Figure-5_Partial_Shuffling.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Take early V1 fMRI with early layers, first 2.\n",
    "Then take a later one that doesn't show effect.\n",
    "So, plot of (1,3) \n",
    "the partial shuffling will show a gradient in two but not in the third.\n",
    "\n",
    "Shuffle levels: 0% 20% 40% 60% 80% 100%."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "sys.path.append(\"/home/amurphy3/code/Github/cka_alignment/src\")\n",
    "\n",
    "import numpy as np \n",
    "import pandas as pd\n",
    "import pickle\n",
    "\n",
    "from itertools import chain\n",
    "from metrics import cka, gram_linear, RV2\n",
    "from pathlib import Path\n",
    "\n",
    "resnet_activations_path = Path('/home/amurphy3/data/THINGS/model_activations/val_fMRI/resnet18')\n",
    "fMRI_folder = Path('/home/amurphy3/data/THINGS/fMRI/fMRI_ROIs_val_THINGS')\n",
    "assert resnet_activations_path.exists()\n",
    "assert fMRI_folder.exists()\n",
    "\n",
    "layer_names_resnet18 = [ 'conv1', 'maxpool',\n",
    "                        'layer1.0.conv1', 'layer1.0.conv2', 'layer1.1.conv1', 'layer1.1.conv2',\n",
    "                        'layer2.0.conv1', 'layer2.0.conv2', 'layer2.1.conv1', 'layer2.1.conv2',\n",
    "                        'layer3.0.conv1', 'layer3.0.conv2', 'layer3.1.conv1', 'layer3.1.conv2',\n",
    "                        'layer4.0.conv1', 'layer4.0.conv2', 'layer4.1.conv1', 'layer4.1.conv2',\n",
    "                        'avgpool', 'fc']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "sub01_V1 = np.load(fMRI_folder / 'fMRI_val_sub01_subset_V1.npy')\n",
    "sub02_V1 = np.load(fMRI_folder / 'fMRI_val_sub02_subset_V1.npy')\n",
    "sub03_V1 = np.load(fMRI_folder / 'fMRI_val_sub03_subset_V1.npy')\n",
    "\n",
    "sub01_PPA = np.load(fMRI_folder / 'fMRI_val_sub01_subset_lPPA.npy')\n",
    "sub02_PPA = np.load(fMRI_folder / 'fMRI_val_sub02_subset_lPPA.npy')\n",
    "sub03_PPA = np.load(fMRI_folder / 'fMRI_val_sub03_subset_lPPA.npy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "import random\n",
    "\n",
    "def partial_shuffle(a, part=0.5):\n",
    "    # which characters are to be shuffled:\n",
    "    idx_todo = random.sample(range(len(a)), int(len(a) * part))\n",
    "\n",
    "    # what are the new positions of these to-be-shuffled characters:\n",
    "    idx_target = idx_todo[:]\n",
    "    random.shuffle(idx_target)\n",
    "\n",
    "    # map all \"normal\" character positions {0:0, 1:1, 2:2, ...}\n",
    "    mapper = dict((i, i) for i in range(len(a)))\n",
    "\n",
    "    # update with all shuffles in the string: {old_pos:new_pos, old_pos:new_pos, ...}\n",
    "    mapper.update(zip(idx_todo, idx_target))\n",
    "\n",
    "    # use mapper to modify the string:\n",
    "    return [mapper[i] for i in range(len(a))]\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['conv1', 'maxpool', 'layer1.0.conv1']"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "layer_names_resnet18[:3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "resnet_conv1 = np.load(resnet_activations_path / 'conv1' / 'features.npy')\n",
    "resnet_maxpool = np.load(resnet_activations_path / 'maxpool' / 'features.npy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "seeds = [3254,25334,32453,253253,75434]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from collections import defaultdict "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_alignment_scores_partial_shuffle(model_layers, model_act_path, roi, seeds):\n",
    "    \"\"\" This function returns the CKA (biased), CKA (debiased) and RV2 scores for \n",
    "        a specific set of biological responses and model activations for the same stimuli \"\"\"\n",
    "    \n",
    "    # subid = \"01\"\n",
    "    \n",
    "    results = defaultdict(list)\n",
    "    results['biased'] = defaultdict(list)\n",
    "    results['debiased'] = defaultdict(list)\n",
    "    \n",
    "    for i, seed in enumerate(seeds):\n",
    "        np.random.seed(seed)\n",
    "        print(f'Seed: {seed}')\n",
    "\n",
    "        # Random\n",
    "        for sub in [1, 2, 3]:\n",
    "            brain_data =  np.load(fMRI_folder / f'fMRI_val_sub0{sub}_subset_{roi}.npy')\n",
    "            \n",
    "            for j, layer_name in enumerate(model_layers):\n",
    "                X = np.load(model_act_path / layer_name / 'features.npy')\n",
    "                Y = brain_data\n",
    "\n",
    "                for shuffle_percent in [0.0, 0.2, 0.4, 0.6, 0.8, 1.0]:\n",
    "\n",
    "                    idx = list(range(Y.shape[0]))\n",
    "                    idx_shuffled = partial_shuffle(idx, part=shuffle_percent)\n",
    "                    Y = Y[idx_shuffled]\n",
    "                    cka_biased = cka(gram_linear(X), gram_linear(Y))\n",
    "                    cka_debiased = cka(gram_linear(X), gram_linear(Y), debiased=True)\n",
    "\n",
    "                    res_string = f'roi-{roi}_layer-{layer_name}_shuffle-{shuffle_percent}'\n",
    "                    results['biased'][res_string].append(cka_biased)\n",
    "                    results['debiased'][res_string].append(cka_debiased)\n",
    "\n",
    "                    print(f\"sp:{shuffle_percent}, layer: {layer_name}, CKA (biased): {cka_biased:.5f}, CKA (unbiased): {cka_debiased:.5f}\")\n",
    "            \n",
    "    return results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Seed: 3254\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.08564, CKA (unbiased): 0.04018\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.06866, CKA (unbiased): 0.02223\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.05037, CKA (unbiased): 0.00293\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.04837, CKA (unbiased): 0.00091\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.04934, CKA (unbiased): 0.00212\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.04553, CKA (unbiased): -0.00233\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.11538, CKA (unbiased): 0.06322\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.09260, CKA (unbiased): 0.03869\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.07716, CKA (unbiased): 0.02218\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.05988, CKA (unbiased): 0.00299\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.05552, CKA (unbiased): -0.00150\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.05890, CKA (unbiased): 0.00230\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.11856, CKA (unbiased): 0.05991\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.09490, CKA (unbiased): 0.03398\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.07136, CKA (unbiased): 0.00811\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.06677, CKA (unbiased): 0.00259\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.06607, CKA (unbiased): 0.00207\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.06431, CKA (unbiased): -0.00035\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09157, CKA (unbiased): 0.03809\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.08092, CKA (unbiased): 0.02682\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.06590, CKA (unbiased): 0.01115\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.05516, CKA (unbiased): -0.00043\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.05602, CKA (unbiased): 0.00042\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.06350, CKA (unbiased): 0.00842\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.12687, CKA (unbiased): 0.06550\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.10384, CKA (unbiased): 0.04085\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.07908, CKA (unbiased): 0.01402\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.06812, CKA (unbiased): 0.00200\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.06487, CKA (unbiased): -0.00136\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.06559, CKA (unbiased): -0.00074\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.13486, CKA (unbiased): 0.06608\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.11273, CKA (unbiased): 0.04206\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.08452, CKA (unbiased): 0.01073\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.07669, CKA (unbiased): 0.00245\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.07403, CKA (unbiased): -0.00054\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.07577, CKA (unbiased): 0.00120\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.04860, CKA (unbiased): 0.00703\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.04490, CKA (unbiased): 0.00330\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.04021, CKA (unbiased): -0.00159\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.04100, CKA (unbiased): -0.00135\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.04222, CKA (unbiased): 0.00006\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.04160, CKA (unbiased): 0.00026\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.06658, CKA (unbiased): 0.01795\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.06129, CKA (unbiased): 0.01244\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.05323, CKA (unbiased): 0.00328\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.05170, CKA (unbiased): 0.00139\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.05422, CKA (unbiased): 0.00455\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.04979, CKA (unbiased): -0.00047\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.07522, CKA (unbiased): 0.02082\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.06782, CKA (unbiased): 0.01275\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.05784, CKA (unbiased): 0.00141\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.06086, CKA (unbiased): 0.00412\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.05800, CKA (unbiased): 0.00155\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.05664, CKA (unbiased): -0.00040\n",
      "Seed: 25334\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.08564, CKA (unbiased): 0.04018\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.06584, CKA (unbiased): 0.01924\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.05304, CKA (unbiased): 0.00540\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.04616, CKA (unbiased): -0.00172\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.04769, CKA (unbiased): -0.00018\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.04743, CKA (unbiased): 0.00003\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.11538, CKA (unbiased): 0.06322\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.09405, CKA (unbiased): 0.04015\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.07039, CKA (unbiased): 0.01501\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.06082, CKA (unbiased): 0.00460\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.05692, CKA (unbiased): 0.00001\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.05480, CKA (unbiased): -0.00220\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.11856, CKA (unbiased): 0.05991\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.09448, CKA (unbiased): 0.03354\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.07572, CKA (unbiased): 0.01255\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.06593, CKA (unbiased): 0.00152\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.06402, CKA (unbiased): -0.00026\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.06330, CKA (unbiased): -0.00059\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09157, CKA (unbiased): 0.03809\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.08418, CKA (unbiased): 0.03034\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.06759, CKA (unbiased): 0.01263\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.05615, CKA (unbiased): 0.00078\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.05452, CKA (unbiased): -0.00118\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.05535, CKA (unbiased): -0.00060\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.12687, CKA (unbiased): 0.06550\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.10615, CKA (unbiased): 0.04302\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.08052, CKA (unbiased): 0.01571\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.06738, CKA (unbiased): 0.00157\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.06403, CKA (unbiased): -0.00226\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.06468, CKA (unbiased): -0.00168\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.13486, CKA (unbiased): 0.06608\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.11594, CKA (unbiased): 0.04506\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.08804, CKA (unbiased): 0.01410\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.08252, CKA (unbiased): 0.00851\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.07943, CKA (unbiased): 0.00529\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.07344, CKA (unbiased): -0.00214\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.04860, CKA (unbiased): 0.00703\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.04881, CKA (unbiased): 0.00732\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.04702, CKA (unbiased): 0.00517\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.04151, CKA (unbiased): -0.00066\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.04441, CKA (unbiased): 0.00265\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.04122, CKA (unbiased): -0.00050\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.06658, CKA (unbiased): 0.01795\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.05664, CKA (unbiased): 0.00727\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.05586, CKA (unbiased): 0.00587\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.05385, CKA (unbiased): 0.00386\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.05028, CKA (unbiased): 0.00020\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.04960, CKA (unbiased): -0.00031\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.07522, CKA (unbiased): 0.02082\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.06425, CKA (unbiased): 0.00868\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.05858, CKA (unbiased): 0.00225\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.05560, CKA (unbiased): -0.00039\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.05689, CKA (unbiased): 0.00047\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.05560, CKA (unbiased): -0.00076\n",
      "Seed: 32453\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.08564, CKA (unbiased): 0.04018\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.07293, CKA (unbiased): 0.02683\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.05299, CKA (unbiased): 0.00595\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.05135, CKA (unbiased): 0.00437\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.04792, CKA (unbiased): 0.00029\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.04656, CKA (unbiased): -0.00095\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.11538, CKA (unbiased): 0.06322\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.09767, CKA (unbiased): 0.04425\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.06423, CKA (unbiased): 0.00725\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.05528, CKA (unbiased): -0.00182\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.05516, CKA (unbiased): -0.00152\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.05750, CKA (unbiased): 0.00043\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.11856, CKA (unbiased): 0.05991\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.10269, CKA (unbiased): 0.04232\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.07817, CKA (unbiased): 0.01584\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.06806, CKA (unbiased): 0.00418\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.06397, CKA (unbiased): -0.00002\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.06395, CKA (unbiased): 0.00008\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09157, CKA (unbiased): 0.03809\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.07758, CKA (unbiased): 0.02307\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.06160, CKA (unbiased): 0.00617\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.05486, CKA (unbiased): -0.00116\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.05547, CKA (unbiased): -0.00004\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.05496, CKA (unbiased): -0.00068\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.12687, CKA (unbiased): 0.06550\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.11040, CKA (unbiased): 0.04791\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.08080, CKA (unbiased): 0.01587\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.06384, CKA (unbiased): -0.00231\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.06472, CKA (unbiased): -0.00167\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.06781, CKA (unbiased): 0.00143\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.13486, CKA (unbiased): 0.06608\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.11801, CKA (unbiased): 0.04774\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.08735, CKA (unbiased): 0.01434\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.07471, CKA (unbiased): -0.00042\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.07405, CKA (unbiased): -0.00102\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.07428, CKA (unbiased): -0.00039\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.04860, CKA (unbiased): 0.00703\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.04600, CKA (unbiased): 0.00445\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.04227, CKA (unbiased): 0.00060\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.03919, CKA (unbiased): -0.00248\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.04092, CKA (unbiased): -0.00079\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.04217, CKA (unbiased): 0.00038\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.06658, CKA (unbiased): 0.01795\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.06176, CKA (unbiased): 0.01279\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.05212, CKA (unbiased): 0.00243\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.04864, CKA (unbiased): -0.00160\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.04938, CKA (unbiased): -0.00042\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.04954, CKA (unbiased): -0.00017\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.07522, CKA (unbiased): 0.02082\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.06879, CKA (unbiased): 0.01394\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.06439, CKA (unbiased): 0.00895\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.05649, CKA (unbiased): 0.00031\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.06012, CKA (unbiased): 0.00403\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.05910, CKA (unbiased): 0.00322\n"
     ]
    }
   ],
   "source": [
    "results = get_alignment_scores_partial_shuffle(model_layers=layer_names_resnet18[:3],\n",
    "                     model_act_path=resnet_activations_path,\n",
    "                    roi='V1', seeds=seeds[:3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# conv1 \n",
    "\n",
    "sp0_m_b = np.mean(results['biased']['roi-V1_layer-conv1_shuffle-0.0'])\n",
    "sp20_m_b = np.mean(results['biased']['roi-V1_layer-conv1_shuffle-0.2'])\n",
    "sp40_m_b = np.mean(results['biased']['roi-V1_layer-conv1_shuffle-0.4'])\n",
    "sp60_m_b = np.mean(results['biased']['roi-V1_layer-conv1_shuffle-0.6'])\n",
    "sp80_m_b = np.mean(results['biased']['roi-V1_layer-conv1_shuffle-0.8'])\n",
    "sp100_m_b = np.mean(results['biased']['roi-V1_layer-conv1_shuffle-1.0'])\n",
    "\n",
    "sp0_m_d = np.mean(results['debiased']['roi-V1_layer-conv1_shuffle-0.0'])\n",
    "sp20_m_d = np.mean(results['debiased']['roi-V1_layer-conv1_shuffle-0.2'])\n",
    "sp40_m_d = np.mean(results['debiased']['roi-V1_layer-conv1_shuffle-0.4'])\n",
    "sp60_m_d = np.mean(results['debiased']['roi-V1_layer-conv1_shuffle-0.6'])\n",
    "sp80_m_d = np.mean(results['debiased']['roi-V1_layer-conv1_shuffle-0.8'])\n",
    "sp100_m_d = np.mean(results['debiased']['roi-V1_layer-conv1_shuffle-1.0'])\n",
    "\n",
    "sp0_s_b = np.std(results['biased']['roi-V1_layer-conv1_shuffle-0.0'])\n",
    "sp20_s_b = np.std(results['biased']['roi-V1_layer-conv1_shuffle-0.2'])\n",
    "sp40_s_b = np.std(results['biased']['roi-V1_layer-conv1_shuffle-0.4'])\n",
    "sp60_s_b = np.std(results['biased']['roi-V1_layer-conv1_shuffle-0.6'])\n",
    "sp80_s_b = np.std(results['biased']['roi-V1_layer-conv1_shuffle-0.8'])\n",
    "sp100_s_b = np.std(results['biased']['roi-V1_layer-conv1_shuffle-1.0'])\n",
    "\n",
    "sp0_s_d = np.std(results['debiased']['roi-V1_layer-conv1_shuffle-0.0'])\n",
    "sp20_s_d = np.std(results['debiased']['roi-V1_layer-conv1_shuffle-0.2'])\n",
    "sp40_s_d = np.std(results['debiased']['roi-V1_layer-conv1_shuffle-0.4'])\n",
    "sp60_s_d = np.std(results['debiased']['roi-V1_layer-conv1_shuffle-0.6'])\n",
    "sp80_s_d = np.std(results['debiased']['roi-V1_layer-conv1_shuffle-0.8'])\n",
    "sp100_s_d = np.std(results['debiased']['roi-V1_layer-conv1_shuffle-1.0'])\n",
    "\n",
    "conv1_list_V1 = [sp0_m_b, sp0_m_d, sp20_m_b, sp20_m_d, sp40_m_b, sp40_m_d, sp60_m_b, sp60_m_d,sp80_m_b, sp80_m_d,sp100_m_b, sp100_m_d]\n",
    "\n",
    " # maxpool \n",
    "\n",
    "sp0_m_b = np.mean(results['biased']['roi-V1_layer-maxpool_shuffle-0.0'])\n",
    "sp20_m_b = np.mean(results['biased']['roi-V1_layer-maxpool_shuffle-0.2'])\n",
    "sp40_m_b = np.mean(results['biased']['roi-V1_layer-maxpool_shuffle-0.4'])\n",
    "sp60_m_b = np.mean(results['biased']['roi-V1_layer-maxpool_shuffle-0.6'])\n",
    "sp80_m_b = np.mean(results['biased']['roi-V1_layer-maxpool_shuffle-0.8'])\n",
    "sp100_m_b = np.mean(results['biased']['roi-V1_layer-maxpool_shuffle-1.0'])\n",
    "\n",
    "sp0_m_d = np.mean(results['debiased']['roi-V1_layer-maxpool_shuffle-0.0'])\n",
    "sp20_m_d = np.mean(results['debiased']['roi-V1_layer-maxpool_shuffle-0.2'])\n",
    "sp40_m_d = np.mean(results['debiased']['roi-V1_layer-maxpool_shuffle-0.4'])\n",
    "sp60_m_d = np.mean(results['debiased']['roi-V1_layer-maxpool_shuffle-0.6'])\n",
    "sp80_m_d = np.mean(results['debiased']['roi-V1_layer-maxpool_shuffle-0.8'])\n",
    "sp100_m_d = np.mean(results['debiased']['roi-V1_layer-maxpool_shuffle-1.0'])\n",
    "\n",
    "sp0_s_b = np.std(results['biased']['roi-V1_layer-maxpool_shuffle-0.0'])\n",
    "sp20_s_b = np.std(results['biased']['roi-V1_layer-maxpool_shuffle-0.2'])\n",
    "sp40_s_b = np.std(results['biased']['roi-V1_layer-maxpool_shuffle-0.4'])\n",
    "sp60_s_b = np.std(results['biased']['roi-V1_layer-maxpool_shuffle-0.6'])\n",
    "sp80_s_b = np.std(results['biased']['roi-V1_layer-maxpool_shuffle-0.8'])\n",
    "sp100_s_b = np.std(results['biased']['roi-V1_layer-maxpool_shuffle-1.0'])\n",
    "\n",
    "sp0_s_d = np.std(results['debiased']['roi-V1_layer-maxpool_shuffle-0.0'])\n",
    "sp20_s_d = np.std(results['debiased']['roi-V1_layer-maxpool_shuffle-0.2'])\n",
    "sp40_s_d = np.std(results['debiased']['roi-V1_layer-maxpool_shuffle-0.4'])\n",
    "sp60_s_d = np.std(results['debiased']['roi-V1_layer-maxpool_shuffle-0.6'])\n",
    "sp80_s_d = np.std(results['debiased']['roi-V1_layer-maxpool_shuffle-0.8'])\n",
    "sp100_s_d = np.std(results['debiased']['roi-V1_layer-maxpool_shuffle-1.0'])\n",
    "\n",
    "maxpool_list_V1 = [sp0_m_b, sp0_m_d, sp20_m_b, sp20_m_d, sp40_m_b, sp40_m_d, sp60_m_b, sp60_m_d,sp80_m_b, sp80_m_d,sp100_m_b, sp100_m_d]\n",
    "\n",
    " # layer1.0.conv1\n",
    "\n",
    "sp0_m_b = np.mean(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.0'])\n",
    "sp20_m_b = np.mean(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.2'])\n",
    "sp40_m_b = np.mean(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.4'])\n",
    "sp60_m_b = np.mean(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.6'])\n",
    "sp80_m_b = np.mean(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.8'])\n",
    "sp100_m_b = np.mean(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-1.0'])\n",
    "\n",
    "sp0_m_d = np.mean(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.0'])\n",
    "sp20_m_d = np.mean(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.2'])\n",
    "sp40_m_d = np.mean(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.4'])\n",
    "sp60_m_d = np.mean(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.6'])\n",
    "sp80_m_d = np.mean(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.8'])\n",
    "sp100_m_d = np.mean(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-1.0'])\n",
    "\n",
    "sp0_s_b = np.std(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.0'])\n",
    "sp20_s_b = np.std(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.2'])\n",
    "sp40_s_b = np.std(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.4'])\n",
    "sp60_s_b = np.std(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.6'])\n",
    "sp80_s_b = np.std(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.8'])\n",
    "sp100_s_b = np.std(results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-1.0'])\n",
    "\n",
    "sp0_s_d = np.std(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.0'])\n",
    "sp20_s_d = np.std(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.2'])\n",
    "sp40_s_d = np.std(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.4'])\n",
    "sp60_s_d = np.std(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.6'])\n",
    "sp80_s_d = np.std(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.8'])\n",
    "sp100_s_d = np.std(results['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-1.0'])\n",
    "\n",
    "layer1_0_conv1_list_V1 = [sp0_m_b, sp0_m_d, sp20_m_b, sp20_m_d, sp40_m_b, sp40_m_d, sp60_m_b, sp60_m_d,sp80_m_b, sp80_m_d,sp100_m_b, sp100_m_d]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "# conv1 \n",
    "\n",
    "sp0_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-conv1_shuffle-0.0'])\n",
    "sp20_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-conv1_shuffle-0.2'])\n",
    "sp40_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-conv1_shuffle-0.4'])\n",
    "sp60_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-conv1_shuffle-0.6'])\n",
    "sp80_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-conv1_shuffle-0.8'])\n",
    "sp100_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-conv1_shuffle-1.0'])\n",
    "\n",
    "sp0_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-conv1_shuffle-0.0'])\n",
    "sp20_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-conv1_shuffle-0.2'])\n",
    "sp40_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-conv1_shuffle-0.4'])\n",
    "sp60_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-conv1_shuffle-0.6'])\n",
    "sp80_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-conv1_shuffle-0.8'])\n",
    "sp100_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-conv1_shuffle-1.0'])\n",
    "\n",
    "conv1_list_lPPA = [sp0_m_b_lPPA, sp0_m_d_lPPA, sp20_m_b_lPPA, sp20_m_d_lPPA, sp40_m_b_lPPA, sp40_m_d_lPPA, sp60_m_b_lPPA, sp60_m_d_lPPA,sp80_m_b_lPPA, sp80_m_d_lPPA,sp100_m_b_lPPA, sp100_m_d_lPPA]\n",
    "\n",
    "#sp0_s_b = np.std(results_lPPA['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.0'])\n",
    "#sp20_s_b = np.std(results_lPPA['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.2'])\n",
    "#sp40_s_b = np.std(results_lPPA['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.4'])\n",
    "#sp60_s_b = np.std(results_lPPA['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.6'])\n",
    "#sp80_s_b = np.std(results_lPPA['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.8'])\n",
    "#sp100_s_b = np.std(results_lPPA['biased']['roi-V1_layer-layer1.0.conv1_shuffle-1.0'])\n",
    "\n",
    "#sp0_s_d = np.std(results_lPPA['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.0'])\n",
    "#sp20_s_d = np.std(results_lPPA['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.2'])\n",
    "#sp40_s_d = np.std(results_lPPA['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.4'])\n",
    "#sp60_s_d = np.std(results_lPPA['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.6'])\n",
    "#sp80_s_d = np.std(results_lPPA['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-0.8'])\n",
    "#sp100_s_d = np.std(results_lPPA['debiased']['roi-V1_layer-layer1.0.conv1_shuffle-1.0'])\n",
    "\n",
    "# maxpool \n",
    "\n",
    "sp0_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-maxpool_shuffle-0.0'])\n",
    "sp20_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-maxpool_shuffle-0.2'])\n",
    "sp40_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-maxpool_shuffle-0.4'])\n",
    "sp60_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-maxpool_shuffle-0.6'])\n",
    "sp80_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-maxpool_shuffle-0.8'])\n",
    "sp100_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-maxpool_shuffle-1.0'])\n",
    "\n",
    "sp0_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-maxpool_shuffle-0.0'])\n",
    "sp20_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-maxpool_shuffle-0.2'])\n",
    "sp40_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-maxpool_shuffle-0.4'])\n",
    "sp60_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-maxpool_shuffle-0.6'])\n",
    "sp80_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-maxpool_shuffle-0.8'])\n",
    "sp100_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-maxpool_shuffle-1.0'])\n",
    "\n",
    "maxpool_list_lPPA = [sp0_m_b_lPPA, sp0_m_d_lPPA, sp20_m_b_lPPA, sp20_m_d_lPPA, sp40_m_b_lPPA, sp40_m_d_lPPA, sp60_m_b_lPPA, sp60_m_d_lPPA,sp80_m_b_lPPA, sp80_m_d_lPPA,sp100_m_b_lPPA, sp100_m_d_lPPA]\n",
    "\n",
    "# layer1.0.conv1\n",
    "\n",
    "sp0_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.0'])\n",
    "sp20_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.2'])\n",
    "sp40_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.4'])\n",
    "sp60_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.6'])\n",
    "sp80_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.8'])\n",
    "sp100_m_b_lPPA = np.mean(results_lPPA['biased']['roi-lPPA_layer-layer1.0.conv1_shuffle-1.0'])\n",
    "\n",
    "sp0_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.0'])\n",
    "sp20_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.2'])\n",
    "sp40_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.4'])\n",
    "sp60_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.6'])\n",
    "sp80_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-layer1.0.conv1_shuffle-0.8'])\n",
    "sp100_m_d_lPPA = np.mean(results_lPPA['debiased']['roi-lPPA_layer-layer1.0.conv1_shuffle-1.0'])\n",
    "\n",
    "layer1_0_conv1_list_lPPA = [sp0_m_b_lPPA, sp0_m_d_lPPA, sp20_m_b_lPPA, sp20_m_d_lPPA, sp40_m_b_lPPA, sp40_m_d_lPPA, sp60_m_b_lPPA, sp60_m_d_lPPA,sp80_m_b_lPPA, sp80_m_d_lPPA,sp100_m_b_lPPA, sp100_m_d_lPPA]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.11790842,\n",
       " 0.15689422,\n",
       " 0.10963023,\n",
       " 0.11790842,\n",
       " 0.15689422,\n",
       " 0.10963023,\n",
       " 0.11790842,\n",
       " 0.15689422,\n",
       " 0.10963023]"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results_lPPA['biased']['roi-lPPA_layer-maxpool_shuffle-0.0']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.0023072674,\n",
       " 0.0023072674,\n",
       " 0.0019855925,\n",
       " 0.0019855925,\n",
       " 0.0016312553,\n",
       " 0.0016312553,\n",
       " 0.00018394663,\n",
       " 0.00018394663,\n",
       " -0.0005958044,\n",
       " -0.0005958044,\n",
       " -0.00021666203,\n",
       " -0.00021666203]"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "maxpool_list_lPPA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Seed: 3254\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09910, CKA (unbiased): 0.00339\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.09846, CKA (unbiased): 0.00259\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.09974, CKA (unbiased): 0.00397\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.09674, CKA (unbiased): 0.00060\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.10032, CKA (unbiased): 0.00503\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.09753, CKA (unbiased): 0.00154\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.11791, CKA (unbiased): 0.00366\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.11845, CKA (unbiased): 0.00454\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.11830, CKA (unbiased): 0.00422\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.11413, CKA (unbiased): -0.00036\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.11325, CKA (unbiased): -0.00156\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.11254, CKA (unbiased): -0.00212\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.13326, CKA (unbiased): 0.00366\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.12979, CKA (unbiased): -0.00030\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.12845, CKA (unbiased): -0.00170\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.12934, CKA (unbiased): -0.00018\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.12921, CKA (unbiased): -0.00018\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.12892, CKA (unbiased): 0.00035\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.13399, CKA (unbiased): 0.00318\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.13503, CKA (unbiased): 0.00449\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.13241, CKA (unbiased): 0.00130\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.13429, CKA (unbiased): 0.00374\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.13024, CKA (unbiased): -0.00136\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.13158, CKA (unbiased): 0.00036\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.15689, CKA (unbiased): 0.00046\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.15920, CKA (unbiased): 0.00320\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.15732, CKA (unbiased): 0.00104\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.15386, CKA (unbiased): -0.00311\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.15685, CKA (unbiased): 0.00050\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.15431, CKA (unbiased): -0.00260\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.17591, CKA (unbiased): -0.00151\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.17660, CKA (unbiased): -0.00047\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.17699, CKA (unbiased): 0.00029\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.17552, CKA (unbiased): -0.00145\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.17636, CKA (unbiased): -0.00052\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.17567, CKA (unbiased): -0.00136\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09252, CKA (unbiased): 0.00280\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.09060, CKA (unbiased): 0.00046\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.09112, CKA (unbiased): 0.00108\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.09073, CKA (unbiased): 0.00052\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.08946, CKA (unbiased): -0.00058\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.08931, CKA (unbiased): -0.00143\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.10963, CKA (unbiased): 0.00280\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.11018, CKA (unbiased): 0.00348\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.10767, CKA (unbiased): 0.00029\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.10829, CKA (unbiased): 0.00105\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.10669, CKA (unbiased): -0.00068\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.10741, CKA (unbiased): -0.00012\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.12278, CKA (unbiased): 0.00178\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.12225, CKA (unbiased): 0.00123\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.12064, CKA (unbiased): -0.00066\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.11855, CKA (unbiased): -0.00379\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.12177, CKA (unbiased): 0.00116\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.12000, CKA (unbiased): -0.00210\n",
      "Seed: 25334\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09910, CKA (unbiased): 0.00339\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.09978, CKA (unbiased): 0.00432\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.09609, CKA (unbiased): 0.00016\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.09906, CKA (unbiased): 0.00364\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.09774, CKA (unbiased): 0.00225\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.09442, CKA (unbiased): -0.00169\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.11791, CKA (unbiased): 0.00366\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.11697, CKA (unbiased): 0.00242\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.11569, CKA (unbiased): 0.00124\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.11390, CKA (unbiased): -0.00085\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.11103, CKA (unbiased): -0.00348\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.11197, CKA (unbiased): -0.00266\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.13326, CKA (unbiased): 0.00366\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.13197, CKA (unbiased): 0.00192\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.13148, CKA (unbiased): 0.00162\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.12811, CKA (unbiased): -0.00160\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.12930, CKA (unbiased): 0.00010\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.12691, CKA (unbiased): -0.00277\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.13399, CKA (unbiased): 0.00318\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.13377, CKA (unbiased): 0.00275\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.13102, CKA (unbiased): -0.00068\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.13045, CKA (unbiased): -0.00131\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.13220, CKA (unbiased): 0.00089\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.13073, CKA (unbiased): -0.00085\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.15689, CKA (unbiased): 0.00046\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.15775, CKA (unbiased): 0.00118\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.15909, CKA (unbiased): 0.00309\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.15792, CKA (unbiased): 0.00180\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.15602, CKA (unbiased): -0.00048\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.15636, CKA (unbiased): -0.00066\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.17591, CKA (unbiased): -0.00151\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.17462, CKA (unbiased): -0.00307\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.17746, CKA (unbiased): -0.00000\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.17455, CKA (unbiased): -0.00365\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.18021, CKA (unbiased): 0.00393\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.17563, CKA (unbiased): -0.00151\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09252, CKA (unbiased): 0.00280\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.09216, CKA (unbiased): 0.00225\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.08706, CKA (unbiased): -0.00336\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.09012, CKA (unbiased): -0.00014\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.08855, CKA (unbiased): -0.00224\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.09265, CKA (unbiased): 0.00260\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.10963, CKA (unbiased): 0.00280\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.10723, CKA (unbiased): -0.00011\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.10868, CKA (unbiased): 0.00138\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.10632, CKA (unbiased): -0.00151\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.10814, CKA (unbiased): 0.00027\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.11196, CKA (unbiased): 0.00516\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.12278, CKA (unbiased): 0.00178\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.12356, CKA (unbiased): 0.00237\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.12442, CKA (unbiased): 0.00362\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.12290, CKA (unbiased): 0.00171\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.12128, CKA (unbiased): 0.00056\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.12328, CKA (unbiased): 0.00192\n",
      "Seed: 32453\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09910, CKA (unbiased): 0.00339\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.09728, CKA (unbiased): 0.00145\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.09484, CKA (unbiased): -0.00097\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.09317, CKA (unbiased): -0.00304\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.09378, CKA (unbiased): -0.00273\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.09668, CKA (unbiased): 0.00078\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.11791, CKA (unbiased): 0.00366\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.11668, CKA (unbiased): 0.00220\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.11552, CKA (unbiased): 0.00103\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.11749, CKA (unbiased): 0.00303\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.11625, CKA (unbiased): 0.00192\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.11379, CKA (unbiased): -0.00087\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.13326, CKA (unbiased): 0.00366\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.13227, CKA (unbiased): 0.00269\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.13104, CKA (unbiased): 0.00124\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.13034, CKA (unbiased): 0.00136\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.12956, CKA (unbiased): -0.00027\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.12855, CKA (unbiased): -0.00069\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.13399, CKA (unbiased): 0.00318\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.13259, CKA (unbiased): 0.00162\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.13093, CKA (unbiased): -0.00073\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.13198, CKA (unbiased): 0.00062\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.13007, CKA (unbiased): -0.00142\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.13135, CKA (unbiased): 0.00012\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.15689, CKA (unbiased): 0.00046\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.15712, CKA (unbiased): 0.00072\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.15862, CKA (unbiased): 0.00260\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.15649, CKA (unbiased): -0.00035\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.15617, CKA (unbiased): -0.00008\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.15823, CKA (unbiased): 0.00207\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.17591, CKA (unbiased): -0.00151\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.17579, CKA (unbiased): -0.00175\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.17735, CKA (unbiased): 0.00001\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.18006, CKA (unbiased): 0.00383\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.17588, CKA (unbiased): -0.00204\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.17529, CKA (unbiased): -0.00284\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09252, CKA (unbiased): 0.00280\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.09257, CKA (unbiased): 0.00307\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.09043, CKA (unbiased): 0.00005\n",
      "sp:0.6, layer: conv1, CKA (biased): 0.09430, CKA (unbiased): 0.00432\n",
      "sp:0.8, layer: conv1, CKA (biased): 0.09039, CKA (unbiased): 0.00021\n",
      "sp:1.0, layer: conv1, CKA (biased): 0.08798, CKA (unbiased): -0.00249\n",
      "sp:0.0, layer: maxpool, CKA (biased): 0.10963, CKA (unbiased): 0.00280\n",
      "sp:0.2, layer: maxpool, CKA (biased): 0.10747, CKA (unbiased): 0.00023\n",
      "sp:0.4, layer: maxpool, CKA (biased): 0.10695, CKA (unbiased): -0.00021\n",
      "sp:0.6, layer: maxpool, CKA (biased): 0.10937, CKA (unbiased): 0.00196\n",
      "sp:0.8, layer: maxpool, CKA (biased): 0.10611, CKA (unbiased): -0.00177\n",
      "sp:1.0, layer: maxpool, CKA (biased): 0.10788, CKA (unbiased): -0.00016\n",
      "sp:0.0, layer: layer1.0.conv1, CKA (biased): 0.12278, CKA (unbiased): 0.00178\n",
      "sp:0.2, layer: layer1.0.conv1, CKA (biased): 0.12394, CKA (unbiased): 0.00251\n",
      "sp:0.4, layer: layer1.0.conv1, CKA (biased): 0.12168, CKA (unbiased): -0.00001\n",
      "sp:0.6, layer: layer1.0.conv1, CKA (biased): 0.12081, CKA (unbiased): -0.00089\n",
      "sp:0.8, layer: layer1.0.conv1, CKA (biased): 0.12105, CKA (unbiased): -0.00144\n",
      "sp:1.0, layer: layer1.0.conv1, CKA (biased): 0.12138, CKA (unbiased): -0.00107\n"
     ]
    }
   ],
   "source": [
    "results_lPPA = get_alignment_scores_partial_shuffle(model_layers=layer_names_resnet18[:3],\n",
    "                     model_act_path=resnet_activations_path,\n",
    "                    roi='lPPA', seeds=seeds[:3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pickle\n",
    "pickle.dump(results, open('fMRI_V1_partial_shuffle_results.pkl', 'wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Doesn't give same results as long version, need to check this out\n",
    "def get_mean_std(results, name):\n",
    "    sps =  [0.2, 0.4, 0.6, 0.8, 1.0]\n",
    "    means_b = [np.mean(results['biased'][f'{name}-{x}']) for x in sps]\n",
    "    stds_b = [np.std(results['biased'][f'{name}-{x}']) for x in sps]\n",
    "    means_d = [np.mean(results['debiased'][f'{name}-{x}']) for x in sps]\n",
    "    stds_d = [np.std(results['debiased'][f'{name}-{x}']) for x in sps]\n",
    "    print(means_b)\n",
    "    print(means_d)\n",
    "    a = list(zip(means_b, means_d))\n",
    "    b = list(zip(stds_b, stds_d))\n",
    "    print(a)\n",
    "    print(b)\n",
    "    res1 = list(chain.from_iterable(a))\n",
    "    res2 = list(chain.from_iterable(b))\n",
    "    print(res1)\n",
    "    print(res2)\n",
    "    return res1, res2\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.06628909, 0.053594984, 0.049072422, 0.0479189, 0.047660507]\n",
      "[0.018932577, 0.0056203436, 0.00071149936, -0.00051165576, -0.0006479858]\n",
      "[(0.06628909, 0.018932577), (0.053594984, 0.0056203436), (0.049072422, 0.00071149936), (0.0479189, -0.00051165576), (0.047660507, -0.0006479858)]\n",
      "[(0.0152934585, 0.011358516), (0.009880139, 0.0050973655), (0.0074063903, 0.002468361), (0.005743353, 0.0014449576), (0.0056424504, 0.0015699931)]\n",
      "[0.06628909, 0.018932577, 0.053594984, 0.0056203436, 0.049072422, 0.00071149936, 0.0479189, -0.00051165576, 0.047660507, -0.0006479858]\n",
      "[0.0152934585, 0.011358516, 0.009880139, 0.0050973655, 0.0074063903, 0.002468361, 0.005743353, 0.0014449576, 0.0056424504, 0.0015699931]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.054225877"
      ]
     },
     "execution_count": 99,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "conv1_m, conv1_s = get_mean_std(results, 'roi-V1_layer-conv1_shuffle')\n",
    "\n",
    "results['biased']['roi-V1_layer-layer1.0.conv1_shuffle-0.8'][5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [],
   "source": [
    "conv1_m, conv1_s = get_mean_std(results, 'roi-V1_layer-conv1_shuffle')\n",
    "maxpool_m, maxpool_s = get_mean_std(results, 'roi-V1_layer-maxpool_shuffle')\n",
    "layer1_0_conv1_m, layer1_0_conv1_s = get_mean_std(results, 'roi-V1_layer-layer1.0.conv1_shuffle')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Seed: 3254\n",
      "sp:0.0, layer: conv1, CKA (biased): 0.09910, CKA (unbiased): 0.00339\n",
      "sp:0.2, layer: conv1, CKA (biased): 0.09688, CKA (unbiased): 0.00100\n",
      "sp:0.4, layer: conv1, CKA (biased): 0.09359, CKA (unbiased): -0.00245\n"
     ]
    }
   ],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then make this nice, add error bars, do the same for next 2 layers.\n",
    "\n",
    "Then do the same with lPPA and things should look equal, make claim that because no debiased CKA anyway, things don't change depending on shuffle amount."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_7103/2492261995.py:29: MatplotlibDeprecationWarning: savefig() got unexpected keyword argument \"type\" which is no longer supported as of 3.3 and will become an error in 3.6\n",
      "  plt.savefig('partial_shuffling_figure.svg', type='svg')\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1400x500 with 6 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt \n",
    "plt.rcParams['svg.fonttype'] = 'none'\n",
    "\n",
    "width = 0.3\n",
    "alpha_biased = 0.4\n",
    "alpha_debiased = 1.0\n",
    "errorbar_alpha = 0.4\n",
    "offset = 0.3\n",
    "\n",
    "alphas = [0.5, 1, 0.5, 1, 0.5, 1, 0.5, 1, 0.5, 1]\n",
    "\n",
    "fig, axes = plt.subplots(2,3, figsize=(14,5), sharex=True, sharey=True)\n",
    "\n",
    "axes[0,0].bar(range(len(conv1_list_V1)), conv1_list_V1)\n",
    "axes[0,1].bar(range(len(maxpool_list_V1)), maxpool_list_V1)\n",
    "axes[0,2].bar(range(len(layer1_0_conv1_list_V1)), layer1_0_conv1_list_V1)\n",
    "\n",
    "axes[1,0].bar(range(len(conv1_list_lPPA)), conv1_list_lPPA)\n",
    "axes[1,1].bar(range(len(maxpool_list_lPPA)), maxpool_list_lPPA)\n",
    "axes[1,2].bar(range(len(layer1_0_conv1_list_lPPA)), layer1_0_conv1_list_lPPA)\n",
    "\n",
    "axes[0,1].set_title('Partial Shuffling Experiment')\n",
    "axes[0,0].set_ylabel('CKA')\n",
    "axes[1,0].set_ylabel('CKA')\n",
    "\n",
    "axes[1,0].set_xticks([0,2,4,6,8,10])\n",
    "axes[1,0].set_xticklabels(['0%', '20%', '40%', '60%', '80%', '100%'])\n",
    "\n",
    "plt.savefig('partial_shuffling_figure.svg', type='svg')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.0013103224,\n",
       " 0.0013103224,\n",
       " 0.0005711557,\n",
       " 0.0005711557,\n",
       " 0.000488891,\n",
       " 0.000488891,\n",
       " -0.0005176265,\n",
       " -0.0005176265,\n",
       " 0.00014386939,\n",
       " 0.00014386939,\n",
       " -0.0011188588,\n",
       " -0.0011188588]"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "layer1_0_conv1_list_lPPA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<BarContainer object of 10 artists>"
      ]
     },
     "execution_count": 95,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA04AAAGsCAYAAADqs/chAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8qNh9FAAAACXBIWXMAAA9hAAAPYQGoP6dpAAA7lUlEQVR4nO3df3RU9Z3/8VcSyAx4SKzmkCEQiLpIQCIpPxITqWn35HSoaTX2FGLqgTRlsW6JgnGjBIG0i22oRzihkpqlW6RdZUPZSpZCTmyMgtUEKUk4SmXxFxoqnQl01wRjTTBzv3/4ZXRk4HLzc27m+ThnjuXO+17fH+/Mu3lxJ3ciDMMwBAAAAAC4qMjhbgAAAAAAQh3BCQAAAABMEJwAAAAAwATBCQAAAABMEJwAAAAAwATBCQAAAABMEJwAAAAAwMSo4W5gIPh8Pp06dUrjxo1TRETEcLcDhDXDMHT27FklJCQoMtI+fzfDHAFCB3MEQH8M1gwZEcHp1KlTSkxMHO42AHzOyZMnNWnSpOFu47IxR4DQwxwB0B8DPUNGRHAaN26cpE//48TExAxzN0B46+zsVGJiov99aRfMESB0MEcA9MdgzZAREZzOXw6PiYlhUAEhwm4fU2GOAKGHOQKgPwZ6htjng8MAAAAAMEwITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACZGxF31AAAA7CRp1b5+H+PdDTkD0AmAy8UVJwAAAAAwQXACAAAAABMEJwAAAAAwQXACAAAAABMEJwAAAAAwQXACAAAAABMEJwAAAAAwQXACAAAAABMEJwAAAAAwQXACAAAAABMEJwAAAAAwMWq4GwAAu0lata/fx3h3Q84AdAIAAIYKV5wAAAAAwATBCQAAAABMEJwAAAAAwATBCQAAAABMEJwAAAAAwATBCQAAAABMEJwAAAAAwESfglNlZaWSkpLkdDqVnp6uQ4cOXbJ+165dSk5OltPpVEpKimprawOe//DDD1VUVKRJkyZpzJgxmjFjhqqqqvrSGgAAAAAMOMvBaefOnSouLlZZWZlaWlo0a9Ysud1utbe3B61vbGxUfn6+li5dqtbWVuXm5io3N1dHjx711xQXF6uurk5PPfWUjh07ppUrV6qoqEh79uzp+8oAAAAAYIBYDk6bNm3SsmXLVFhY6L8yNHbsWG3bti1o/ebNm7VgwQKVlJRo+vTpWr9+vWbPnq0tW7b4axobG1VQUKCvfvWrSkpK0t13361Zs2Zd9EpWd3e3Ojs7Ax4AYAVzBEB/MUeA8GIpOPX09Ki5uVnZ2dmfHSAyUtnZ2Wpqagq6T1NTU0C9JLnd7oD6zMxM7dmzR++//74Mw9ALL7ygN954Q1//+teDHrO8vFyxsbH+R2JiopVlAABzBEC/MUeA8GIpOJ05c0a9vb2Kj48P2B4fHy+PxxN0H4/HY1r/+OOPa8aMGZo0aZKio6O1YMECVVZW6pZbbgl6zNLSUnV0dPgfJ0+etLIMAGCOAOg35ggQXkYNdwPSp8Hp4MGD2rNnj6ZMmaIXX3xRy5cvV0JCwgVXqyTJ4XDI4XAMQ6cARgrmCID+Yo4A4cVScIqLi1NUVJS8Xm/Adq/XK5fLFXQfl8t1yfq///3vWr16tXbv3q2cnBxJ0o033qgjR47oscceCxqcAAAAAGAoWfqoXnR0tObMmaOGhgb/Np/Pp4aGBmVkZATdJyMjI6Bekurr6/31586d07lz5xQZGdhKVFSUfD6flfYAAAAAYFBY/qhecXGxCgoKNHfuXKWlpamiokJdXV0qLCyUJC1ZskQTJ05UeXm5JGnFihXKysrSxo0blZOTo+rqah0+fFhbt26VJMXExCgrK0slJSUaM2aMpkyZogMHDug3v/mNNm3aNIBLBQAAAIC+sRyc8vLydPr0aa1bt04ej0epqamqq6vz3wCira0t4OpRZmamduzYoTVr1mj16tWaOnWqampqNHPmTH9NdXW1SktLddddd+l///d/NWXKFP3kJz/RPffcMwBLBAAAQF8krdrX72O8uyFnADoBhl+fbg5RVFSkoqKioM/t37//gm0LFy7UwoULL3o8l8ulJ598si+tAAAAAMCgs/wFuAAAAAAQbghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJkYNdwMAgMGTtGpfv4/x7oacAegEAAB744oTAAAAAJjgihMAAABsjavrGApccQIAAAAAE2F5xYm/lQAAAABgRVgGJwAAAAAXx4WGC/FRPQAAAAAwwRUnAAAAYIhwJce+uOIEAAAAACYITgAAAABgguAEAAAAACb6FJwqKyuVlJQkp9Op9PR0HTp06JL1u3btUnJyspxOp1JSUlRbW3tBzbFjx3TbbbcpNjZWV1xxhebNm6e2tra+tAcAAAAAA8pycNq5c6eKi4tVVlamlpYWzZo1S263W+3t7UHrGxsblZ+fr6VLl6q1tVW5ubnKzc3V0aNH/TVvv/225s+fr+TkZO3fv1+vvvqq1q5dK6fT2feVAQAAAMAAsRycNm3apGXLlqmwsFAzZsxQVVWVxo4dq23btgWt37x5sxYsWKCSkhJNnz5d69ev1+zZs7VlyxZ/zcMPP6xbb71Vjz76qL785S/ruuuu02233abx48f3fWUAAAAAMEAs3Y68p6dHzc3NKi0t9W+LjIxUdna2mpqagu7T1NSk4uLigG1ut1s1NTWSJJ/Pp3379unBBx+U2+1Wa2urrrnmGpWWlio3NzfoMbu7u9Xd3e3/c2dnp5VlAABzJMRwe17YEXMECC+WrjidOXNGvb29io+PD9geHx8vj8cTdB+Px3PJ+vb2dn344YfasGGDFixYoD/84Q+644479O1vf1sHDhwIeszy8nLFxsb6H4mJiVaWAQDMEQD9xhwBwsuw31XP5/NJkm6//Xbdf//9Sk1N1apVq/TNb35TVVVVQfcpLS1VR0eH/3Hy5MmhbBnACMAcAdBfzBEgvFj6qF5cXJyioqLk9XoDtnu9XrlcrqD7uFyuS9bHxcVp1KhRmjFjRkDN9OnT9dJLLwU9psPhkMPhsNI6AARgjgDoL+YIEF4sBafo6GjNmTNHDQ0N/t8/8vl8amhoUFFRUdB9MjIy1NDQoJUrV/q31dfXKyMjw3/MefPm6fjx4wH7vfHGG5oyZYqV9gAAGFT8LhYAhC9LwUmSiouLVVBQoLlz5yotLU0VFRXq6upSYWGhJGnJkiWaOHGiysvLJUkrVqxQVlaWNm7cqJycHFVXV+vw4cPaunWr/5glJSXKy8vTLbfcoq997Wuqq6vT73//e+3fv39gVgkAQJjpb8gj4AFAIMvBKS8vT6dPn9a6devk8XiUmpqquro6/w0g2traFBn52a9OZWZmaseOHVqzZo1Wr16tqVOnqqamRjNnzvTX3HHHHaqqqlJ5ebnuu+8+TZs2Tb/73e80f/78AVgiAAAAAPSP5eAkSUVFRRf9aF6wq0QLFy7UwoULL3nM73//+/r+97/fl3YAAAAAYFAN+131AAAAACDUEZwAAAAAwATBCQAAAABMEJwAAAAAwESfbg6BocH3hQAAAAChgStOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJrgdOfqN26YDAABgpOOKEwAAAACY4IoTwgpXxwAAANAXBCcgBBHwAAAAQgsf1QMAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADDRp+BUWVmppKQkOZ1Opaen69ChQ5es37Vrl5KTk+V0OpWSkqLa2tqL1t5zzz2KiIhQRUVFX1oDAAAAgAFnOTjt3LlTxcXFKisrU0tLi2bNmiW326329vag9Y2NjcrPz9fSpUvV2tqq3Nxc5ebm6ujRoxfU7t69WwcPHlRCQoL1lQAAAADAILEcnDZt2qRly5apsLBQM2bMUFVVlcaOHatt27YFrd+8ebMWLFigkpISTZ8+XevXr9fs2bO1ZcuWgLr3339f9957r55++mmNHj26b6sBAAAAgEFgKTj19PSoublZ2dnZnx0gMlLZ2dlqamoKuk9TU1NAvSS53e6Aep/Pp8WLF6ukpEQ33HCDaR/d3d3q7OwMeACAFcwRAP3FHAHCi6XgdObMGfX29io+Pj5ge3x8vDweT9B9PB6Paf3PfvYzjRo1Svfdd99l9VFeXq7Y2Fj/IzEx0coyAIA5AqDfmCNAeBn2u+o1Nzdr8+bN2r59uyIiIi5rn9LSUnV0dPgfJ0+eHOQuAYw0zBEA/cUcAcLLKCvFcXFxioqKktfrDdju9XrlcrmC7uNyuS5Z/8c//lHt7e2aPHmy//ne3l498MADqqio0LvvvnvBMR0OhxwOh5XWASAAcwRAfzFHgPBi6YpTdHS05syZo4aGBv82n8+nhoYGZWRkBN0nIyMjoF6S6uvr/fWLFy/Wq6++qiNHjvgfCQkJKikp0bPPPmt1PQAAAAAw4CxdcZKk4uJiFRQUaO7cuUpLS1NFRYW6urpUWFgoSVqyZIkmTpyo8vJySdKKFSuUlZWljRs3KicnR9XV1Tp8+LC2bt0qSbr66qt19dVXB/w7Ro8eLZfLpWnTpvV3fQAAAADQb5aDU15enk6fPq1169bJ4/EoNTVVdXV1/htAtLW1KTLyswtZmZmZ2rFjh9asWaPVq1dr6tSpqqmp0cyZMwduFQAAAAAwiCwHJ0kqKipSUVFR0Of2799/wbaFCxdq4cKFl338YL/XBAAAAADDZdjvqgcAAAAAoY7gBAAAAAAmCE4AAAAAYILgBAAAAAAmCE4AAAAAYILgBAAAAAAmCE4AAAAAYKJP3+MEAACA0JK0al+/j/HuhpwB6AQYmQhOABAi+vtDDz/wAAAwePioHgAAAACYIDgBAAAAgAmCEwAAAACYIDgBAAAAgAmCEwAAAACYIDgBAAAAgAmCEwAAAACYIDgBAAAAgAmCEwAAAACYIDgBAAAAgAmCEwAAAACYIDgBAAAAgAmCEwAAAACY6FNwqqysVFJSkpxOp9LT03Xo0KFL1u/atUvJyclyOp1KSUlRbW2t/7lz587poYceUkpKiq644golJCRoyZIlOnXqVF9aAwAAAIABZzk47dy5U8XFxSorK1NLS4tmzZolt9ut9vb2oPWNjY3Kz8/X0qVL1draqtzcXOXm5uro0aOSpI8++kgtLS1au3atWlpa9Mwzz+j48eO67bbb+rcyAAAAABggloPTpk2btGzZMhUWFmrGjBmqqqrS2LFjtW3btqD1mzdv1oIFC1RSUqLp06dr/fr1mj17trZs2SJJio2NVX19vRYtWqRp06bppptu0pYtW9Tc3Ky2trb+rQ4AAAAABoCl4NTT06Pm5mZlZ2d/doDISGVnZ6upqSnoPk1NTQH1kuR2uy9aL0kdHR2KiIjQlVdeGfT57u5udXZ2BjwAwArmCID+Yo4A4cVScDpz5ox6e3sVHx8fsD0+Pl4ejyfoPh6Px1L9xx9/rIceekj5+fmKiYkJWlNeXq7Y2Fj/IzEx0coyAIA5AqDfmCNAeAmpu+qdO3dOixYtkmEYeuKJJy5aV1paqo6ODv/j5MmTQ9glgJGAOQKgv5gjQHgZZaU4Li5OUVFR8nq9Adu9Xq9cLlfQfVwu12XVnw9N7733np5//vmLXm2SJIfDIYfDYaX1QZe0al+/j/HuhpwB6ATA5QjFOQLAXpgjQHixdMUpOjpac+bMUUNDg3+bz+dTQ0ODMjIygu6TkZERUC9J9fX1AfXnQ9Obb76p5557TldffbWVtgAAAABgUFm64iRJxcXFKigo0Ny5c5WWlqaKigp1dXWpsLBQkrRkyRJNnDhR5eXlkqQVK1YoKytLGzduVE5Ojqqrq3X48GFt3bpV0qeh6Tvf+Y5aWlq0d+9e9fb2+n//6aqrrlJ0dPRArRUAAAAA+sRycMrLy9Pp06e1bt06eTwepaamqq6uzn8DiLa2NkVGfnYhKzMzUzt27NCaNWu0evVqTZ06VTU1NZo5c6Yk6f3339eePXskSampqQH/rhdeeEFf/epX+7g0AAAAABgYloOTJBUVFamoqCjoc/v3779g28KFC7Vw4cKg9UlJSTIMoy9tAAAAAMCQCKm76gEAAABAKCI4AQAAAIAJghMAAAAAmCA4AQAAAIAJghMAAAAAmCA4AQAAAICJPt2OHAAAAOiLpFX7+n2MdzfkDEAngDUEJwCAJfzQAwAIR3xUDwAAAABMEJwAAAAAwATBCQAAAABMEJwAAAAAwATBCQAAAABMEJwAAAAAwATBCQAAAABMEJwAAAAAwATBCQAAAABMjBruBjC0klbt6/cx3t2QMwCdAAAAAPbBFScAAAAAMEFwAgAAAAATBCcAAAAAMMHvOAEARiR+pxMAMJAITgAAAEAQ/AUMPq9PH9WrrKxUUlKSnE6n0tPTdejQoUvW79q1S8nJyXI6nUpJSVFtbW3A84ZhaN26dZowYYLGjBmj7Oxsvfnmm31pDQAAAAAGnOXgtHPnThUXF6usrEwtLS2aNWuW3G632tvbg9Y3NjYqPz9fS5cuVWtrq3Jzc5Wbm6ujR4/6ax599FH9/Oc/V1VVlV555RVdccUVcrvd+vjjj/u+MgAAAAAYIJaD06ZNm7Rs2TIVFhZqxowZqqqq0tixY7Vt27ag9Zs3b9aCBQtUUlKi6dOna/369Zo9e7a2bNki6dOrTRUVFVqzZo1uv/123XjjjfrNb36jU6dOqaampl+LAwAAAICBYOl3nHp6etTc3KzS0lL/tsjISGVnZ6upqSnoPk1NTSouLg7Y5na7/aHoxIkT8ng8ys7O9j8fGxur9PR0NTU16c4777zgmN3d3eru7vb/ubOz08oyYBP9/VwxnynGpTBHECr4HQr7Yo4A1th93kUYhmFcbvGpU6c0ceJENTY2KiMjw7/9wQcf1IEDB/TKK69csE90dLR+/etfKz8/37/tF7/4hX784x/L6/WqsbFRN998s06dOqUJEyb4axYtWqSIiAjt3LnzgmP+6Ec/0o9//OMLtnd0dCgmJuZylwMMCLsMgaHqs7OzU7GxsSH/fmSOAKGLOQJYY5efRYbKYM0QW95Vr7S0NOAqVmdnpxITE4exI4Qzuwwau/Q5VJgjAPqLOYJQwf/HDw1LwSkuLk5RUVHyer0B271er1wuV9B9XC7XJevP/9Pr9QZccfJ6vUpNTQ16TIfDIYfDYaV1AAjAHAHQX8wRILxYujlEdHS05syZo4aGBv82n8+nhoaGgI/ufV5GRkZAvSTV19f766+55hq5XK6Ams7OTr3yyisXPSYAAAAADCXLH9UrLi5WQUGB5s6dq7S0NFVUVKirq0uFhYWSpCVLlmjixIkqLy+XJK1YsUJZWVnauHGjcnJyVF1drcOHD2vr1q2SpIiICK1cuVKPPPKIpk6dqmuuuUZr165VQkKCcnNzB26lAAAAANBHloNTXl6eTp8+rXXr1snj8Sg1NVV1dXWKj4+XJLW1tSky8rMLWZmZmdqxY4fWrFmj1atXa+rUqaqpqdHMmTP9NQ8++KC6urp0991364MPPtD8+fNVV1cnp9M5AEsEAAAAgP6xdFe9UGWXu+8A4cCu70e79g2MRHZ9P9q1b2CkGaz3ouUvwAUAAACAcGPL25F/0fmLZnzxHDD8zr8P7XYxmzkChA7mCID+GKwZMiKC09mzZyWJ704AQsjZs2cVGxs73G1cNuYIEHqYIwD6Y6BnyIj4HSefz6dTp05p3LhxioiI6Nexzn953cmTJ0fM55NZkz2MlDUZhqGzZ88qISEh4EYxoY45cmmsKfSNpPWE+xwZSefyPNZkDyNlTYM1Q0bEFafIyEhNmjRpQI8ZExNj6xdMMKzJHkbCmuz0N8TnMUcuD2sKfSNlPcyRkXMuP4812cNIWNNgzBD7/DUOAAAAAAwTghMAAAAAmCA4fYHD4VBZWZkcDsdwtzJgWJM9jMQ1hauReC5ZU+gbaesJZyPxXLImexiJaxpII+LmEAAAAAAwmLjiBAAAAAAmCE4AAAAAYILgBAAAAAAmCE4AAAAAYILgBAAAAAAmwjI4VVZWKikpSU6nU+np6Tp06NAl63ft2qXk5GQ5nU6lpKSotrZ2iDo1V15ernnz5mncuHEaP368cnNzdfz48Uvus337dkVERAQ8nE7nEHVs7kc/+tEF/SUnJ19yn1A+R5KUlJR0wZoiIiK0fPnyoPWhfo7AHAn11yhzJPTPUbhjhoT265MZEvrnaDiEXXDauXOniouLVVZWppaWFs2aNUtut1vt7e1B6xsbG5Wfn6+lS5eqtbVVubm5ys3N1dGjR4e48+AOHDig5cuX6+DBg6qvr9e5c+f09a9/XV1dXZfcLyYmRn/961/9j/fee2+IOr48N9xwQ0B/L7300kVrQ/0cSdKf/vSngPXU19dLkhYuXHjRfUL9HIUz5sinQv01yhwJ/XMUrpghnwr11yczJPTP0ZAzwkxaWpqxfPly/597e3uNhIQEo7y8PGj9okWLjJycnIBt6enpxg9+8INB7bOv2tvbDUnGgQMHLlrz5JNPGrGxsUPXlEVlZWXGrFmzLrvebufIMAxjxYoVxnXXXWf4fL6gz4f6OQp3zJHQf40yR0L/HIUzZkjovz6ZIaF/joZDWF1x6unpUXNzs7Kzs/3bIiMjlZ2draampqD7NDU1BdRLktvtvmj9cOvo6JAkXXXVVZes+/DDDzVlyhQlJibq9ttv15///OehaO+yvfnmm0pISNC1116ru+66S21tbRettds56unp0VNPPaXvf//7ioiIuGhdqJ+jcMUc+Uyov0aZI6F/jsIRM+Qzof76ZIaE/jkaamEVnM6cOaPe3l7Fx8cHbI+Pj5fH4wm6j8fjsVQ/nHw+n1auXKmbb75ZM2fOvGjdtGnTtG3bNv33f/+3nnrqKfl8PmVmZuovf/nLEHZ7cenp6dq+fbvq6ur0xBNP6MSJE/rKV76is2fPBq230zmSpJqaGn3wwQf63ve+d9GaUD9H4Yw58qlQf40yR0L/HIUrZsinQv31yQwJ/XM0HEYNdwMYOMuXL9fRo0cv+RlcScrIyFBGRob/z5mZmZo+fbr+7d/+TevXrx/sNk194xvf8P/vG2+8Uenp6ZoyZYp++9vfaunSpcPY2cD41a9+pW984xtKSEi4aE2onyOMXMwRe2COIFQxQ+yBGdI3YRWc4uLiFBUVJa/XG7Dd6/XK5XIF3cflclmqHy5FRUXau3evXnzxRU2aNMnSvqNHj9aXv/xlvfXWW4PUXf9ceeWVuv766y/an13OkSS99957eu655/TMM89Y2i/Uz1E4YY4EF+qvUeZI6J+jcMEMCS7UX5/MkNA/R0MhrD6qFx0drTlz5qihocG/zefzqaGhISBRf15GRkZAvSTV19dftH6oGYahoqIi7d69W88//7yuueYay8fo7e3Va6+9pgkTJgxCh/334Ycf6u23375of6F+jj7vySef1Pjx45WTk2Npv1A/R+GEORJcqL9GmSOhf47CBTMkuFB/fTJDQv8cDYlhvjnFkKuurjYcDoexfft24/XXXzfuvvtu48orrzQ8Ho9hGIaxePFiY9WqVf76l19+2Rg1apTx2GOPGceOHTPKysqM0aNHG6+99tpwLSHAP//zPxuxsbHG/v37jb/+9a/+x0cffeSv+eKafvzjHxvPPvus8fbbbxvNzc3GnXfeaTidTuPPf/7zcCzhAg888ICxf/9+48SJE8bLL79sZGdnG3FxcUZ7e7thGPY7R+f19vYakydPNh566KELnrPbOQp3zJHQf40yR0L/HIUzZkjovz6ZIaF/joZD2AUnwzCMxx9/3Jg8ebIRHR1tpKWlGQcPHvQ/l5WVZRQUFATU//a3vzWuv/56Izo62rjhhhuMffv2DXHHFycp6OPJJ5/013xxTStXrvSvPz4+3rj11luNlpaWoW/+IvLy8owJEyYY0dHRxsSJE428vDzjrbfe8j9vt3N03rPPPmtIMo4fP37Bc3Y7R2COhPprlDkS+uco3DFDQvv1yQwJ/XM0HCIMwzCG9BIXAAAAANhMWP2OEwAAAAD0BcEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADAxKjhbmAg+Hw+nTp1SuPGjVNERMRwtwOENcMwdPbsWSUkJCgy0j5/N8McAUIHcwRAfwzWDBkRwenUqVNKTEwc7jYAfM7Jkyc1adKk4W7jsjFHgNDDHAHQHwM9Q0ZEcBo3bpykT//jxMTEDHM3QHjr7OxUYmKi/31pF8wRIHQwRwD0x2DNkBERnM5fDo+JiWFQASHCbh9TYY4AoYc5AqA/BnqG2OeDwwAAAAAwTAhOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJvoUnCorK5WUlCSn06n09HQdOnTokvW7du1ScnKynE6nUlJSVFtbG/D8hx9+qKKiIk2aNEljxozRjBkzVFVV1ZfWAAAAAGDAWQ5OO3fuVHFxscrKytTS0qJZs2bJ7Xarvb09aH1jY6Py8/O1dOlStba2Kjc3V7m5uTp69Ki/pri4WHV1dXrqqad07NgxrVy5UkVFRdqzZ0/fVwYAAAAAAyTCMAzDyg7p6emaN2+etmzZIkny+XxKTEzUvffeq1WrVl1Qn5eXp66uLu3du9e/7aabblJqaqr/qtLMmTOVl5entWvX+mvmzJmjb3zjG3rkkUcuOGZ3d7e6u7v9f+7s7FRiYqI6OjoUExNjZTkABlhnZ6diY2ND/v3IHAFCF3MEQH8M1gyxdMWpp6dHzc3Nys7O/uwAkZHKzs5WU1NT0H2ampoC6iXJ7XYH1GdmZmrPnj16//33ZRiGXnjhBb3xxhv6+te/HvSY5eXlio2N9T8SExOtLAMAmCMA+o05AoQXS8HpzJkz6u3tVXx8fMD2+Ph4eTyeoPt4PB7T+scff1wzZszQpEmTFB0drQULFqiyslK33HJL0GOWlpaqo6PD/zh58qSVZQAAcwRAvzFHgPAyargbkD4NTgcPHtSePXs0ZcoUvfjii1q+fLkSEhIuuFolSQ6HQw6HYxg6BTBSMEcA9BdzBAgvloJTXFycoqKi5PV6A7Z7vV65XK6g+7hcrkvW//3vf9fq1au1e/du5eTkSJJuvPFGHTlyRI899ljQ4AQAAAAAQ8nSR/Wio6M1Z84cNTQ0+Lf5fD41NDQoIyMj6D4ZGRkB9ZJUX1/vrz937pzOnTunyMjAVqKiouTz+ay0BwAAAACDwvJH9YqLi1VQUKC5c+cqLS1NFRUV6urqUmFhoSRpyZIlmjhxosrLyyVJK1asUFZWljZu3KicnBxVV1fr8OHD2rp1qyQpJiZGWVlZKikp0ZgxYzRlyhQdOHBAv/nNb7Rp06YBXCoAAAAA9I3l4JSXl6fTp09r3bp18ng8Sk1NVV1dnf8GEG1tbQFXjzIzM7Vjxw6tWbNGq1ev1tSpU1VTU6OZM2f6a6qrq1VaWqq77rpL//u//6spU6boJz/5ie65554BWCIAAAAA9I/l73EKRXb5vgcgHNj1/WjXvoGRyK7vR7v2DYw0IfE9TgAAAAAQjghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCC4AQAAAAAJghOAAAAAGCiT8GpsrJSSUlJcjqdSk9P16FDhy5Zv2vXLiUnJ8vpdColJUW1tbUX1Bw7dky33XabYmNjdcUVV2jevHlqa2vrS3sAAAAAMKAsB6edO3equLhYZWVlamlp0axZs+R2u9Xe3h60vrGxUfn5+Vq6dKlaW1uVm5ur3NxcHT161F/z9ttva/78+UpOTtb+/fv16quvau3atXI6nX1fGQAAAAAMkAjDMAwrO6Snp2vevHnasmWLJMnn8ykxMVH33nuvVq1adUF9Xl6eurq6tHfvXv+2m266SampqaqqqpIk3XnnnRo9erT+4z/+o0+L6OzsVGxsrDo6OhQTE9OnYwAYGHZ9P9q1b2Aksuv70a59AyPNYL0XLV1x6unpUXNzs7Kzsz87QGSksrOz1dTUFHSfpqamgHpJcrvd/nqfz6d9+/bp+uuvl9vt1vjx45Wenq6ampqL9tHd3a3Ozs6ABwBYwRwB0F/MESC8WApOZ86cUW9vr+Lj4wO2x8fHy+PxBN3H4/Fcsr69vV0ffvihNmzYoAULFugPf/iD7rjjDn3729/WgQMHgh6zvLxcsbGx/kdiYqKVZQAAcwRAvzFHgPAy7HfV8/l8kqTbb79d999/v1JTU7Vq1Sp985vf9H+U74tKS0vV0dHhf5w8eXIoWwYwAjBHAPQXcwQIL6OsFMfFxSkqKkperzdgu9frlcvlCrqPy+W6ZH1cXJxGjRqlGTNmBNRMnz5dL730UtBjOhwOORwOK60DQADmCID+Yo4A4cXSFafo6GjNmTNHDQ0N/m0+n08NDQ3KyMgIuk9GRkZAvSTV19f766OjozVv3jwdP348oOaNN97QlClTrLQHAAAAAIPC0hUnSSouLlZBQYHmzp2rtLQ0VVRUqKurS4WFhZKkJUuWaOLEiSovL5ckrVixQllZWdq4caNycnJUXV2tw4cPa+vWrf5jlpSUKC8vT7fccou+9rWvqa6uTr///e+1f//+gVklAAAAAPSD5eCUl5en06dPa926dfJ4PEpNTVVdXZ3/BhBtbW2KjPzsQlZmZqZ27NihNWvWaPXq1Zo6dapqamo0c+ZMf80dd9yhqqoqlZeX67777tO0adP0u9/9TvPnzx+AJQIAAABA/1j+HqdQxPcmAKHDru9Hu/YNjER2fT/atW9gpAmJ73ECAAAAgHBEcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADBBcAIAAAAAEwQnAAAAADDRp+BUWVmppKQkOZ1Opaen69ChQ5es37Vrl5KTk+V0OpWSkqLa2tqL1t5zzz2KiIhQRUVFX1oDAAAAgAFnOTjt3LlTxcXFKisrU0tLi2bNmiW326329vag9Y2NjcrPz9fSpUvV2tqq3Nxc5ebm6ujRoxfU7t69WwcPHlRCQoL1lQAAAADAILEcnDZt2qRly5apsLBQM2bMUFVVlcaOHatt27YFrd+8ebMWLFigkpISTZ8+XevXr9fs2bO1ZcuWgLr3339f9957r55++mmNHj26b6sBAAAAgEFgKTj19PSoublZ2dnZnx0gMlLZ2dlqamoKuk9TU1NAvSS53e6Aep/Pp8WLF6ukpEQ33HCDaR/d3d3q7OwMeACAFcwRAP3FHAHCi6XgdObMGfX29io+Pj5ge3x8vDweT9B9PB6Paf3PfvYzjRo1Svfdd99l9VFeXq7Y2Fj/IzEx0coyAIA5AqDfmCNAeBn2u+o1Nzdr8+bN2r59uyIiIi5rn9LSUnV0dPgfJ0+eHOQuAYw0zBEA/cUcAcLLKCvFcXFxioqKktfrDdju9XrlcrmC7uNyuS5Z/8c//lHt7e2aPHmy//ne3l498MADqqio0LvvvnvBMR0OhxwOh5XWASAAcwRAfzFHgPBi6YpTdHS05syZo4aGBv82n8+nhoYGZWRkBN0nIyMjoF6S6uvr/fWLFy/Wq6++qiNHjvgfCQkJKikp0bPPPmt1PQAAAAAw4CxdcZKk4uJiFRQUaO7cuUpLS1NFRYW6urpUWFgoSVqyZIkmTpyo8vJySdKKFSuUlZWljRs3KicnR9XV1Tp8+LC2bt0qSbr66qt19dVXB/w7Ro8eLZfLpWnTpvV3fQAAAADQb5aDU15enk6fPq1169bJ4/EoNTVVdXV1/htAtLW1KTLyswtZmZmZ2rFjh9asWaPVq1dr6tSpqqmp0cyZMwduFQAAAAAwiCIMwzCGu4n+6uzsVGxsrDo6OhQTEzPc7QBhza7vR7v2DYxEdn0/2rVvYKQZrPfisN9VDwAAAABCHcEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEwQnAAAAADABMEJAAAAAEz0KThVVlYqKSlJTqdT6enpOnTo0CXrd+3apeTkZDmdTqWkpKi2ttb/3Llz5/TQQw8pJSVFV1xxhRISErRkyRKdOnWqL60BAAAAwICzHJx27typ4uJilZWVqaWlRbNmzZLb7VZ7e3vQ+sbGRuXn52vp0qVqbW1Vbm6ucnNzdfToUUnSRx99pJaWFq1du1YtLS165plndPz4cd122239WxkAAAAADJAIwzAMKzukp6dr3rx52rJliyTJ5/MpMTFR9957r1atWnVBfV5enrq6urR3717/tptuukmpqamqqqoK+u/405/+pLS0NL333nuaPHmyaU+dnZ2KjY1VR0eHYmJirCwHwACz6/vRrn0DI5Fd34927RsYaQbrvWjpilNPT4+am5uVnZ392QEiI5Wdna2mpqag+zQ1NQXUS5Lb7b5ovSR1dHQoIiJCV155ZdDnu7u71dnZGfAAACuYIwD6izkChBdLwenMmTPq7e1VfHx8wPb4+Hh5PJ6g+3g8Hkv1H3/8sR566CHl5+dfNCGWl5crNjbW/0hMTLSyDABgjgDoN+YIEF5C6q56586d06JFi2QYhp544omL1pWWlqqjo8P/OHny5BB2CWAkYI4A6C/mCBBeRlkpjouLU1RUlLxeb8B2r9crl8sVdB+Xy3VZ9edD03vvvafnn3/+kp9HdDgccjgcVloHgADMEQD9xRwBwoulK07R0dGaM2eOGhoa/Nt8Pp8aGhqUkZERdJ+MjIyAekmqr68PqD8fmt58800999xzuvrqq620BQAAAACDytIVJ0kqLi5WQUGB5s6dq7S0NFVUVKirq0uFhYWSpCVLlmjixIkqLy+XJK1YsUJZWVnauHGjcnJyVF1drcOHD2vr1q2SPg1N3/nOd9TS0qK9e/eqt7fX//tPV111laKjowdqrQAAAADQJ5aDU15enk6fPq1169bJ4/EoNTVVdXV1/htAtLW1KTLyswtZmZmZ2rFjh9asWaPVq1dr6tSpqqmp0cyZMyVJ77//vvbs2SNJSk1NDfh3vfDCC/rqV7/ax6UBAAAAwMCw/D1OoYjvTQBCh13fj3btGxiJ7Pp+tGvfwEgTEt/jBAAAAADhiOAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgguAEAAAAACYITgAAAABgok/BqbKyUklJSXI6nUpPT9ehQ4cuWb9r1y4lJyfL6XQqJSVFtbW1Ac8bhqF169ZpwoQJGjNmjLKzs/Xmm2/2pTUAAAAAGHCWg9POnTtVXFyssrIytbS0aNasWXK73Wpvbw9a39jYqPz8fC1dulStra3Kzc1Vbm6ujh496q959NFH9fOf/1xVVVV65ZVXdMUVV8jtduvjjz/u+8oAAAAAYIBEGIZhWNkhPT1d8+bN05YtWyRJPp9PiYmJuvfee7Vq1aoL6vPy8tTV1aW9e/f6t910001KTU1VVVWVDMNQQkKCHnjgAf3Lv/yLJKmjo0Px8fHavn277rzzTtOeOjs7FRsbq46ODsXExFhZDoABZtf3o137BkYiu74f7do3MNIM1nvR0hWnnp4eNTc3Kzs7+7MDREYqOztbTU1NQfdpamoKqJckt9vtrz9x4oQ8Hk9ATWxsrNLT0y96zO7ubnV2dgY8AMAK5giA/mKOAOHFUnA6c+aMent7FR8fH7A9Pj5eHo8n6D4ej+eS9ef/aeWY5eXlio2N9T8SExOtLAMAmCMA+o05AoQXW95Vr7S0VB0dHf7HyZMnh7slADbDHAHQX8wRILyMslIcFxenqKgoeb3egO1er1culyvoPi6X65L15//p9Xo1YcKEgJrU1NSgx3Q4HHI4HFZaB4AAzBEA/cUcAcKLpStO0dHRmjNnjhoaGvzbfD6fGhoalJGREXSfjIyMgHpJqq+v99dfc801crlcATWdnZ165ZVXLnpMAAAAABhKlq44SVJxcbEKCgo0d+5cpaWlqaKiQl1dXSosLJQkLVmyRBMnTlR5ebkkacWKFcrKytLGjRuVk5Oj6upqHT58WFu3bpUkRUREaOXKlXrkkUc0depUXXPNNVq7dq0SEhKUm5s7cCsFAAAAgD6yHJzy8vJ0+vRprVu3Th6PR6mpqaqrq/Pf3KGtrU2RkZ9dyMrMzNSOHTu0Zs0arV69WlOnTlVNTY1mzpzpr3nwwQfV1dWlu+++Wx988IHmz5+vuro6OZ3OAVgiAAAAAPSP5e9xCkV8bwIQOuz6frRr38BIZNf3o137BkaakPgeJwAAAAAIR5Y/qheKzl8044vngOF3/n1ot4vZzBEgdDBHAPTHYM2QERGc/va3v0kSXzwHhJC//e1vio2NHe42LhtzBAg9zBEA/THQM2REBKerrrpK0qc3prDTgO3s7FRiYqJOnjxpq89C27FvO/Ys2bPvjo4OTZ482f++tAvmyNCxY8+SPfu2Y88Sc2Qo2fU1Yse+7dizZM++B2uGjIjgdP4ufrGxsbY5oZ8XExND30PEjj1L9uz783fXtAPmyNCzY8+SPfu2Y88Sc2Qo2fU1Yse+7dizZM++B3qG2GsiAQAAAMAwIDgBAAAAgIkREZwcDofKysrkcDiGuxVL6Hvo2LFnyZ5927Fnib6Hkh17luzZtx17luh7KNmxZ8mefduxZ8mefQ9WzyPiC3ABAAAAYDCNiCtOAAAAADCYCE4AAAAAYILgBAAAAAAmCE4AAAAAYILgBAAAAAAmbBOcKisrlZSUJKfTqfT0dB06dOiS9bt27VJycrKcTqdSUlJUW1s7RJ0GstL3L3/5S33lK1/Rl770JX3pS19Sdna26ToHi9X/3udVV1crIiJCubm5g9tgEFZ7/uCDD7R8+XJNmDBBDodD119//bC8Tqz2XVFRoWnTpmnMmDFKTEzU/fffr48//niIupVefPFFfetb31JCQoIiIiJUU1Njus/+/fs1e/ZsORwO/cM//IO2b98+6H0GwxwZOnacIZI954jdZohk3znCDBladpwjdpwhkv3myLDNEMMGqqurjejoaGPbtm3Gn//8Z2PZsmXGlVdeaXi93qD1L7/8shEVFWU8+uijxuuvv26sWbPGGD16tPHaa6+FdN/f/e53jcrKSqO1tdU4duyY8b3vfc+IjY01/vKXv4R03+edOHHCmDhxovGVr3zFuP3224em2f/Pas/d3d3G3LlzjVtvvdV46aWXjBMnThj79+83jhw5EtJ9P/3004bD4TCefvpp48SJE8azzz5rTJgwwbj//vuHrOfa2lrj4YcfNp555hlDkrF79+5L1r/zzjvG2LFjjeLiYuP11183Hn/8cSMqKsqoq6sbmob/P+bI0M0RO84Qw7DnHLHjDDEMe84RZgg/i5ix4wzpS9+hMEeGa4bYIjilpaUZy5cv9/+5t7fXSEhIMMrLy4PWL1q0yMjJyQnYlp6ebvzgBz8Y1D6/yGrfX/TJJ58Y48aNM379618PVotB9aXvTz75xMjMzDT+/d//3SgoKBjyYWW15yeeeMK49tprjZ6enqFqMSirfS9fvtz4x3/8x4BtxcXFxs033zyofV7M5QyrBx980LjhhhsCtuXl5Rlut3sQO7sQc2To5ogdZ4hh2HOO2H2GGIZ95ggzhJ9FzNhxhhiG/efIUM6QkP+oXk9Pj5qbm5Wdne3fFhkZqezsbDU1NQXdp6mpKaBektxu90XrB0Nf+v6ijz76SOfOndNVV101WG1eoK99/+u//qvGjx+vpUuXDkWbAfrS8549e5SRkaHly5crPj5eM2fO1E9/+lP19vYOVdt96jszM1PNzc3+S+jvvPOOamtrdeuttw5Jz31h1/ejXfv+oqGeI3acIZI950i4zBBp+N+PzBB+FjFjxxkihc8cGaj346iBbGownDlzRr29vYqPjw/YHh8fr//5n/8Juo/H4wla7/F4Bq3PL+pL31/00EMPKSEh4YITPZj60vdLL72kX/3qVzpy5MgQdHihvvT8zjvv6Pnnn9ddd92l2tpavfXWW/rhD3+oc+fOqaysbCja7lPf3/3ud3XmzBnNnz9fhmHok08+0T333KPVq1cPRct9crH3Y2dnp/7+979rzJgxg94Dc2To5ogdZ4hkzzkSLjNEGv45wgzhZxEzdpwhUvjMkYGaISF/xSlcbdiwQdXV1dq9e7ecTudwt3NRZ8+e1eLFi/XLX/5ScXFxw93OZfP5fBo/fry2bt2qOXPmKC8vTw8//LCqqqqGu7VL2r9/v37605/qF7/4hVpaWvTMM89o3759Wr9+/XC3hhBkhzli1xki2XOOMENghR1miGTfOWLHGSKF9xwJ+StOcXFxioqKktfrDdju9XrlcrmC7uNyuSzVD4a+9H3eY489pg0bNui5557TjTfeOJhtXsBq32+//bbeffddfetb3/Jv8/l8kqRRo0bp+PHjuu6660KqZ0maMGGCRo8eraioKP+26dOny+PxqKenR9HR0YPas9S3vteuXavFixfrn/7pnyRJKSkp6urq0t13362HH35YkZGh93chF3s/xsTEDMnVJok5MpRzxI4zRLLnHAmXGSIN/xxhhvCzyED3LA3/DJHCZ44M1AwJvZV9QXR0tObMmaOGhgb/Np/Pp4aGBmVkZATdJyMjI6Bekurr6y9aPxj60rckPfroo1q/fr3q6uo0d+7coWg1gNW+k5OT9dprr+nIkSP+x2233aavfe1rOnLkiBITE0OuZ0m6+eab9dZbb/kHqyS98cYbmjBhwpAMKqlvfX/00UcXDKTzA/fT348MPXZ9P9q1b2l454gdZ0hf+paGf46EywyRhv/9yAwZWnacI3acIVL4zJEBez9aupXEMKmurjYcDoexfft24/XXXzfuvvtu48orrzQ8Ho9hGIaxePFiY9WqVf76l19+2Rg1apTx2GOPGceOHTPKysqG7RagVvresGGDER0dbfzXf/2X8de//tX/OHv2bEj3/UXDcScbqz23tbUZ48aNM4qKiozjx48be/fuNcaPH2888sgjId13WVmZMW7cOOM///M/jXfeecf4wx/+YFx33XXGokWLhqzns2fPGq2trUZra6shydi0aZPR2tpqvPfee4ZhGMaqVauMxYsX++vP3wK0pKTEOHbsmFFZWTlstyNnjoRmz180XHfVs+McseMMMQx7zhFmCD+LmLHjDOlL36EwR4ZrhtgiOBmGYTz++OPG5MmTjejoaCMtLc04ePCg/7msrCyjoKAgoP63v/2tcf311xvR0dHGDTfcYOzbt2+IO/6Ulb6nTJliSLrgUVZWFtJ9f9Fw/dBjtefGxkYjPT3dcDgcxrXXXmv85Cc/MT755JMh7tpa3+fOnTN+9KMfGdddd53hdDqNxMRE44c//KHxf//3f0PW7wsvvBD0dXq+z4KCAiMrK+uCfVJTU43o6Gjj2muvNZ588skh6/fzmCOh2fMXDdcMMQx7zhG7zRDDsO8cYYaEbt9fxM8i1thtjgzXDIkwjBC9pgYAAAAAISLkf8cJAAAAAIYbwQkAAAAATBCcAAAAAMAEwQkAAAAATBCcAAAAAMAEwQkAAAAATBCcAAAAAMAEwQkAAAAATBCcAAAAAMAEwQkAAAAATBCcAAAAAMDE/wPTSQhhStxQ5AAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 1000x500 with 6 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axes = plt.subplots(2,3, figsize=(10,5), sharey=True)\n",
    "\n",
    "axes[0,0].bar(range(len(conv1_m)), conv1_m)\n",
    "axes[0,1].bar(range(len(maxpool_m)), maxpool_m)\n",
    "axes[0,2].bar(range(len(layer1_0_conv1_m)), layer1_0_conv1_m)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "alex",
   "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": "3.8.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}