",
        "def vectorized_correlation(x,y):\n",
        "    dim = 0\n",
        "\n",
        "    centered_x = x - x.mean(axis=dim, keepdims=True)\n",
        "    centered_y = y - y.mean(axis=dim, keepdims=True)\n",
        "\n",
        "    covariance = (centered_x * centered_y).sum(axis=dim, keepdims=True)\n",
        "\n",
        "    bessel_corrected_covariance = covariance / (x.shape[dim] - 1)\n",
        "\n",
        "    x_std = x.std(axis=dim, keepdims=True)+1e-8\n",
        "    y_std = y.std(axis=dim, keepdims=True)+1e-8\n",
        "\n",
        "    corr = bessel_corrected_covariance / (x_std * y_std)\n",
        "\n",
        "    return corr.ravel()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "42Gdl4cwDcS-"
      },
      "source": [
        "def perform_encoding(pca_dir, fmri_dir, results_dir, layer, sub, ROI,\n",
        "                     batch_size=1000):\n",
        "  \"\"\"This function linearly regresses pca-reduced features/activations of a\n",
        "  given layer of a neural network to the fMRI activity of a given brain region\n",
        "  (ROI) in a given subject. The fitted model is used to predict part of the\n",
        "  training set (validation mode). The model predictions are saved in a specified\n",
        "  directory. \n",
        "  \n",
        "  # Arguments\n",
        "  -------------\n",
        "  pca_dir : str\n",
        "      path to PCA features.\n",
        "  fmri_dir : str\n",
        "      path to fMRI data.\n",
        "  results_dir : str\n",
        "      saving directory for results.\n",
        "  layer : str\n",
        "      layer name from which to extract activations.\n",
        "  sub : str\n",
        "      participant number path.\n",
        "  ROI : str\n",
        "      region of interest (brain region) from which to extract fMRI data.\n",
        "  batch_size : int\n",
        "      Number of voxels processed when fitting the linear regressor. 1000 by\n",
        "      default.\n",
        "\n",
        "  \"\"\"\n",
        "\n",
        "  # load train activations\n",
        "  train_activations = get_activations(pca_dir, layer)\n",
        "  \n",
        "  # load fMRI data for a given subject and ROI\n",
        "  fmri_train_all = get_fmri(fmri_dir, sub, ROI)\n",
        "  # get number of voxels for given ROI\n",
        "  num_voxels = fmri_train_all.shape[1]\n",
        "\n",
        "  # create training and validation sets\n",
        "  val_activations = train_activations[900:,:]       ################ adjust splits according to #videos ###############\n",
        "  train_activations = train_activations[:900,:]    \n",
        "  fmri_train = fmri_train_all[:900,:]\n",
        "  fmri_val = fmri_train_all[900:,:]      \n",
        "\n",
        "  # initialise results - #validation videos x #voxels\n",
        "  pred_fmri = np.zeros_like(fmri_val)\n",
        "  pred_fmri_save_path = os.path.join(results_dir)  ##### removed subject and ROI #####\n",
        "\n",
        "  # perform multiple multivariate regression\n",
        "  reg = MultiOutputRegressor(LinearRegression()).fit(train_activations, fmri_train)\n",
        "  pred_fmri = reg.predict(val_activations)\n",
        "  \n",
        "  # correlation between validation predictions and ground truth\n",
        "  corr = vectorized_correlation(fmri_val, pred_fmri)\n",
        "  print(\"----------------------------------------------------------------------------\")\n",
        "  # get mean correlation\n",
        "  voxelwise_corr = round(corr.mean(), 6)\n",
        "  print(\"Mean correlation for ROI : \",ROI,\" in \",sub,\" using \",layer,\" is :\", voxelwise_corr)\n",
        "  \n",
        "  # save results\n",
        "  np.save(pred_fmri_save_path, pred_fmri)\n",
        "\n",
        "  return voxelwise_corr\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XZzgBLKzDeyH"
      },
      "source": [
        "# specify create fmri directory\n",
        "fmri_dir = '/content/participants_data_v2021/mini_track' \n",
        "\n",
        "# create saving directory for results\n",
        "prediction_dir = '/content/model_predictions'\n",
        "if not os.path.exists(prediction_dir):\n",
        "  os.makedirs(prediction_dir)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0o5dvac9DgXL"
      },
      "source": [
        "# list of subjects, ROIs, model and layers\n",
        "subs = [\"sub01\",\"sub02\",\"sub03\",\"sub04\",\"sub05\",\"sub06\",\"sub07\",\"sub08\",\"sub09\",\"sub10\"]  \n",
        "ROIs = [\"V1\",\"V2\",\"V3\",\"V4\",\"LOC\",\"EBA\",\"FFA\",\"STS\",\"PPA\"]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CSFgV72nDiGH"
      },
      "source": [
        "##############################\n",
        "#### CHANGE LAYER NUMBER #####\n",
        "layer = 'layer_19'\n",
        "##############################"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LlGHxbcxDjnx",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "b3efc863-2c56-49a7-8a36-5df0355cb7eb"
      },
      "source": [
        "# loop across subjects and layers\n",
        "voxelwise_corrs = np.zeros((len(subs), len(ROIs)))\n",
        "\n",
        "for i, sub in enumerate(subs):\n",
        "  for j, ROI in enumerate(ROIs):\n",
        "    # make prediction\n",
        "    print(\"Starting ROI: \", ROI, \"sub: \", sub)\n",
        "    voxelwise_corrs[i, j] = perform_encoding(pca_dir, fmri_dir, prediction_dir,\n",
        "                                        layer = layer, sub = sub, ROI = ROI)\n",
        "    print (\"Completed ROI: \", ROI, \"sub: \", sub)\n",
        "    print(\"----------------------------------------------------------------------------\")\n",
        "\n",
        "# create saving directory\n",
        "all_prediction_dir = os.path.join(prediction_dir, 'mean_correlation_all_subs_all_ROIs_with_' + layer)\n",
        "np.save(all_prediction_dir, voxelwise_corrs)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Starting ROI:  V1 sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub01  using  layer_19  is : 0.106415\n",
            "Completed ROI:  V1 sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub01  using  layer_19  is : 0.139367\n",
            "Completed ROI:  V2 sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub01  using  layer_19  is : 0.165812\n",
            "Completed ROI:  V3 sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub01  using  layer_19  is : 0.153617\n",
            "Completed ROI:  V4 sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub01  using  layer_19  is : 0.231041\n",
            "Completed ROI:  LOC sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub01  using  layer_19  is : 0.231025\n",
            "Completed ROI:  EBA sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub01  using  layer_19  is : 0.207018\n",
            "Completed ROI:  FFA sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub01  using  layer_19  is : 0.079394\n",
            "Completed ROI:  STS sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub01  using  layer_19  is : 0.172835\n",
            "Completed ROI:  PPA sub:  sub01\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V1 sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub02  using  layer_19  is : 0.15009\n",
            "Completed ROI:  V1 sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub02  using  layer_19  is : 0.127038\n",
            "Completed ROI:  V2 sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub02  using  layer_19  is : 0.118618\n",
            "Completed ROI:  V3 sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub02  using  layer_19  is : 0.189301\n",
            "Completed ROI:  V4 sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub02  using  layer_19  is : 0.195839\n",
            "Completed ROI:  LOC sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub02  using  layer_19  is : 0.205142\n",
            "Completed ROI:  EBA sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub02  using  layer_19  is : 0.150283\n",
            "Completed ROI:  FFA sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub02  using  layer_19  is : 0.080297\n",
            "Completed ROI:  STS sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub02  using  layer_19  is : 0.152117\n",
            "Completed ROI:  PPA sub:  sub02\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V1 sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub03  using  layer_19  is : 0.072968\n",
            "Completed ROI:  V1 sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub03  using  layer_19  is : 0.109164\n",
            "Completed ROI:  V2 sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub03  using  layer_19  is : 0.113978\n",
            "Completed ROI:  V3 sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub03  using  layer_19  is : 0.076521\n",
            "Completed ROI:  V4 sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub03  using  layer_19  is : 0.203071\n",
            "Completed ROI:  LOC sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub03  using  layer_19  is : 0.201146\n",
            "Completed ROI:  EBA sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub03  using  layer_19  is : 0.19282\n",
            "Completed ROI:  FFA sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub03  using  layer_19  is : 0.102606\n",
            "Completed ROI:  STS sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub03  using  layer_19  is : 0.165013\n",
            "Completed ROI:  PPA sub:  sub03\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V1 sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub04  using  layer_19  is : 0.204729\n",
            "Completed ROI:  V1 sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub04  using  layer_19  is : 0.170966\n",
            "Completed ROI:  V2 sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub04  using  layer_19  is : 0.184912\n",
            "Completed ROI:  V3 sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub04  using  layer_19  is : 0.133017\n",
            "Completed ROI:  V4 sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub04  using  layer_19  is : 0.247552\n",
            "Completed ROI:  LOC sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub04  using  layer_19  is : 0.254932\n",
            "Completed ROI:  EBA sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub04  using  layer_19  is : 0.188463\n",
            "Completed ROI:  FFA sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub04  using  layer_19  is : 0.101285\n",
            "Completed ROI:  STS sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub04  using  layer_19  is : 0.113173\n",
            "Completed ROI:  PPA sub:  sub04\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V1 sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub05  using  layer_19  is : 0.138532\n",
            "Completed ROI:  V1 sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub05  using  layer_19  is : 0.11329\n",
            "Completed ROI:  V2 sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub05  using  layer_19  is : 0.131175\n",
            "Completed ROI:  V3 sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub05  using  layer_19  is : 0.136824\n",
            "Completed ROI:  V4 sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub05  using  layer_19  is : 0.218177\n",
            "Completed ROI:  LOC sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub05  using  layer_19  is : 0.267621\n",
            "Completed ROI:  EBA sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub05  using  layer_19  is : 0.219273\n",
            "Completed ROI:  FFA sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub05  using  layer_19  is : 0.198496\n",
            "Completed ROI:  STS sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub05  using  layer_19  is : 0.194614\n",
            "Completed ROI:  PPA sub:  sub05\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V1 sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub06  using  layer_19  is : 0.140382\n",
            "Completed ROI:  V1 sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub06  using  layer_19  is : 0.123665\n",
            "Completed ROI:  V2 sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub06  using  layer_19  is : 0.108036\n",
            "Completed ROI:  V3 sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub06  using  layer_19  is : 0.104811\n",
            "Completed ROI:  V4 sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub06  using  layer_19  is : 0.240413\n",
            "Completed ROI:  LOC sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub06  using  layer_19  is : 0.235177\n",
            "Completed ROI:  EBA sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub06  using  layer_19  is : 0.206928\n",
            "Completed ROI:  FFA sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub06  using  layer_19  is : 0.122821\n",
            "Completed ROI:  STS sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub06  using  layer_19  is : 0.117857\n",
            "Completed ROI:  PPA sub:  sub06\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V1 sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub07  using  layer_19  is : 0.09519\n",
            "Completed ROI:  V1 sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub07  using  layer_19  is : 0.120754\n",
            "Completed ROI:  V2 sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub07  using  layer_19  is : 0.130745\n",
            "Completed ROI:  V3 sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub07  using  layer_19  is : 0.095777\n",
            "Completed ROI:  V4 sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub07  using  layer_19  is : 0.129615\n",
            "Completed ROI:  LOC sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub07  using  layer_19  is : 0.163941\n",
            "Completed ROI:  EBA sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub07  using  layer_19  is : 0.117821\n",
            "Completed ROI:  FFA sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub07  using  layer_19  is : 0.049093\n",
            "Completed ROI:  STS sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub07  using  layer_19  is : 0.116753\n",
            "Completed ROI:  PPA sub:  sub07\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V1 sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub08  using  layer_19  is : 0.126547\n",
            "Completed ROI:  V1 sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub08  using  layer_19  is : 0.067674\n",
            "Completed ROI:  V2 sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub08  using  layer_19  is : 0.077432\n",
            "Completed ROI:  V3 sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub08  using  layer_19  is : 0.086612\n",
            "Completed ROI:  V4 sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub08  using  layer_19  is : 0.182972\n",
            "Completed ROI:  LOC sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub08  using  layer_19  is : 0.189587\n",
            "Completed ROI:  EBA sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub08  using  layer_19  is : 0.171821\n",
            "Completed ROI:  FFA sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub08  using  layer_19  is : 0.174954\n",
            "Completed ROI:  STS sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub08  using  layer_19  is : 0.142204\n",
            "Completed ROI:  PPA sub:  sub08\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V1 sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub09  using  layer_19  is : 0.105605\n",
            "Completed ROI:  V1 sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub09  using  layer_19  is : 0.123403\n",
            "Completed ROI:  V2 sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub09  using  layer_19  is : 0.141165\n",
            "Completed ROI:  V3 sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub09  using  layer_19  is : 0.155708\n",
            "Completed ROI:  V4 sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub09  using  layer_19  is : 0.188283\n",
            "Completed ROI:  LOC sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub09  using  layer_19  is : 0.224622\n",
            "Completed ROI:  EBA sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub09  using  layer_19  is : 0.177898\n",
            "Completed ROI:  FFA sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub09  using  layer_19  is : 0.059013\n",
            "Completed ROI:  STS sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub09  using  layer_19  is : 0.094499\n",
            "Completed ROI:  PPA sub:  sub09\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V1 sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V1  in  sub10  using  layer_19  is : 0.161751\n",
            "Completed ROI:  V1 sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V2 sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V2  in  sub10  using  layer_19  is : 0.112142\n",
            "Completed ROI:  V2 sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V3 sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V3  in  sub10  using  layer_19  is : 0.10944\n",
            "Completed ROI:  V3 sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  V4 sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  V4  in  sub10  using  layer_19  is : 0.098523\n",
            "Completed ROI:  V4 sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  LOC sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  LOC  in  sub10  using  layer_19  is : 0.227131\n",
            "Completed ROI:  LOC sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  EBA sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  EBA  in  sub10  using  layer_19  is : 0.237711\n",
            "Completed ROI:  EBA sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  FFA sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  FFA  in  sub10  using  layer_19  is : 0.132456\n",
            "Completed ROI:  FFA sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  STS sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  STS  in  sub10  using  layer_19  is : 0.117842\n",
            "Completed ROI:  STS sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Starting ROI:  PPA sub:  sub10\n",
            "----------------------------------------------------------------------------\n",
            "Mean correlation for ROI :  PPA  in  sub10  using  layer_19  is : 0.130753\n",
            "Completed ROI:  PPA sub:  sub10\n",
            "----------------------------------------------------------------------------\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WyEWbmQyDlla",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "e16e0587-5dcf-488f-ed54-6fd2639e3e03"
      },
      "source": [
        "subjs_mean = np.mean(voxelwise_corrs, axis = 0)\n",
        "subjs_std = np.std(voxelwise_corrs, axis = 0)\n",
        "plt.plot(ROIs, subjs_mean)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[<matplotlib.lines.Line2D at 0x7fdb67a2a290>]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 100
        },
        {
          "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": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 319
        },
        "id": "HtuF2y6qWbC2",
        "outputId": "639f7282-d34a-4cdb-c06e-0ad529c0bab2"
      },
      "source": [
        "plt.plot(ROIs, subjs_mean, linewidth=2)\n",
        "plt.title('FC #3', fontsize = 20)\n",
        "plt.xlabel('Region of interest', fontsize=18, labelpad=15)\n",
        "plt.ylabel('Correlation', fontsize=18, labelpad=15)\n",
        "plt.ylim([0, 0.25])\n",
        "plt.savefig('Prediction_with_VGG19_layer_19.pdf', bbox_inches='tight')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "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"
          }
        }
      ]
    }
  ]
}