analysis.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import torch\n",
    "import os\n",
    "import math\n",
    "import pickle\n",
    "\n",
    "plt.rcParams[\"font.family\"] = \"Liberation Sans\"\n",
    "plt.rcParams[\"font.size\"] = 12\n",
    "plt.rcParams['xtick.labelsize'] = 10\n",
    "plt.rcParams['ytick.labelsize'] = 10\n",
    "plt.rcParams['lines.linewidth'] = 0.5\n",
    "plt.rcParams['figure.autolayout'] = True\n",
    "plt.rcParams['axes.spines.right'] = False\n",
    "plt.rcParams['axes.spines.top'] = False\n",
    "plt.rcParams['axes.xmargin'] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(13910, 128)\n"
     ]
    }
   ],
   "source": [
    "from preprocess import load_features_labels\n",
    "features, labels, batch_ind = load_features_labels()\n",
    "\n",
    "print(features.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "GAS_IDENTITIES = {\n",
    "    1: \"Acetone\",\n",
    "    2: \"Acetaldehyde\",\n",
    "    3: \"Ethanol\",\n",
    "    4: \"Ethylene\",\n",
    "    5: \"Ammonia\",\n",
    "    6: \"Toluene\"\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Basic statistics\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Number of samples per label per batch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Batch 1: 445 samples\n",
      "    gas 1: 90 samples\n",
      "    gas 2: 98 samples\n",
      "    gas 3: 83 samples\n",
      "    gas 4: 30 samples\n",
      "    gas 5: 70 samples\n",
      "    gas 6: 74 samples\n",
      "Batch 2: 1244 samples\n",
      "    gas 1: 164 samples\n",
      "    gas 2: 334 samples\n",
      "    gas 3: 100 samples\n",
      "    gas 4: 109 samples\n",
      "    gas 5: 532 samples\n",
      "    gas 6: 5 samples\n",
      "Batch 3: 1586 samples\n",
      "    gas 1: 365 samples\n",
      "    gas 2: 490 samples\n",
      "    gas 3: 216 samples\n",
      "    gas 4: 240 samples\n",
      "    gas 5: 275 samples\n",
      "    gas 6: 0 samples\n",
      "Batch 4: 161 samples\n",
      "    gas 1: 64 samples\n",
      "    gas 2: 43 samples\n",
      "    gas 3: 12 samples\n",
      "    gas 4: 30 samples\n",
      "    gas 5: 12 samples\n",
      "    gas 6: 0 samples\n",
      "Batch 5: 197 samples\n",
      "    gas 1: 28 samples\n",
      "    gas 2: 40 samples\n",
      "    gas 3: 20 samples\n",
      "    gas 4: 46 samples\n",
      "    gas 5: 63 samples\n",
      "    gas 6: 0 samples\n",
      "Batch 6: 2300 samples\n",
      "    gas 1: 514 samples\n",
      "    gas 2: 574 samples\n",
      "    gas 3: 110 samples\n",
      "    gas 4: 29 samples\n",
      "    gas 5: 606 samples\n",
      "    gas 6: 467 samples\n",
      "Batch 7: 3613 samples\n",
      "    gas 1: 649 samples\n",
      "    gas 2: 662 samples\n",
      "    gas 3: 360 samples\n",
      "    gas 4: 744 samples\n",
      "    gas 5: 630 samples\n",
      "    gas 6: 568 samples\n",
      "Batch 8: 294 samples\n",
      "    gas 1: 30 samples\n",
      "    gas 2: 30 samples\n",
      "    gas 3: 40 samples\n",
      "    gas 4: 33 samples\n",
      "    gas 5: 143 samples\n",
      "    gas 6: 18 samples\n",
      "Batch 9: 470 samples\n",
      "    gas 1: 61 samples\n",
      "    gas 2: 55 samples\n",
      "    gas 3: 100 samples\n",
      "    gas 4: 75 samples\n",
      "    gas 5: 78 samples\n",
      "    gas 6: 101 samples\n",
      "Batch 10: 3600 samples\n",
      "    gas 1: 600 samples\n",
      "    gas 2: 600 samples\n",
      "    gas 3: 600 samples\n",
      "    gas 4: 600 samples\n",
      "    gas 5: 600 samples\n",
      "    gas 6: 600 samples\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "n_labels = np.unique(labels).shape[0]\n",
    "df = pd.DataFrame()\n",
    "for batch_num, (start, end) in enumerate(batch_ind):\n",
    "    print(f\"Batch {batch_num+1}: {end-start} samples\")\n",
    "    for label in range(n_labels):\n",
    "        n_matching = (labels[start:end]==label).astype(np.long).sum()\n",
    "        df = df.append({\n",
    "            # add one to be consistent with original data indexing\n",
    "            \"label\": label+1, \n",
    "            \"batch\": batch_num+1,\n",
    "            \"count\": n_matching\n",
    "        }, ignore_index=True)\n",
    "        print(f\"    gas {label+1}: {n_matching} samples\")\n",
    "\n",
    "df.pivot(\"batch\", \"label\", \"count\").plot(kind='bar', figsize=(12, 6))\n",
    "plt.ylabel(\"Count\")\n",
    "plt.title(\"Amount of data in each batch\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Readings from a single odor class over time"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "A basic plot reveals temporal dynamics which do not depend on the "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 504x180 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "class_1_indices, = np.where(labels==0)\n",
    "readings = features[class_1_indices]\n",
    "\n",
    "x = np.arange(readings.shape[0])\n",
    "feature_ids = [0]\n",
    "n_plots = len(feature_ids)\n",
    "\n",
    "fig, axes = plt.subplots(n_plots, 1, figsize=(7, 2.5), sharex=True)\n",
    "if axes is not list:\n",
    "    axes = [axes]\n",
    "for plot_id, feat_id in enumerate(feature_ids):\n",
    "    y = readings[:, feat_id]\n",
    "    \n",
    "    ax = axes[plot_id]\n",
    "    ax.plot(x, y, c=\"black\", linewidth=1)\n",
    "    ax.set_ylabel(f\"Feature {feat_id+1} Z-score\")\n",
    "    \n",
    "    for start, _ in batch_ind:\n",
    "        # Draw a line at the first sample whose index is equal to order greater than the batch start\n",
    "        if start == 0:\n",
    "            continue\n",
    "        for sub_i, i in enumerate(class_1_indices):\n",
    "            if i >= start:\n",
    "                break\n",
    "        ax.axvline(sub_i, linestyle='--', linewidth=1)\n",
    "        \n",
    "ax.set_xlabel(\"Samples of Acetone\")\n",
    "# plt.tight_layout()\n",
    "fig.savefig(\"writeup/figure_sources/fig_1.svg\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "numpy.ndarray"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fdf7a1ab400>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from preprocess import _load_raw_data\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "features, labels, batch_ind = _load_raw_data(include_gas_6=False)\n",
    "\n",
    "anamolous_sensor = 1\n",
    "\n",
    "readings = features[labels==0]\n",
    "x = np.arange(readings.shape[0])\n",
    "y = readings[:, anamolous_sensor]\n",
    "\n",
    "plt.figure(figsize=(12, 4))\n",
    "plt.plot(x, y)\n",
    "plt.ylabel(\"Raw Reading\")\n",
    "plt.xlabel(\"Samples of Odor 1\")\n",
    "plt.title(\"Verifying that an outlier in the z-scores is present in the raw data\")\n",
    "plt.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Identifying outliers\n",
    "Print the index of any feature with a standard deviation of more than 5, and also print that feature value."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([ 663,  666,  666,  666,  666,  666,  666,  666,  812,  824,  824,\n",
       "         824,  825,  825,  825,  825, 9871]),\n",
       " array([ 20,   1,  33,  36,  41,  44,  49,  57,  43,  20,  27,  28,  36,\n",
       "         44, 100, 108,   9]))"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(features > 50).nonzero()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Visualizing features and labels for figures"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 285,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x shape (10, 6, 10) y shape (10, 6)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fecdc05b7b8>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Visualize features and labels from each batch\n",
    "\n",
    "import numpy as np\n",
    "import math\n",
    "import cairo\n",
    "from mpl_toolkits.axes_grid1 import make_axes_locatable\n",
    "from batches import features, labels, batch_ind\n",
    "\n",
    "n_samples = 6\n",
    "n_feats = 10\n",
    "batches = list(range(10))\n",
    "\n",
    "x = []\n",
    "y = []\n",
    "for batch in batches:\n",
    "    samp = np.random.choice(np.arange(*batch_ind[batch]), n_samples, replace=False)\n",
    "    x.append(features[samp, :n_feats])\n",
    "    y.append(labels[samp])\n",
    "x = np.stack(x)\n",
    "y = np.stack(y)\n",
    "print(\"x shape\", x.shape, \"y shape\", y.shape)\n",
    "\n",
    "fig, axes = plt.subplots(len(batches)+1, 2, figsize=(4, 16))\n",
    "for row in range(len(batches)+1):\n",
    "    for col in range(2):\n",
    "        ax = axes[row, col]\n",
    "        if row == len(batches):\n",
    "            if col == 0:\n",
    "                plt.colorbar(im_gray, cax=ax, orientation=\"horizontal\")\n",
    "            else:\n",
    "                plt.colorbar(im_color, cax=ax, orientation=\"horizontal\")\n",
    "            continue\n",
    "        if col == 0:\n",
    "            im_gray = ax.imshow(x[row], cmap=\"gray\")\n",
    "        else:\n",
    "            im_color = ax.imshow(y[row].reshape(n_samples, 1), cmap=\"rainbow\")\n",
    "    #     divider = make_axes_locatable(ax)\n",
    "    #     cax = divider.append_axes('top', size=0.5, pad=0.35)\n",
    "        \n",
    "        ax.set_xticks([])\n",
    "        ax.set_yticks([])\n",
    "# plt.colorbar(im_gray)\n",
    "fig.savefig(\"writeup/fig_data_matrix_raw.svg\", format=\"svg\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 290,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fecdb1f4240>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axes = plt.subplots(len(batches)+1, 2, figsize=(4, 4))\n",
    "for row in range(len(batches)+1):\n",
    "    for col in range(2):\n",
    "        ax = axes[row, col]\n",
    "        if row == len(batches):\n",
    "            if col == 0:\n",
    "                plt.colorbar(im_gray, cax=ax, orientation=\"horizontal\")\n",
    "            else:\n",
    "                plt.colorbar(im_color, cax=ax, orientation=\"horizontal\")\n",
    "            continue\n",
    "        if col == 0:\n",
    "            im_gray = ax.imshow(x[row], cmap=\"gray\")\n",
    "        else:\n",
    "            im_color = ax.imshow(y[row].reshape(n_samples, 1), cmap=\"rainbow\")\n",
    "            im_color.set_clim(0.0, 4.0)\n",
    "    #     divider = make_axes_locatable(ax)\n",
    "    #     cax = divider.append_axes('top', size=0.5, pad=0.35)\n",
    "        \n",
    "        ax.set_xticks([])\n",
    "        ax.set_yticks([])\n",
    "# plt.colorbar(im_gray)\n",
    "fig.savefig(\"writeup/fig_data_matrix_raw.svg\", format=\"svg\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x shape (10, 5, 3, 10) y shape (10, 5, 3)\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAdMAAANzCAYAAADyffJqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4xLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvAOZPmwAAIABJREFUeJzt3WmQ5WV5N/7rTK/TM92z0TALwwzIIqICxQhCKUQx4pqISqrUCKKoMaXG3YiIj4pAFpeoQcoycSNlBQzGUksTjYTCqOigQwgaVmeAYYDZ956Z7jnPC8rUU/L/x77uyzmK9fm8vr91/c459/l9+/SL+9fpdrsBALSb8Zu+AAB4tFOmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFDUn1k8c+bM7tjYWGrA/v37U+srWo5GnJqaSmdGR0fTmb6+vp5kIiI2b96czsyYkfu7avv27TExMdFJD0o66KCDusuWLUtl1q9f3zInnYmI6HTyb8GOHTvSmdtvvz2dmTVrVjrT+j60zNqyZUs6c//992/odrvj6WDS3Llzu4sWLUpltm7dmp6TvZ/+Qsu+27VrVzrT8l0aHh5OZ+bMmZPORLTdI3fu3Jlav3Xr1ti1a9evfMNTZTo2NhYve9nLUhfS8gG2FvCePXvSme3bt6czZ5xxRjrTslnmz5+fzkREXHPNNelM9mb4T//0T+kZLZYtWxbf//73U5lPfepT6TmveMUr0pmIthvH9ddfn84885nPTGeOP/74dOb8889PZyIinvzkJ6cz//zP/5zOvOc971mTDjVYtGhRfP7zn09lvvrVr6bnPOMZz0hnIiIGBgbSmZtvvjmdufLKK9OZY489Np15znOek85ERMyePTudufHGG1PrP/e5z01rnX/zAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAotRxgtu3b4/rrrsuNWD58uWp9RERq1evTmciIiYnJ9OZ/v7UWxAREe9+97vTmZZzgzdu3JjORLQdV3faaael1recadziv//7v+PUU09NZQ477LD0nM9+9rPpTETb+9BynugPfvCDdKbFpk2bmnLPe97z0pmWIwh7Ze3atXHhhRemMnPnzk3Paf1cW+5be/fuTWf+6q/+Kp1pOepw9+7d6UxExMUXX5zOHHPMMan1033f/DIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABSlTkvu6+tLH+bccujx+Ph4OhPRdtB9y0Hld911VzozY0b+75bh4eF0JiJiwYIF6czQ0FBqfcvradHX1xfz5s1LZVr23KJFi9KZiLb903Lg+B133JHOtHxGs2fPTmci2r6z2T3XS1NTU7Ft27ZUZnR0ND2n9cECvXrAQsseatnfrbZs2ZLO7N+//wBciV+mAFCmTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARamnxhxzzDFx3XXXpQZ87GMfS62PiPiTP/mTdCYiYvfu3enMDTfckM48//nPT2danHPOOU25H/3oR+nMZZddllrfqyd+HH300fHtb387lWl5wspRRx2VzvRSp9PpyZyzzz67Kdey5770pS+lM5///OfTmRbHHnts3HjjjalMy9NSBgcH05le6tW+W7hwYVNu3bp16cy//uu/ptavXLlyWuv8MgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFKUOum/RcoB46+HPLbnnPe95TbN64ZprrmnKXX311enMnDlzUuv7+vrSM3rlvvvuS2d+2w+675Uvf/nLPZu1YcOGns3qhSuvvDKdeeMb33gAruTRZ/369T2b9b3vfS+1fseOHdNa55cpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaCo0+12p7+401kfEWsO3OXwKLKs2+2OH+gh9hy/xL6j16a151JlCgA8kn/zAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFPVnFh900EHd5cuXH6BLefS46aabejJn0aJFTbnFixenM3v37k2tv++++2LTpk2d9KCksbGx7vj4eCqzb9++9JwZM9r+rpyamkpnsu91RNtrGhgYSGdmzZqVzkS0vaZut5vO3H///Ru63W5uQzRwr3tYr+51CxcubMotWbIkndmzZ09q/dq1a6d1r0uV6fLly2PlypWpC/ld1Okc8A6JiIjXvva1Tbn3vve96cy9996bWv/c5z43PaPF+Ph4/OVf/mUqc99996XnzJ49O52JiNi+fXs6c88996QzLa/p0EMPTWdOPvnkdCYi4uc//3k601Km7373u9ekQw3c6x7Wq3vd+eef35S79NJL05m77rortf4FL3jBtNb5Ny8AFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQlDqb94477oizzjorNWDBggWp9RERJ510UjoTEbF169Z05qGHHkpnrrvuunSmr68vndm5c2c6ExHx5Cc/OZ153OMel1p///33p2e0eOihh+KjH/1oKvPYxz42PWfVqlXpTETE5ORkOtNyqP7f/u3fpjMt56q2fIciIi6//PJ05vTTT2+a1Qu33357PP3pT09lWh4w8frXvz6diYg4+OCD05kf/vCH6cz111+fzrTs75YzriMiTjzxxHTm+OOPT61ft27dtNb5ZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKEoddL9v3770wfDz5s1LrY+I+MxnPpPORERMTU2lM/39qbcgIiLe+MY3pjMtB6JPTEykMxHTP5j5/9VyOHwvdDqdGBoaSmW63W56zvz589OZiLY915LZtWtXOtOiZZ9GtD3IoeVz6pVutxv79u1LZQYGBtJz3va2t6UzEe2fU9ZHPvKRdKblc215KENEpD+jiIiRkZHU+uke3O+XKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQFHqkSlz5syJZz3rWakBS5YsSa2PiDjiiCPSmYiIwcHBdObBBx9MZ772ta+lM6Ojo+nM+Ph4OhMRcfbZZ6czixcvTq3/9re/nZ7RYtasWbFixYpU5tBDD03POfjgg9OZiLY9l33yUkTETTfdlM7Mnj07nWl9H57ylKekMy33hl5ZunRpfOxjH0tl7rrrrvSc008/PZ2JyD/5JCLie9/7Xjpz7rnnpjOHH354OvPMZz4znYmI+OQnP5nO3HLLLan1X//616e1zi9TAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkBR6qD7JUuWxGWXXZYacPHFF6fWR0Rceuml6UwvdTqd3/Ql/K+63W468453vCO1fnJyMj2jxZIlS+Lyyy9PZf7P//k/6Tn2XE3LnvvzP//zA3Alvx4jIyNx4oknpjL/8A//kJ7z4he/OJ1p1XKY/J133tmTzLe+9a10JqJt31177bWp9RMTE9Na55cpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaCokzkouNPprI+INQfucngUWdbtdscP9BB7jl9i39Fr09pzqTIFAB7Jv3kBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACK+jOLDzrooO7y5csP0KX8ZnS73XTmxz/+8QG4kkdauHBhU258fDydGRwcTK1fvXp1bNiwoZMelDQ8PNydPXt2KjN37tz0nG3btqUzERGdTv4tmJycTGdGR0fTmRkz8n8rt2QiIjZs2JDOzJs3L51ZvXr1hm63m9/gSSMjI93sPmr53m3cuDGdabVv376ezOnvT9VKREQMDw83zdq+fXs6s2DBgtT6devWxZYtW37lFz31qpcvXx4rV65MXchvuz179qQzrR981vnnn9+Ue93rXpfOLF26NLV+xYoV6RktZs+eHc9//vNTmbPPPjs955vf/GY6ExExNDSUzrQUzxlnnJHOtOzTltKOiPj7v//7dObFL35xOnPuueeuSYcazJ07N1796lenMq997WvTc77whS+kMxFtf/SsXbu2J3Pmz5+fzhx11FHpTETE9ddfn868/OUvT61/5StfOa11/s0LAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKUscJ3nrrrfGEJzwhNSC7PiLiJz/5SToTEbF///50ZmBgIJ255ZZb0pmWa3vwwQfTmYiI0047LZ35/d///dT61atXp2e02LVrV6xatSqVmTVrVnrODTfckM5EtH2uLWeXvuUtb0lnWs6dbjnqMOLhe0PWMccc0zSrFzZv3hzXXnttKnP33Xen57Qez9ry2bbc6774xS+mM1NTU+nMQw89lM5ERHzta19LZyYmJlLr169fP611fpkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIpSJ2739fWlDxFfsGBBan1ExOzZs9OZiLZDxzudTjqze/fudKblYOqWA6Mj2t6/uXPnptb39fWlZ7To6+tLv56Wa5szZ046E9H2GbXshZY916LlOxTRtucmJyebZvXC4OBgLF68OJVZtGhRes7SpUvTmYi2z6nl/V63bl060/L927dvXzoTEXHYYYelMwcffHBq/XQfEOCXKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQFHqqTEzZ86ME044ITXg6KOPTq2PaH9aSsvTCjZu3JjO/OQnP0lnZs6cmc6Mjo6mMxERT3rSk9KZ7BMyBgcH0zNajI6Oxu/93u+lMmeeeWZ6TuvradmrLXvuO9/5TjozNjaWzrQ8+SQi4ulPf3o6c+qpp6YzH/nIR9KZFocccki8+c1vTmVanuRyzjnnpDMRbfeGlvvWZZddls60PMmlZS9ERLz1rW9NZ7JPYJruvdsvUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAUeqg+2XLlsWVV16ZGvCOd7wjtT4i4oorrkhneqnT6fymL+F/1e1205m3vOUtqfV79+5Nz2ixePHi+MAHPpDKZF9LRMSHP/zhdKaXfhf33Jve9KYDcCW/HmNjY/GsZz0rlWl5PR/96EfTmVaPfexj05mXvvSlB+BKHulzn/tcU65l373xjW9Mrd+5c+e01vllCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAo6mQOCu50OusjYs2BuxweRZZ1u93xAz3EnuOX2Hf02rT2XKpMAYBH8m9eAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAov7M4gULFnQPO+yw1IC+vr7U+l6bnJxMZ26++eZ0ptPppDMLFy5MZyIi5s6dm86MjIyk1q9evTo2bNiQf1FJ/f393YGBgVRm1qxZ6Tn79u1LZyIi5s2bl87s378/nZmYmEhn1q9fn860yn5GERFz5sxJZzZs2LCh2+2Op4NJLftu5syZ6Tl79uxJZyIihoeHm3JZY2Nj6czq1at//Rfy/6O/P1VhEZHfdzt27IiJiYlfea9LXclhhx0W//7v/566kJYvTC9t3LgxnTnooIPSmZabzQUXXJDORES88IUvTGdOOOGE1PoVK1akZ7QYGBiII488MpU56aST0nMeeOCBdCYi4kUvelE601KMP/3pT9OZK6+8Mp1pNT6e77fnPOc56cynP/3pNelQg4GBgVi+fHkqc/zxx6fn3HXXXelMRMQxxxyTzrT8EXfWWWelM694xSvSmVYtf8w++9nPTq3/+te/Pq11/s0LAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFKXO5r3tttvijDPOSA049dRTU+sjIm644YZ0JiJiamoqnWk5KPk///M/05mWa3vooYfSmYiIF7zgBenM2WefnVp/7733pme06Ha7sXv37lRm06ZN6Tl33HFHOhMRcfnll6czg4OD6cyXv/zldObVr351OnPnnXemMxERr3nNa9KZrVu3Ns3qhU6nk/6cWs6J3bFjRzoTEfGjH/0onWnZdxdeeGE6s2rVqnSm9V73qle9Kp1ZsGBBav10H9bilykAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoCh1yntfX1+Mjo4eqGv5H3PmzGnK7d+//9d8Jf/fdu3alc50u910pvX1tLx/LdfXC319femDqQ855JD0nHXr1qUzEW2f0YwZ+b9hW/Zcy7W1fvda3vOFCxc2zeqFGTNmxMjISCrTcnB/dm//Qq++r9u2bevJnMnJyabcrFmz0pnNmzen1k/3ISV+mQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABR1Mk8fOPzww7vvfe97UwOOOeaY7DXFxo0b05mIiMHBwXTmtttuS2euu+66dKblqRonnnhiOhPR9lSS7BM83vzmN8cdd9zRSQ9KWrx4cfdVr3pVKvOBD3wgPefCCy9MZyIiBgYG0pkHH3wwnZk/f346MzY2ls4cccQR6UxExKpVq9KZiy66KJ2ZNWvWTd1ud0U6mHT44Yd33/e+96Uyy5cvT8+ZmJhIZyIihoaG0pmf/exn6cw3v/nNdKblXnfSSSelMxER/f2pB59FRP679La3vS3uvPPOX3mv88sUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCUOuh+xYoV3ZUrV6YGZA+LjojIHqbfa53OAT/fvSTzmf5C9qD3z3zmM7Fu3boD/ka07LmLL744Pef9739/OtNLv4t77l3velc6c/nll/fkoHv3uof9Lu67A3Wv88sUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCUOui+0+msj4g1B+5yeBRZ1u12xw/0EHuOX2Lf0WvT2nOpMgUAHsm/eQGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARf2ZxaOjo93x8fHUgNmzZ6fWR0Ts3bs3nYmImDEj/7fBrl270pmNGzemMwMDA+nM6OhoOhMRsX///nRm1qxZqfUPPPBAbNmypZMelDR//vzuoYcemsoMDg4eoKv59WjZ37fccks609fXl87MmzcvnYmImD9/fjozMjKSzqxatWpDt9vN3YQatNzrWr6ve/bsSWciIjqd/FdvYmIinenVvS57//mFbrebzmQ7ad26ddO616XKdHx8PC699NLUhTzlKU9JrY+IuPvuu9OZiLYvZ8tN6vOf/3w6s3DhwnTm9NNPT2ci2r40T3rSk1LrL7jggvSMFoceemh8/etfT2WWLl16gK7m12P16tXpzOGHH57OjI2NpTMveMEL0pmIiJe+9KXpzAknnJDOzJ8/f0061GB8fDw++MEPpjJPf/rT03Nuu+22dCYiYubMmenMrbfems706l53yimnpDMREVNTU+nMqaeemlr/yle+clrr/JsXAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKEqdzbtu3bp43/velxpw/PHHp9ZHRNx0003pTETbocdDQ0PpzLXXXpvO7Nu3L51Zt25dOhMR8cd//MfpzLOf/ezU+g0bNqRntLjzzjvjD/7gD1KZ7NmbERE33HBDOhPR9lCBloPAV61alc60nFva+rlO9/zS/9fZZ5/dNKsX1q1bF+9///tTma9+9avpOa33upZ913Kv+9KXvpTOTE5OpjOt97pzzz03nXnuc5+bWj/d74RfpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAotRB9/39/bFw4cLUgEWLFqXWR0QsX748nYmI6HQ66UzLoczr169PZ1q0XFtExNKlS9OZ8fHx1Pr+/tTWaTZv3rx40YtelMoMDg6m5zzucY9LZ1pn3XvvvenMV77ylXRm9uzZ6cxBBx2UzkREvPzlL09nWg7875Xh4eE47rjjUpmjjz46PWfGjLbfM319fenM9u3b05nbb789nWn5XPfu3ZvORESceOKJ6cyhhx6aWj/d1+OXKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQFHq0R8LFiyIc889NzXgzDPPTK2PiFi1alU6E9H2tII777wznfnQhz6UzhxyyCHpzAknnJDORERccMEFB3zW1772tfSMFoODg+mn4PzhH/5hes53v/vddCaibc/t2bMnnbnuuuvSmfnz56czrU/PyT6JIyLitNNOS2cuvvjidKbF/Pnz45xzzkll/uiP/ig954YbbkhnIh5+qk3WzTffnM5cccUV6UzLk4dOOeWUdCYi4qUvfWk6k+2kL3/5y9Na55cpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaCo0+12p714xYoV3ZUrV6YGXHrppdlrigsvvDCd6aVOp/ObvoT/VeYz/YXsAeKf/vSn4/777z/gb0TLnms5FP5pT3taOtNLv4t77qqrrkpnXv7yl9/U7XZXpINJLfvukksuSc+56KKL0plWU1NT6Ux/f+pZKM0e//jHN+VuueWWdOY973lPav3f/d3fTete55cpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaAoddB9p9NZHxFrDtzl8CiyrNvtjh/oIfYcv8S+o9emtedSZQoAPJJ/8wJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABT1ZxaPjIx058yZkxowODiYWh8RMTU1lc5ERHQ6nXSm2+2mMy3X13JtAwMD6UxExI4dO9KZkZGR1PrNmzfHjh078i8qaWBgoDs8PJzKDA0Npefs378/nYmI6Ovra8plLViwIJ3p1d6OiFi7dm06M3v27HRmw4YNG7rd7ng6mDRz5szu2NhYKjNr1qz0nImJiXQmom3f9eq+1ZLp709V0f/Ytm1bOpP9nDZt2hQ7d+78lS8q9QrmzJkT559/fupCli1bllof8fCNukXLTXTPnj3pTMsH2HJtCxcuTGciIq6//vp05uSTT06t/9CHPpSe0WJ4eDhWrFiRyhx++OHpObt3705nIiKyN9xWL3vZy9KZlpvn1q1b05mIiHe9613pzFOe8pR05tOf/vSadKjB2NhYvOQlL0llsvs0IuKuu+5KZyLa/hDZvn17OtNSci2Zgw8+OJ2JiPiXf/mXdObJT35yav2HP/zhaa3zb14AKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCUOvdp8+bNcc0116QGHH/88an1ERE333xzOhPRdhbp6OhoOnP11VenMy3HFracdxoR8e53vzudyZ71uWvXrvSMFpOTk7Fhw4ZUZsmSJek5q1atSmdatRwF94Y3vCGdmZycTGdaz4ptOaqu5fvaK1u3bo1vfOMbqczq1avTc2677bZ0JiJi7ty56UzLve7jH/94OtPLe913vvOddCZ7dvDOnTuntc4vUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAUeqg+xkzZqQP6Z45c2ZqfUTbQeCthoeH05mWw8D37duXzuzfvz+diYiYN29eOjMyMpJaP2NGb/4OmzFjRvozyh5kHdHbPdffn/raRUTE1NRUTzKtWr5HLZ9Tr/T19aUPhm85SH7WrFnpTETb/aTl4RQth9a3XFurlgP/x8bGUuun+xAQv0wBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKOt1ud9qLly5d2n3rW9+aGnDcccdlrynWrl2bzkS0PY1j48aN6cxPf/rTdKbl6RBLlixJZyIidu/enc4ceeSRqfUXXnhh3H333Qf8sR+LFy/uXnDBBanM0UcfnZ6zevXqdCai7ek5LU/i2Lt3bzrT8sSmRYsWpTMRET/72c/SmWOPPTadec1rXnNTt9tdkQ4mLVy4sHveeeelMi33ujvuuCOdiWh74k7LU2NanoTT8gShliddRUSsWrUqnXniE5+YWn/ZZZfFmjVrfuUb7pcpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaAodTL8IYccEm9605tSAz70oQ+l1kdEZA/Tr7jnnnvSmex7ENF2+PqZZ56ZzkREXHHFFenMJZdcklrfclh7i06nE4ODg6nMt771rfScvr6+dKY117IXWg7Ub7F169am3MDAQDrz/e9/v2lWLxx66KHxF3/xF6lMdn1ExAc+8IF0ptWmTZvSmQULFqQzy5YtS2cDrJ5tAAAUb0lEQVTOOOOMdCYi4nOf+1w6k33PJycnp7XOL1MAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQFGn2+1Of3Gnsz4i1hy4y+FRZFm32x0/0EPsOX6JfUevTWvPpcoUAHgk/+YFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAo6s8sHhgY6A4PD6cGZNdHROzbty+diYiYMSP/t0Gn00lnxsfH05kWk5OTTbm1a9emM7Nnz06t3759e0xMTOTfvKS+vr7uwMBAKjNr1qz0nJ07d6YzrVr26RFHHJHOdLvddKb1u3fvvfemM6Ojo+nM+vXrN3S73QP+Bezv7+8ODg6mMjNnzkzPmZiYSGcienevW7p0aTrzu7bvpnuvS5Xp8PBwnHDCCakLOeaYY1LrIyIeeuihdCaibTP39fWlM3/6p3+azuzfvz+d2bJlSzoTEfHOd74znTn99NNT66+99tr0jBYDAwOxfPnyVGbFihXpOT/60Y/SmVYt+/Tqq69OZ/bu3ZvOtPwhFhHxpje9KZ35vd/7vXTmyiuvXJMONRgcHIyjjjoqlTnuuOPSc+644450JqLtR8rQ0FA685GPfCSdaSnGBx54IJ2JiHjLW96Szpxxxhmp9dO91/k3LwAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCUOpt3xowZ6UOCW84hHRkZSWci2g5/zh6iHvHwwcdZLYdMt5ytGhExZ86cdCb7ubacadxqamoqtb7l82k5O7lVy55r2T8tWg4pj4jYuHFjOrN58+amWb2SvZ9kD8aPaP+OtxyQn32YRUTb96J1D7VoeU3Ze9d0v3t+mQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAilIH3e/ZsyfuuOOO1IC5c+em1kdErFy5Mp2JaDtgueVQ/Ysuuiid2bNnTzpz//33pzMREffdd186s2nTptT6ycnJ9IxWvThwvJda3rteHTjeeqD+vHnz0pmxsbGmWb2wf//+9AMTdu/enZ6zY8eOdCaidw9mmDVrVjrTcnh/f3+qiv7H2rVr05mdO3em1k/3vfbLFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoCh1VH+n00mf7t/LJwi0PCVjaGgonZmamkpnWq6t9ckQo6Oj6UzL9fVCf39/jI+PpzJLly5Nz7nnnnvSmYi2z2h4eDid2bZtWzrTsk/37duXzkREnHHGGenM0Ucf3TSrV/r6+lLrW9671ntdy9N9Wp6m1LK/e5WJaPsuZWdN997olykAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoCh1yvLIyEiceOKJqQELFy5MrY+ImJycTGciIvbs2ZPOLFq0KJ1Zs2ZNOtNyCPaWLVvSmYiI173udenMpk2bUutbDs1u8djHPja++93vpjLvec970nNuvPHGdKaXenWw+cknn5zORER88pOfTGeuuuqqplm9sHjx4vQ++vGPf5yec84556QzEb17cMbHP/7xdKbl8PmBgYF0JiLi5z//eTrz3ve+N7V+ug9D8csUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFDUyRyY3Ol01kdE/pR3fhct63a74wd6iD3HL7Hv6LVp7blUmQIAj+TfvABQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEX9qcX9/d2BgYHUgFmzZqXWR0Ts3LkznWk1Y0b+74nDDz88nel2u+nMvn370pmIiHvvvTedmTNnTmr91q1bY/fu3Z30oCR77mG/7Xvu7rvvTmfmzp2bzmzcuHFDt9sdTweT7LuH/bbvu5Z73bx581Lrt2zZErt27fqV97pUmQ4MDMRjHvOY1IWcfPLJqfURET/84Q/TmVYzZ85MZ77whS+kMy2bZe3atelMRMTb3/72dOass85Krb/qqqvSM1oMDAzEkUcemco86UlPSs+x5x7Wuude8pKXpDPPf/7z05nPfvaza9KhBgMDA7F8+fJU5rf9XjcyMpLOfPGLX0xn9u7dm87cd9996UxE273uRS96UWr9pz71qWmt829eAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQlDpOsNvtxsTERGrAxo0bU+sjInbv3p3ORETs378/nenvT70FzZmWa2s5SzMiYseOHenMtm3bUuunpqbSM1q07LlNmzal52Rn/ELLOaTZM18jerfnOp2245azZztHtL2mXul2uzE5OZnKbN26NT2n9Uzaln03NDSUzrTuh17Nafmur169OrV+uscj+mUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChKH3SfPWC59bD2Xmk5sL3lUOZeHRgdEdHX15fOTPcw519oOWi7RbfbTR/Y3nJtra+n5TD57AHqEb3bc6379BOf+EQ6s3LlyqZZvZLdE2NjYwd8xi+07LuWe13L/bsl07rvDj744HQm+6CJ6V7bb3fTAcCjgDIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABSlnhrT6XTSTwRoeSpC65NmWp48MDQ0lM60PH2hJdPyZIiIiMMOOyyd6dVTYHrhd+m1PFpcccUV6UzLEz96qb8/dXuMbdu2HfAZFbNnz05nsk+TiojYt29fOtPyJKWItte0Z8+e1Prp3of9MgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFKVPWc4eIt5yaH1fX186ExExd+7cdGbmzJnpTMv1tWRaD/x/+9vfns6sXLkytb7lAQEtOp1O+gEGLZ9pL7V+rlm9PPC/5dD6+fPnH4Ar+fXo6+uLOXPmpDLj4+PpOWvWrElnItoegtFyqH6vDq1vfajHG97whnTmrrvuSq3/zne+M611fpkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIo6mcOwO53O+ohoO5mZ3zXLut1u/mTvJHuOX2Lf0WvT2nOpMgUAHsm/eQGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIr6M4sHBwe7w8PDqQEzZ85MrY+ImJycTGciIjqdTlMu66CDDkpnut1uOjM1NZXORESsW7cunRkbG0ut37p1a+zateuAv+H23MPsuYc98MADG7rd7ng6mPTbvu9mzOjN76AFCxakM79r+26697pUmQ4PD8cpp5ySupBjjz02tT4iYtOmTelMRMTg4GA60/LBv/KVr0xnWr40mzdvTmciIj74wQ+mM2eddVZq/Wc+85n0jBa92nMbNmxIZyIevr5eOP/889OZlhtU65675JJL0pnsnouIuOyyy9akQw16te9a3++hoaGmXNZ5552XzrTsu9Z7/m/Tvc6/eQGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYCi1Nm8+/bti/vuuy81YHR0NLU+IuLWW29NZyIi9u/fn87MnTs3nRkfz5+zvXfv3nRmYmIinYmIuP/++9OZe+65J7W+5fW02LdvX/r1zJkzJz2ndc+1nO3csudaDhxv+Yx27dqVzkREPPDAA+lM9l7SS7+L97qWPdTygIWWfbd79+50JqLtXnfvvfem1k/39fhlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAipQpABQpUwAoSh103+l0YmhoKDVgYGAgtT4iYnBwMJ2JaDt0vGVWyyHTLZmpqal0JiJiZGQkncl+Tp1OJz2jxYwZM9Kvp+UzHR4eTmci2vZcy6zJycl0pmX/tO65sbGxdCZ7L+mlGTNmxOzZs1OZls+15bsa0bt9t2/fvnSmV3s1ImL+/PnpzMyZM1PrZ8yY3m9Ov0wBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKUk+NGRoaiiOOOOJAXcv/OOqoo5pyfX196cwhhxySztx0003pTK+eshIR6addRETs2bMntb7lqRUt+vr6YnR0NJXp5dM7Wp4G1PJUm23btqUzLZ9R69M7li9fns5M92kcvwkDAwOxZMmSVKblvTvssMPSmYi2+8nSpUvTmf/6r/9KZ1q0PJ0mou192L59e2r9dD/X397dDACPEsoUAIqUKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKEoddD8xMRG33XZbasBJJ52UWh8RcfPNN6czEW0He2cPUY+IeMMb3pDO7N27N51Zu3ZtOhMRcdppp6Uzn/zkJ1PrV6xYkZ7RYsGCBXHeeeelMi2Hcz/rWc9KZyIidu7cmc4cd9xx6cxXv/rVdGZgYCCdGRoaSmciIp7znOekM/fcc0/TrF6YN29evPCFL0xlWvbd2NhYOhMRsWvXrnTmcY97XDrTci9ueeBIy8MfItr23e7du1Pr+/unV5N+mQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAijqZw+E7nc76iFhz4C6HR5Fl3W53/EAPsef4JfYdvTatPZcqUwDgkfybFwCKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoKg/s3hkZKQ7d+7c1IA9e/ak1kdEDA8PpzMREZ1Opyez+vtTb1uzqampptzGjRvTmex7t3PnzpiYmMi/4Umjo6Pd8fHxVGbmzJnpORMTE+lMRES3201n9u3bl87s3LkznRkYGOhJJqLtuzcyMpLO3H777Ru63W5uQzSYM2dO9+CDD05lRkdH03N27NiRzrRquRdv3rw5nRkcHOxJpjWXvT+sW7cutmzZ8is3eKoV5s6dG69+9atTF3L33Xen1kdEHHPMMelMRMTQ0FA6c+SRR6YzCxYsSGdabjZbt25NZyIivvCFL6Qz2ZvoN77xjfSMFuPj4/H+978/lXnCE56QnnPHHXekMxERk5OT6cz999+fzvzwhz9MZxYuXJjOLFmyJJ2JaCvhJz7xienMmWeeuSYdanDwwQfH3/zN36QyT33qU9NzbrzxxnQmImL//v3pzOrVq9OZL33pS+nMoYcems4sXbo0nYmIWLZsWTrz+Mc/PrX+vPPOm9Y6/+YFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFqeME9+zZkz527eijj06tj4j46U9/ms5EtJ2v+h//8R/pzLnnnpvOtBy3tnfv3nQmou24wxUrVqTWf/e7303PaLFly5b4yle+ksr8+Mc/Ts9pOfYyou0zajlP9Pzzz09nZszI/628ffv2dCYi4uqrr05nWo/L7IV77703/uzP/iyVOe6449Jzbr311nQmou1M6IMOOiidueiii9KZlnOkH3zwwXQmIuKd73xnOvOMZzwjtX661+aXKQAUKVMAKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKFKmAFCkTAGgKHXQ/a5du+InP/lJasDk5GRqfUT74c/79+9PZ4aHh9OZI488Mp2ZmppKZx566KF0JiLiW9/6VjqTPTh7165d6Rktdu3aFbfccksq03II+M9+9rN0JqJtz82ePTud+eIXv5jOzJ8/P50ZGRlJZyIifvCDH6Qzc+bMaZrVCzNmzEh/Ti2f69y5c9OZiLY93nKvO+yww9KZlnt+6/vwmMc8Jp1ZsmRJav10H0zhlykAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkBR6qkxo6Oj8bSnPS01YGJiIrU+IuL0009PZyIi9uzZk84MDQ2lM//2b/+WzsyaNSudGRgYSGciIk455ZR0Jvs5tTwtpcXIyEgcf/zxqUxfX196znHHHZfORLQ9DahlL1xwwQXpTMtntHnz5nQmIuLmm29OZ3q1h1qMjo7GU5/61FRmx44d6TmnnnpqOhPRdq9r+V6sXbs2nWn5TmzZsiWdiYg44ogj0pmNGzem1k/3KTh+mQJAkTIFgCJlCgBFyhQAipQpABQpUwAoUqYAUKRMAaBImQJAkTIFgCJlCgBFyhQAilIH3S9evDguueSS1IBvf/vbqfURkT5g+hdaDn9+6KGH0pm5c+emM3v37k1n1q1bl85ERFx11VUHPHP99denZ7Q44ogj4h//8R9TmU984hPpOa9//evTmV7qdDq/6Uv4X3W73XTmr//6r9OZlr3dYunSpfGxj30slbnmmmvSc84555x0ppd+F/fdgbrX+WUKAEXKFACKlCkAFClTAChSpgBQpEwBoEiZAkCRMgWAImUKAEXKFACKlCkAFClTACjqZA4K7nQ66yNizYG7HB5FlnW73fEDPcSe45fYd/TatPZcqkwBgEfyb14AKFKmAFCkTAGgSJkCQJEyBYAiZQoARcoUAIqUKQAUKVMAKPq/2h8shMDuotMAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f3249018dd8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Visualize features and labels from each batch\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import math\n",
    "import cairo\n",
    "from mpl_toolkits.axes_grid1 import make_axes_locatable\n",
    "from batches import features, labels, batch_ind, samples_in_batch_by_label, N_ODOR_CLASSES\n",
    "\n",
    "n_feats = 10\n",
    "batches = list(range(10))\n",
    "\n",
    "x = []\n",
    "y = []\n",
    "k = 3\n",
    "for batch in batches:\n",
    "    xb = []\n",
    "    yb = []\n",
    "    for c in range(N_ODOR_CLASSES):\n",
    "        choices = samples_in_batch_by_label[batch][c]\n",
    "        samp = np.random.choice(choices, k, replace=False)\n",
    "        xb.append(features[samp, :n_feats])\n",
    "        yb.append(labels[samp])\n",
    "    x.append(np.stack(xb))\n",
    "    y.append(np.stack(yb))\n",
    "# x shape (batches, classes, k, features)\n",
    "# y shape (batches, classes, k,)\n",
    "x = np.stack(x)\n",
    "y = np.stack(y)\n",
    "print(\"x shape\", x.shape, \"y shape\", y.shape)\n",
    "\n",
    "fig, axes = plt.subplots(x.shape[0], x.shape[2], figsize=(8, 16))\n",
    "for batch in range(x.shape[0]):\n",
    "    for samp in range(x.shape[2]):\n",
    "        ax = axes[batch, samp]\n",
    "        data = x[batch, :, samp]\n",
    "        im_gray = ax.imshow(data, cmap=\"gray\")\n",
    "        im_gray.set_clim(-1.0, 1.0)\n",
    "\n",
    "        ax.set_xticks([])\n",
    "        ax.set_yticks([])\n",
    "# plt.colorbar(im_gray)\n",
    "fig.savefig(\"writeup/batch_matrix_raw.svg\", format=\"svg\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Basic classification techniques"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ANOVA\n",
    "\n",
    "Here we run a one-way ANOVA to test the null hypothesis that the data from the different batches have the same mean. This analysis however makes the assumption that samples are independent which isn't evidently true looking at the plots above."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<function scipy.stats.stats.f_oneway(*args)>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import scipy.stats\n",
    "\n",
    "features_z = scipy.stats.zscore(features, axis=0)\n",
    "unique_labels = np.unique(labels)\n",
    "features_by_label = []\n",
    "for label in unique_labels:\n",
    "    features_by_label.append(features_z[labels==label])\n",
    "\n",
    "scipy.stats.f_oneway"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Principal Components"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.52365805 0.15396908]\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from sklearn.decomposition import PCA\n",
    "from batches import split_all\n",
    "\n",
    "pca = PCA(n_components=2)\n",
    "new_features = pca.fit_transform(split_all.features)\n",
    "print(pca.explained_variance_ratio_)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f3d088f10b8>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.colors as mcolors\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "plt.figure(figsize=(8, 8))\n",
    "colors = list(mcolors.TABLEAU_COLORS)\n",
    "for i, (start, end) in enumerate(split_all.batch_ind):\n",
    "    batch = i+1\n",
    "    batch = new_features[start:end]\n",
    "    plt.scatter(batch[:, 0], batch[:, 1], color=colors[i], label=batch, alpha=100/(end-start))\n",
    "# plt.legend()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## LDA"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Inference of the chemical from the features, within-batch\n",
    "\n",
    "An LDA is able to discriminate odors within a batch (no generalization tested)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(12077,) (12077, 128)\n",
      "(371, 128) (371,)\n",
      "(1239, 128) (1239,)\n",
      "(1586, 128) (1586,)\n",
      "(161, 128) (161,)\n",
      "(197, 128) (197,)\n",
      "(1833, 128) (1833,)\n",
      "(3045, 128) (3045,)\n",
      "(276, 128) (276,)\n",
      "(369, 128) (369,)\n",
      "(3000, 128) (3000,)\n",
      "================================================================================\n",
      "Accuracy achieved by batch 0: 1.0\n",
      "Accuracy achieved by batch 1: 1.0\n",
      "Accuracy achieved by batch 2: 0.9993694829760403\n",
      "Accuracy achieved by batch 3: 1.0\n",
      "Accuracy achieved by batch 4: 1.0\n",
      "Accuracy achieved by batch 5: 0.9950900163666121\n",
      "Accuracy achieved by batch 6: 0.9990147783251232\n",
      "Accuracy achieved by batch 7: 1.0\n",
      "Accuracy achieved by batch 8: 1.0\n",
      "Accuracy achieved by batch 9: 0.9956666666666667\n"
     ]
    }
   ],
   "source": [
    "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "print(labels.shape, features.shape)\n",
    "batches_feats = [features[s:t] for s, t in batch_ind]\n",
    "batches_label = [labels[s:t] for s, t in batch_ind]\n",
    "batches_accuracy = []\n",
    "batches_lda_models = []\n",
    "for feats_batch, labels_batch in zip(batches_feats, batches_label):\n",
    "    X = feats_batch\n",
    "    y = labels_batch\n",
    "    print(X.shape, y.shape)\n",
    "    clf = LinearDiscriminantAnalysis()\n",
    "    batches_lda_models.append(clf)\n",
    "    clf.fit(X, y)\n",
    "    y_pred = clf.predict(X)\n",
    "    accuracy = (y_pred==y).astype(np.long).sum().item() / y.shape[0]\n",
    "    batches_accuracy.append(accuracy)\n",
    "print(\"=\"*80)\n",
    "for batch_i, accuracy in enumerate(batches_accuracy):\n",
    "    print(f\"Accuracy achieved by batch {batch_i}: {accuracy}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Inference of the chemical from the features, within-batch (50/50 train/test split)\n",
    "\n",
    "With a 50/50 train/test split, the LDA classifier succeeds to generalize to classify within-batch."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(12077,) (12077, 128)\n",
      "(185, 128) (185,)\n",
      "(619, 128) (619,)\n",
      "(793, 128) (793,)\n",
      "(80, 128) (80,)\n",
      "(98, 128) (98,)\n",
      "(916, 128) (916,)\n",
      "(1522, 128) (1522,)\n",
      "(138, 128) (138,)\n",
      "(184, 128) (184,)\n",
      "(1500, 128) (1500,)\n",
      "Accuracy achieved by batch 0: 0.8454935622317596\n",
      "Accuracy achieved by batch 1: 0.9787798408488063\n",
      "Accuracy achieved by batch 2: 0.9947753396029259\n",
      "Accuracy achieved by batch 3: 0.9791666666666666\n",
      "Accuracy achieved by batch 4: 0.9646017699115044\n",
      "Accuracy achieved by batch 5: 0.990036231884058\n",
      "Accuracy achieved by batch 6: 0.9994609164420485\n",
      "Accuracy achieved by batch 7: 0.9349112426035503\n",
      "Accuracy achieved by batch 8: 0.9955555555555555\n",
      "Accuracy achieved by batch 9: 0.9846238330587589\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/sklearn/discriminant_analysis.py:388: UserWarning: Variables are collinear.\n",
      "  warnings.warn(\"Variables are collinear.\")\n",
      "/usr/local/lib/python3.6/dist-packages/sklearn/discriminant_analysis.py:388: UserWarning: Variables are collinear.\n",
      "  warnings.warn(\"Variables are collinear.\")\n",
      "/usr/local/lib/python3.6/dist-packages/sklearn/discriminant_analysis.py:388: UserWarning: Variables are collinear.\n",
      "  warnings.warn(\"Variables are collinear.\")\n"
     ]
    }
   ],
   "source": [
    "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "print(labels.shape, features.shape)\n",
    "batches_feats = [features[s:t] for s, t in batch_ind]\n",
    "batches_label = [labels[s:t] for s, t in batch_ind]\n",
    "batches_sizes = [batch.shape[0] for batch in batches_label]\n",
    "R_inds = [np.random.choice(np.arange(size), int(size/2)) for size in batches_sizes]\n",
    "T_inds = [np.setdiff1d(np.arange(size), ind, assume_unique=True) for ind, size in zip(R_inds, batches_sizes)]\n",
    "batches_feats_R = [feat[ind] for feat, ind in zip(batches_feats, R_inds)]\n",
    "batches_label_R = [label[ind] for label, ind in zip(batches_label, R_inds)]\n",
    "batches_feats_T = [feat[ind] for feat, ind in zip(batches_feats, T_inds)]\n",
    "batches_label_T = [label[ind] for label, ind in zip(batches_label, T_inds)]\n",
    "batches_lda_models = []\n",
    "for feats_batch, labels_batch in zip(batches_feats_R, batches_label_R):\n",
    "    X = feats_batch\n",
    "    y = labels_batch\n",
    "    print(X.shape, y.shape)\n",
    "    clf = LinearDiscriminantAnalysis()\n",
    "    batches_lda_models.append(clf)\n",
    "    clf.fit(X, y)\n",
    "\n",
    "# test    \n",
    "batches_accuracy = []    \n",
    "for feats_batch, labels_batch, clf in zip(batches_feats_T, batches_label_T, batches_lda_models):\n",
    "    X = feats_batch\n",
    "    y = labels_batch\n",
    "    y_pred = clf.predict(X)\n",
    "    accuracy = (y_pred==y).astype(np.long).sum().item() / y.shape[0]\n",
    "    batches_accuracy.append(accuracy)\n",
    "\n",
    "for batch_i, accuracy in enumerate(batches_accuracy):\n",
    "    print(f\"Accuracy achieved by batch {batch_i}: {accuracy}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "int"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(np.arange(5)[0].item())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Inference of the chemical from the features, between-batch\n",
    "\n",
    "Make a matrix between batches. This shows that, as you would expect, discriminators trained on one batch perform the best on themselves, followed by their neighbors."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.colorbar.Colorbar at 0x7f3cb4c6bef0>"
      ]
     },
     "execution_count": 123,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWAAAAEWCAYAAABYGk2QAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4xLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvAOZPmwAAIABJREFUeJzt3Xu8HVV99/HPNycJ4RKMEBRIAkSgPEQEgRBsbREFfYAqWKWVtF6wKkVFxYoWrC9E1NqLpVIbqxGpFxCKgDb6RILKxSuXcBFNIpiikhAQEu6XkMv5PX+sdWCy2XuffXJmn9mz832/XvM6Zy57zZrZM7+9Zs2aNYoIzMxs7I2rOgNmZlsqB2Azs4o4AJuZVcQB2MysIg7AZmYVcQA2M6tIpQFY0h6SQtL4PP5dSW/ZjHR2k/SYpIHyc2mStpb0bUkPS/pG1fkBkDRL0uIOlvuwpPM2cx1/JenKwnhI2msz0xrVMSrpBkkvbDP/REk/3py060DSWZIuGIP1bBKTum3YACzpt5KezAfP7yX9l6TtupGZiDg6Ir7SYZ6OLHzurojYLiI2diNfxvHA84EdI+LPG2e2OzkKx8+jkh6S9FNJJ0t61rGX0wlJczrI08eBTw+3UET8Q0S8vYP0mn32woh41eZ8tklamxyjkq6RNJJ8fRo4u4y8NNoCgnfPbl+nJeDXRMR2wEHAIcBHGhdQ4iqNEo3Vr3AHdgfuiIgNm/n510TE5JzOPwJ/B3ypuIAkAW8CHgDaXgVJ2gV4OfCtzczPmCrpe1wAvDxvu/WLiGg7AL8FjiyM/wvwnfz/NcAngZ8ATwJ7Ac8hnVz3AHcDnwAG8vIDpF/y1cCdwLuBAMYX0nt7YV3vAJYBjwJLST8AXwMG8/oeAz4E7NGQzq6kA/YBYDnwjkKaZwGXAF/N6S4BZrfZ/nOBFcAjwE3AnxTmDQAfBv43p3UTMCPPeyHwvZyH3wMfztO/DHyikMbhwMqG/f13wG3AU8B44PTCOpYCf9aQx2b76YPAZQ3LfRb4TIvt3Dfv/4fyPjk2T/8YsA5Yn/f325p89izggk6OnzxtTv4O9ytMOyx/p28E1gAT23wnbwa+3zDt70jH26PA7cARjXkrHCdvzd/pg8DJpELFbXnb/6OQ5onAjwvjAeyV//9T4JZ8XKwAziosN7SetwF3AT8sTBtPOmc2AmvzPv0PYB7wrw3b9G3g1ML494C3tNgnJ5LOw88CDwO/GtoHeX7T8zJ/72tzfh7L+2Bm/jsuf/Y84L5CWhcM5atVuoVl/5p0bD4ILAJ2b9ifJwO/zvPnAWqxfWcBlwL/nb/jm4EDCvObniPNti9P3xr4V+B3eX/9OE8b+p7ekr+71cDfF9YzrrCuNaRYskOeNynvmzV5/90IPL9tfB1JAAZmkE7OjxcC5l2kYDMemEAqlXwB2BZ4HnAD8Dd5+ZPzgTED2AG4mhYBGPjz/IUeAogU3Hdv8aOwR0M61wKfyzvkxcD9bHpCrgWOIR2AnwKua7P9bwR2zNv3AeBeYFKe90HgF8A+OY8H5GUnkw7ID+Q8TAYOHUEAvjXvo60L+2LX/OW/AXgc2KXdfgJ2yctNycuNB+4DDm6yjRNIP1QfBiYCryAdyPsMF2A3JwDn6XcB7yyMf4l0ME8gHcCva7O+fwHmFcb3IQXBXQvHw55tAvDn8/fyqnwsfIt0rE7L++hlHQTgw4EX5e9kf9KP7Gsb1vNV0nlQPLFbFTbmAKt4JuhNBZ6gcAID/w6c0yYAbwDen/fhG0iBZSg4tDsvN9nOwvdzcP7/dlKBad/CvAM7SPe1pONqX9Lx9xHgpw378zvAFGA30nl6VJtjbD2pOmwCcBrwG2BCB+dIs+2bl7+DaaQ48EfAVoXv6Yv5ezuAVBAa2vZTgeuA6Xn5LwAX5Xl/Q/rR3CaneTCwfRkBeOiX8XekwDYUGK4Bzi4s+/yc2a0L0+YCV+f/rwJOLsx7Fa0D8CLgfZ2c1GxauphB+rWbXJj/KeDLhS/y+4V5s4Anh9sPheUfJP/ykg7M45osMxe4pcXnv8zwAfivh8nDrUPrHWY/fZdc+gdeDSxtsdyfkH5YxhWmXUQu1dGdAHwduWSRD9hHeCaAfQH4nzbr+yLwj4XxvUiB80jyCdksb4XjZFph/hrgDYXxy3imdHciLQJwkzx9Bvi3hvW8oNkx2nisF5ZZBrwy/38KsLBh/ieB81us/0RSAFdh2g2kap3hzstNtjNP+xrwt8DOpOP8n0kFqKdLxx2k+10KV0z5M0/wTEEqgD8uzL8EOL3NMXZdQ1r3ULgibXOONH6P40hXWwc0+dzQ9zS9YT+eUPiOilcWu5B+GMaTSvs/BfZvd/4Wh07rbF8bEVMiYveIeFdEPFmYt6Lw/+6kX6d78g2Xh0gn0/Py/F0blv9dm3XOIBXzR2pX4IGIeLRhPdMK4/cW/n8CmNSqnk7SByQtyy0AHiJdck0dJo+bm/chxX2EpDdLurWwT/frIA8AXyGV4Ml/v9ZiuV2BFRExWJjWuM/KNo1UPQPwZ6TS28I8fiFwtKSdWnz2QdJVBQARsZxUMjkLuE/SxZJ2bbPu3xf+f7LJ+LA3mSUdKulqSfdLepgUnKY2LLaiyUfbGe77mkwKfq3cHTkqZL8jfbfDnZfNXEsqHBxGqkK5BnhZHn6Uj5Xh0t0dOLcw7wHSVVq7c7Hdvn96f+b1r8zbN9w50mgq6Qqo3TnaKl+7A98srGcZqcD3fNL3tQi4WNIqSf8saUKbdZTSDK34ha8g/SJOzQF7SkRsHxFDzWfuIQWMIbu1SXcFsGcH62y0CthB0uTCtN1Il+kjIulPSHWLfwE8NyKmkC7rNEwe2+X9cVKJb8jOTZZ5evsk7U4q8Z1CaoUwBfhlB3mAdHm4v6T9SCXgC1sstwqY0XATdbP2WSckHUI6CYfuTL+FdIDfJele4BukE3tuiyRuA/6gOCEivh4Rf0w6QQL4py5kvejrpPsMMyLiOaRqDTUs0+44bTbvAuA4SQeQLtsbbzLuC/y8TZrT8s3MIbuRvtvhzstmebmWdGV0eP7/x8BLSQH42rzMcOmuIFVHTCkMW0fET9tsQztPx458rE4HVnVwjjRu32pS1VOr86adFcDRDds0KSLujoj1EfGxiJhFqtJ4Nel+RUultlqIiHuAK4F/lbS9pHGS9pT0srzIJcB7JU2X9FxSZXYr5wGnSTo4t7DYK+9oSCWWF7TIwwrSZcCnJE2StD/pZkir4NPOZFLJ7H5gvKQzge0b8vhxSXvnPO4vaUdSvdbOkk6VtJWkyZIOzZ+5FThG0g6SdiaV3NrZlnQA3Q8g6a2kX/diHprup4hYS7px8XXghoi4q8U6rif9MHxI0gRJhwOvAS4ebgcVjMv7e2jYqnGBfEy8Oqd7QUT8QtI04AjSwfriPBxACqBvabGu7wEHSZqU091H0ivyOteSSrHdbpI4mXSltTY3m/vLEX7+WcdwRKwk3bj5GukG6tNXmnnbDiZteyvPI51fEyT9OSlgL+zgvPw9MF3SxEJefs0zN0V/GBGP5OVeTw7AHaT7eeAM5fbLkp6T87W5Dpb0uny1eiop+F/H8OfIJtuXS8/nA+dI2lXSgKQ/bHbMNvF54JND55iknSQdl/9/uaQXKbX1foRUNdH2OOxGs7E3k27kLCVdKl5KqieB9Cu1iPQrfjNweatEIuIbpDqvr5NuCH2LdOMOUp3uR/JlwGlNPj6XVJezCvgm8NGIaHfgtrKIVI91B+lybi2bXlaeQ/pRuZK0w79Eqg97FHglKYjdS7rL+/L8ma+Rtv+3+XP/3S4DEbGUdLf2Z6QD6UWku91D89vtJ0iXtS+idfUDEbEOOBY4mlQ6+Bzw5oj4Vbu8NZhLOmGHhuLl3bclPUrad39P2m9vzfPeBNwaEVdGxL1DA+mG01DpvTG/vyfdTzguT9qK1LxtNWl/P490Q7Gb3gWcnbfrTNJxMBLnAsdLelDSvxemt/q+jgWuiYhVbdK8HtibtB8+CRwfEWvyvHbn5VWkm+v3SlpdSO9aYE3hh/taUqnylsIyLdONiG+SfkgvlvQIqVR6dJv8D+d/SDfYHiQdN6/Lpc6250iL7TuNdAP9RlLVyD/RWTw8l3Tlc2X+7q8DhgpXO5O2/xFS1cS1pKualrRplZH1G0m7kVqe7JxLMX1B0ixSsJoTfXQQSzqMdNLuUayTl3Q96YbWLyvLnJXOAbiP5Xqyc0hNYf666vxYe/mGzcXAzyOiK0+9WW/xk2t9StK2pEuhVwIfrTg7NgxJ+5JaOOxCatJmY0TS+ZLuk9T06iLfW/l3Scsl3SbpoNLW7RKwmW3JcrXPY8BXI+JZ9xwkHQO8h/Tw1qHAuRFxaONym8MlYDPbokXED3mmTXozx5GCc0TEdcAUldQnR6909tIzpu4wEHvMaNt2erMsu7vVMwWjM7B2cPiFekW3rraeWNuddNXYrLcs5e+HbvWD1Y0r5LXxOOti7ah27v99+bax5oHOWhredNtTS0gtmIbMj4j5I1jdNDZt/bQyT7tnBGk05QDcYI8ZE7hh0YzhFxyhOWe8s/Q0Aab8+onS09RgdwKlnupO09y4ZUlX0tWEicMvtDmi/B9NTexOXmP95naA19p1668YdRprHtjIDYvaPcf1jIFdfr02ImaPYnXNfixKOUkcgM2sdgIYZMyu/lay6RO800nPGIya64DNrHaCYH1s7GgowQLgzbk1xEuAh/NTgKPmErCZ1VJZJWBJF5H6vJgqaSWp2eYEgIj4PKmTqGNIXWs+wTNPcY6aA7CZ1U4QbCzpBmFEtOr0aWh+kF4eUToHYDOrpcEutCYZaw7AZlY7AWzsgwDc9zfhJB0l6fb8GGG77i/NrEYGiY6GXtbXJeDcL+c8Un8IK4EbJS3I3deZWU0FsL4PulHo9xLwHGB5RNyZ+7y9mGf6kDWzmgqCjR0OvazfA3CrRwg3IekkSYslLb5/TbdfpGBmoxawscOhl/V7AO7oEcKImB8RsyNi9k47DoxBtsxsNNKTcJ0Nvayv64Dp4iOEZlYlsbFp+ape+j0A3wjsLWkm6Q2/JzDylyeaWY9JN+EcgHtaRGyQdArp5ZoDwPkR0Z2us8xszKR2wA7APS8iFpKe5TazPjLoErCZ2dhzCdjMrCKB2NgHjbgcgM2sllwFYWZWgUCsi/q32XcANrPaSQ9iuAqi72xkkIcHnyw93Z/9w7zS0wT43EMzS09z9YbtSk8T4LENW3Ul3ctvPqQr6U68r0unR48/Hls08ZHyL/PXf+WHpaTjm3BmZhWIEBvDJWAzs0oMugRsZjb20k24+oev+m+BmW1xfBPOzKxCG90O2Mxs7PlJODOzCg32QSuI+m+BmW1xUmc84zoahjPcm9Ml7S7pB5Juk3SNpOllbYcDsJnVTiDWx0BHQzuFN6cfDcwC5kqa1bDYp4GvRsT+wNnAp8raDgdgM6udCNgY4zoahtHJm9NnAT/I/1/dZP5mcwA2sxoSgx0OwNSht57n4aRCQp28Of3nwOvz/38GTJa0Yxlb4ZtwZlY7ASN5FHl1RMxuMa+TN6efBvyHpBOBH5LeL7mh05W34wBsZrVUUjO0Yd+cHhGrgNcBSNoOeH1EPFzGyl0FYWa1E4jB6GwYxtNvTpc0kfTm9AXFBSRNlTQUK88Azi9rOxyAzax20mvpx3c0tE0nYgMw9Ob0ZcAlEbFE0tmSjs2LHQ7cLukO4PnAJ8vaDldBmFkNqbT+gJu9OT0iziz8fylwaSkra+AAbGa1E/THk3AOwGZWS34jhplZBSLkErCZWRXSTTi/FdnMrAJ+J1xfSpX7XXhtbZeqq9415Telp/lYPFV6mt30uiMWdyXd7z+6X1fS7ca7zNYOTig9TYBv3bF/6WnGNwZHnwZ00sa35zkAm1ktuUN2M7MKDD0JV3cOwGZWS34pp5lZBSJg/aADsJnZmEtVEA7AZmaV8JNwZmYV6JdmaPUvw7chaYakqyUtk7RE0vuqzpOZlSFVQXQy9LJ+LwFvAD4QETdLmgzcJOl7EbG06oyZ2eh044GWsdbXATgi7gHuyf8/KmkZ6YV7DsBmNZZaQbgviNqQtAdwIHB9k3knAScBTJ9W/y/VrN/1y4MYvV1BUpL8Ir3LgFMj4pHG+RExPyJmR8TsHXfcInaJWe2N4LX0PavvS8CSJpCC74URcXnV+TGz0euXVhB9HYAlCfgSsCwizqk6P2ZWnl5v4dCJvg7AwEuBNwG/kHRrnvbh/BI+M6upCLHBAbi3RcSP6VpPvGZWJVdBmJlVoF/qgOtfhjezLdJgqKNhOJKOknS7pOWSTm8yf7f8RO0tkm6TdExZ2+AAbGa1M9QOeLQBWNIAMA84GpgFzJU0q2GxjwCXRMSBwAnA58raDgdgM6ulktoBzwGWR8SdEbEOuBg4rmGZALbP/z8HWFXWNrgOuIGAcapP3dIg5b9AdBtNLD3Nbtp9/BNdSXdAo395ZDNPbSx//27oUufkW221vvQ0pdEfsxGlbfM0YEVhfCVwaMMyZwFXSnoPsC1wZBkrBpeAzaymRlAFMVXS4sJwUiGZZqWtxl+IucCXI2I6cAzwNUmlxE6XgM2sdkbYF8TqiJjdYt5KYEZhfDrPrmJ4G3AUQET8TNIkYCpwX+c5bs4lYDOrpQh1NAzjRmBvSTMlTSTdZFvQsMxdwBEAkvYFJgH3l7ENLgGbWS2V0dFORGyQdAqwCBgAzo+IJZLOBhZHxALgA8AXJb2fVD1xYkSUcvPFAdjMaieivAcxctcECxumnVn4fympW4PSOQCbWQ2JjX4tvZlZNTqo3+15DsBmVjv90heEA7CZ1U+keuC6cwA2s1rq9dcNdcIB2MxqJ3wTzsysOq6CMDOriFtBmJlVIMIB2MysMm6GZmZWEdcBm5lVIBCDbgVhZlaNPigAOwCbWQ35JpyZWYX6oAjsAGxmteQSsFkPeKpLJaEnuvD2YoAnBrvxVuSB0tMEGCjhDcaNynjpeACDgw7AZmZjLwCXgM3MquF2wGZmVXEANjOrQkevnO95DsBmVk8uAZuZVSAg3ArCzKwq9Q/AtenNQtLzJX1J0nfz+CxJb+vwswOSbpH0ne7m0szGTHQ4DEPSUZJul7Rc0ulN5v+bpFvzcIekh8rahNoEYODLwCJg1zx+B3Bqh599H7CsC3kys6qUEIAlDQDzgKOBWcBcSbM2WU3E+yPixRHxYuCzwOVlbUKdAvDUiLgEGASIiA3AxuE+JGk68KfAed3NnpmNmaEHMToZ2psDLI+IOyNiHXAxcFyb5ecCF5WzEfUKwI9L2pH8mybpJcDDHXzuM8CHyIG7GUknSVosafHqNS0XM7Mekl5LNPwATB06v/NwUiGZacCKwvjKPO1ZJO0OzASuKmsb6nQT7m+BBcCekn4C7AQc3+4Dkl4N3BcRN0k6vNVyETEfmA9w4AET+6Bxi9kWoPNWEKsjYnaLec0SaRUDTgAujYhhr7w7VZsAHBE3S3oZsA9pp90eEeuH+dhLgWMlHQNMAraXdEFEvLHL2TWzLiupn6CVwIzC+HRgVYtlTwDeXcpas9pUQUh6N7BdRCyJiF8C20l6V7vPRMQZETE9IvYg7byrHHzN+kCnN+CGD9I3AntLmilpIilOLGhcSNI+wHOBn5WzAUltAjDwjoh4uvlHRDwIvKPC/JhZZTq8ATfMTbh8M/8UUgurZcAlEbFE0tmSji0sOhe4OKLcLoBqUwUBjJOkoR2Qm4903LFqRFwDXNOdrJnZmCspFEbEQmBhw7QzG8bPKmdtm6pTAL4SuETS50m7/mTgimqzZGaV6YMGS3UKwB8CTgLeSboJdyVu22u2ZXKH7GMnVzd8Jd9A+3zV+TGz6nXhbUljrhY34XK7u53yXUozs9L6gqhSLUrA2W+Bn0haADw+NDEizqksR2Zmo1CnALwqD+OAyd1aydoQy9eX/4bZvSY8VXqaABPozttw62SXge5cGL1jh592Jd1u3DtaF925mN1p4qOlp/nFiU+Ukk4/VEHUJgBHxMeqzoOZ9YhgJI8i96zaBGBJV9OkRiciXlFBdsysai4Bj6nTCv9PAl4PbKgoL2ZWMVdBjKGIuKlh0k8kXVtJZsyseg7AY0fSDoXRccDBwM4VZcfMquYAPKZuIu1ykaoefgN09E44M+svCldBjKmImFl1Hsysh7gVxNiRNIHUD8RhedI1wBc66JTdzPqQS8Bj6z+BCcDn8vib8rS3V5YjM6uOA/CYOiQiDiiMXyXp55Xlxsyq0yd1wLXojCfbKGnPoRFJL6CD19KbWZ9yZzxj6oPA1ZLuzON7AG+tLjtmViX1QYfsPV8ClnSIpJ0j4gfA3sDlwCOkDtldBWFmtdXzARj4ArAu/38ocDrwFeD3wPyqMmVmFXMVxJgYiIgH8v9vAOZHxGXAZZJurTBfZlYV34QbMwOShn4ojgCuKsyrww+ImXWDS8Bj4iLgWkmrgSeBHwFI2gt4uMqMmVmFejy4dqLnA3BEfFLSD4BdgCsjYmi3jwPeU13OzKwqwq0gxkxEXBcR34yI4rvg7oiIm6vMl5lVJJ7pkGe4YTiSjpJ0u6Tlkk5vscxfSFoqaYmkr5e1GT1fAjYza6qEKghJA8A84JXASuBGSQsiYmlhmb2BM4CXRsSDkp43+jUntSgBm5k9Szk34eYAyyPizohYB1wMHNewzDuAeRHxIEBE3FfSFrgE3Ojxwa247sk9h19whCaP+1XpaQJMGdcHFWGj1K1SxA4D9Xnj9GB0547UvlvfXXqak8aV04HhCJqhTZW0uDA+PyKGniGYBqwozFtJet6g6A8AJP0EGADOiogrRpzhJhyAzayeOg/AqyNidot5zToVbkx5POkp3MOB6cCPJO0XEQ91nIMWXAVhZvUTqRVEJ8MwVgIzCuPTgVVNlvmfiFgfEb8BbicF5FFzADazeiqnDvhGYG9JMyVNBE4AFjQs8y3g5QCSppKqJO6kBA7AZlZLZTRDi4gNwCnAImAZcElELJF0tqRj82KLgDWSlgJXAx+MiDVlbIPrgM2snkq67xgRC4GFDdPOLPwfwN/moVQOwGZWPzXo56ETDsBmVjvCvaHVgqQpki6V9CtJyyT9YdV5MrPRK+tR5CptCSXgc4ErIuL4fJdzm6ozZGYl6PHg2om+DsCStgcOA04EyI8armv3GTOriT4IwP1eBfEC4H7gvyTdIuk8Sds2LiTpJEmLJS1+7EHHZ7OeV2JvaFXq9wA8HjgI+M+IOBB4nPROuU1ExPyImB0Rs7d77sSxzqOZbY4+eCNGvwfglcDKiLg+j19KCshmVnMlPYpcqb4OwBFxL7BC0j550hHA0jYfMbOa6IcqiL6+CZe9B7gwt4C4E3hrxfkxs9GqQfVCJ/o+AEfErUCrrujMrK4cgM3Mxl6/PAnnAGxmtaTB+kdgB2Azqx/XAZuZVcdVEGZmVXEA7k+D0dfNo/vOQNP3KvZuuhu7EDnGqTt57eVzwSVgM7OqOACbmVUgev8x4044AJtZ7bgdsJlZlaL+EdgB2MxqySVgM7Mq9MmDGL3bxsTMrI2y+gOWdJSk2yUtl/SsFzZIOlHS/ZJuzcPby9oGl4DNrJbKaAUhaQCYB7yS9AKHGyUtiIjGfsP/OyJOGf0aN+USsJnVT5BuwnUytDcHWB4Rd+aX9l4MHNft7A9xADazWhrBGzGmDr10Nw8nFZKZBqwojK/M0xq9XtJtki6VNKOsbXAVhJnVU+c34VZHRKuXMjR7hrsx5W8DF0XEU5JOBr4CvKLjtbfhErCZ1c7QgxglvBNuJVAs0U4HVhUXiIg1EfFUHv0icHBJm+EAbGY1FIEGOxuGcSOwt6SZ+b2RJwALigtI2qUweiywrKzNcBWEmdVTCe2AI2KDpFOARcAAcH5ELJF0NrA4IhYA75V0LLABeAA4cfRrThyAzayWynoSLiIWAgsbpp1Z+P8M4Ixy1rYpB2Azq58A/E44M7OK1D/+OgCbWT25Mx4zs4r4tfRmZlXok97QHIDNxlg3XvbZjRd99rL0IEb9t9kB2Mzqye+EMzOrhkvAZmZVcB2wmVlVOurnoec5AJtZPbkKwsysAlHOK4mq5gBsZvXUByXgvu8PWNL7JS2R9EtJF0maVHWezKwE0eHQw/o6AEuaBrwXmB0R+5H6+zyh2lyZWRk0ONjR0Mu2hCqI8cDWktYD29DwuhEzq6GgLx7E6OsScETcDXwauAu4B3g4Iq5sXE7SSUNvTH3sgfVjnU0zGyERKDobellfB2BJzwWOA2YCuwLbSnpj43IRMT8iZkfE7O12mDDW2TSzzRHR2dDD+joAA0cCv4mI+yNiPXA58EcV58nMytAHAbjf64DvAl4iaRvgSeAIYHG1WTKzUeuTOuC+DsARcb2kS4GbSW80vQWYX22uzKwMvd7CoRN9HYABIuKjwEerzoeZlan3qxc60fcB2Mz6UOAAbGZWmfrXQPR9Kwgz61NltQOWdJSk2yUtl3R6m+WOlxSSZpe1DQ7AZlZPJTRDkzQAzAOOBmYBcyXNarLcZFK3BteXuQkOwGZWPxGwcbCzob05wPKIuDMi1gEXkx7eavRx4J+BtWVuhuuAGzy8fmu+e98LS0/3gK1/V3qaAAM8Vnqa9ftV7s7NmIF+6HB2lA6atLL0NLdRSY/7d34TbqqkYvv/+REx1Bx1GrCiMG8lcGjxw5IOBGZExHcknba52W3GAdjM6qnzALw6IlrV26pZyk/PlMYB/wacOKK8dcgB2MzqJ4By3gm3EphRGJ/Opj0mTgb2A66RBLAzsEDSsREx6qdqHYDNrIYCopQqohuBvSXNBO4m9Rf+l0+vJeJhYOrQuKRrgNPKCL7gAGxmdRR0coNt+GQiNkg6BVhEemHD+RGxRNLZwOKIWDDqlbThAGxm9VTSk3ARsRBY2DDtzBbLHl7KSjMHYDOrJz+KbGZWBXfGY2ZWjQDcHaWZWUVcAjYzq0KU0gqiag7AZlY/AVFOO+BKOQCbWT2V8yRcpRyAzayeXAdsZlaBCLeCMDOrjEvAZmZVCGLjxqozMWoOwGZWP+V1R1kpB2Azqyc3QzMzG3sBhEvAZmYViNI6ZK+UA7CZ1VI/3IThM/QTAAAEr0lEQVRT9EFTjjJJuh/o5BXGU4HVXc5OmeqU3zrlFeqV317I6+4RsdNoEpB0BYVXBQ1jdUQcNZr1dYsD8GaStLjNm1Z7Tp3yW6e8Qr3yW6e8bgnGVZ0BM7MtlQOwmVlFHIA33/yqMzBCdcpvnfIK9cpvnfLa91wHbGZWEZeAzcwq4gBsZlYRB+DNIOkoSbdLWi7p9Krz04qkGZKulrRM0hJJ76s6T52QNCDpFknfqTov7UiaIulSSb/K+/gPq85TO5Len4+DX0q6SNKkqvO0pXMAHiFJA8A84GhgFjBX0qxqc9XSBuADEbEv8BLg3T2c16L3AcuqzkQHzgWuiIj/AxxAD+dZ0jTgvcDsiNgPGABOqDZX5gA8cnOA5RFxZ0SsAy4Gjqs4T01FxD0RcXP+/1FSgJhWba7akzQd+FPgvKrz0o6k7YHDgC8BRMS6iHio2lwNazywtaTxwDbAqorzs8VzAB65acCKwvhKejyoAUjaAzgQuL7anAzrM8CHgF7vaeUFwP3Af+XqkvMkbVt1plqJiLuBTwN3AfcAD0fEldXmyhyAR05NpvV0Wz5J2wGXAadGxCNV56cVSa8G7ouIm6rOSwfGAwcB/xkRBwKPA718P+C5pCu1mcCuwLaS3lhtrswBeORWAjMK49Pp4Us5SRNIwffCiLi86vwM46XAsZJ+S6raeYWkC6rNUksrgZURMXRFcSkpIPeqI4HfRMT9EbEeuBz4o4rztMVzAB65G4G9Jc2UNJF0I2NBxXlqSpJIdZTLIuKcqvMznIg4IyKmR8QepP16VUT0ZCktIu4FVkjaJ086AlhaYZaGcxfwEknb5OPiCHr4puGWwv0Bj1BEbJB0CrCIdCf5/IhYUnG2Wnkp8CbgF5JuzdM+HBELK8xTP3kPcGH+Ib4TeGvF+WkpIq6XdClwM6l1zC34seTK+VFkM7OKuArCzKwiDsBmZhVxADYzq4gDsJlZRRyAzcwq4mZoVilJOwI/yKM7AxtJj/gCzMn9bZS9zoOA50XEFWWnbTYSDsBWqYhYA7wYQNJZwGMR8elOPy9pICI2jnC1BwH7AQ7AVilXQVjPkvRtSTflPmzfnqeNl/SQpE9IugGYI+nY3D/zjyR9VtK38rLbSfqypBtyhzmvkbQ1cCbwV5JulXR8hZtoWziXgK2XvSUiHpC0DbBY0mXAo8BzgJsj4iN53h2kp/7uAi4pfP5MUn+9J+bOaK4H9gfOBvaLiFPHcmPMGrkEbL3s/ZJ+DvyM1OnRnnn6OuCb+f9ZwO0R8btIj3VeVPj8q4C/z49hXw1MAnYbk5ybdcAlYOtJko4kdXj+koh4UtKPSQEU4Ml45hn6Zt2DUpj32oj434a0Dys9w2abwSVg61XPAR7IwfeFwCEtllsC7JPffyfgDYV5i0iv4QFA0oH530eByV3Is9mIOABbr/p/wDa5CuJMWrzJIyKeAE4Bvg/8iNQ388N59sdyGr+QtAQ4K0+/Cjgg35jzTTirjHtDs9qTtF1EPJZLwF8AfhERn606X2bDcQnY+sE78422pcDWwBcrzo9ZR1wCNjOriEvAZmYVcQA2M6uIA7CZWUUcgM3MKuIAbGZWkf8Ptgc7IXe3dzUAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f3cb4cbf908>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Use the models trained in a previous cell\n",
    "n_batches = len(batches_lda_models)\n",
    "cross_accuracy = np.empty((n_batches, n_batches))\n",
    "for i, model_i in enumerate(batches_lda_models):\n",
    "    for j, (feats_batch_j, labels_batch_j) in enumerate(zip(batches_feats, batches_label)):\n",
    "        X = feats_batch_j\n",
    "        y = labels_batch_j\n",
    "        y_pred = model_i.predict(X)\n",
    "        accuracy = (y_pred==y).astype(np.long).sum().item() / y.shape[0]\n",
    "        cross_accuracy[i, j] = accuracy\n",
    "\n",
    "# plt.figure(figsize=(6, 8))\n",
    "plt.imshow(cross_accuracy)\n",
    "# plt.xticks(np.arange(n_batches), labels=np.arange(n_batches)+1)\n",
    "# plt.yticks(np.arange(n_batches), labels=np.arange(n_batches)+1)\n",
    "plt.title(\"Prediction accuracy of LDA (similarity) between batches\")\n",
    "plt.xlabel(\"Target\")\n",
    "plt.ylabel(\"Source\")\n",
    "plt.colorbar()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f3cb4b9ffd0>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Use the models trained in a previous cell\n",
    "n_batches = len(batches_lda_models)\n",
    "cross_accuracy = np.empty((n_batches, n_batches))\n",
    "for i, model_i in enumerate(batches_lda_models):\n",
    "    for j, (feats_batch_j, labels_batch_j) in enumerate(zip(batches_feats, batches_label)):\n",
    "        X = feats_batch_j\n",
    "        y = labels_batch_j\n",
    "        y_pred = model_i.predict(X)\n",
    "        accuracy = (y_pred==y).astype(np.long).sum().item() / y.shape[0]\n",
    "        cross_accuracy[i, j] = accuracy\n",
    "\n",
    "fig, ax = plt.subplots()\n",
    "x = np.arange(1, 11)\n",
    "# ax.plot(x, cross_accuracy[0], c=\"purple\")\n",
    "# ax.axvline(1, c=\"purple\", linestyle=\"--\")\n",
    "ax.plot(x, cross_accuracy[3], c=\"blue\", marker=\"x\", markersize=10)\n",
    "ax.axvline(4, c=\"blue\", linestyle=\"--\")\n",
    "ax.plot(x, cross_accuracy[6], c=\"black\", marker=\"x\", markersize=10)\n",
    "ax.axvline(7, c=\"black\", linestyle=\"--\")\n",
    "ax.set_xticks(x)\n",
    "ax.set_ylim([0.0, 1.0])\n",
    "ax.set_yticks(np.arange(0, 1.1, 0.1))\n",
    "ax.grid(axis='y', color='#F7BFBE')\n",
    "fig.savefig(\"writeup/lda_centers.png\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Inference of chemical from the features, all the data\n",
    "\n",
    "Can a single LDA model classify every smell in all the batches? Random 50/50 train/test split."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(6955, 128) (6955,)\n",
      "Accuracy achieved: 0.948371817643576\n"
     ]
    }
   ],
   "source": [
    "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "\n",
    "size = features.shape[0]\n",
    "R_ind = np.random.choice(np.arange(size), int(size/2))\n",
    "T_ind = np.setdiff1d(np.arange(size), R_ind, assume_unique=True) \n",
    "batches_feats_R = features[R_ind]\n",
    "batches_label_R = labels[R_ind]\n",
    "batches_feats_T = features[T_ind]\n",
    "batches_label_T = labels[T_ind]\n",
    "\n",
    "X = batches_feats_R\n",
    "y = batches_label_R\n",
    "print(X.shape, y.shape)\n",
    "clf = LinearDiscriminantAnalysis()\n",
    "batches_lda_models.append(clf)\n",
    "clf.fit(X, y)\n",
    "\n",
    "# test    \n",
    "accuracy = []\n",
    "X = batches_feats_T\n",
    "y = batches_label_T\n",
    "y_pred = clf.predict(X)\n",
    "accuracy = (y_pred==y).astype(np.long).sum().item() / y.shape[0]\n",
    "\n",
    "print(f\"Accuracy achieved: {accuracy}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train on 1..T-1, test on T\n",
    "For each batch number T=1...9, (0-indexed) train an LDA model on batch 0..T-1; then evaluate on batch T"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(12077,) (12077, 128)\n",
      "(371, 128) (371,)\n",
      "(1610, 128) (1610,)\n",
      "(3196, 128) (3196,)\n",
      "(3357, 128) (3357,)\n",
      "(3554, 128) (3554,)\n",
      "(5387, 128) (5387,)\n",
      "(8432, 128) (8432,)\n",
      "(8708, 128) (8708,)\n",
      "(9077, 128) (9077,)\n",
      "(12077, 128) (12077,)\n",
      "================================================================================\n",
      "Accuracy achieved by batch 0: 1.0\n",
      "Accuracy achieved by batch 1: 0.9962732919254659\n",
      "Accuracy achieved by batch 2: 0.9949937421777222\n",
      "Accuracy achieved by batch 3: 0.9940422996723265\n",
      "Accuracy achieved by batch 4: 0.994091164884637\n",
      "Accuracy achieved by batch 5: 0.9894189715982922\n",
      "Accuracy achieved by batch 6: 0.9837523719165086\n",
      "Accuracy achieved by batch 7: 0.9838079926504364\n",
      "Accuracy achieved by batch 8: 0.9842458962212185\n",
      "Accuracy achieved by batch 9: 0.9605862383042146\n"
     ]
    }
   ],
   "source": [
    "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "from batches import split_all\n",
    "\n",
    "features, labels, batch_ind = split_all.features, split_all.labels, split_all.batch_ind\n",
    "\n",
    "print(labels.shape, features.shape)\n",
    "\n",
    "# batches_X[T] contains all data 0...T (including batch T) (0-indexed) \n",
    "batches_feats = [features[0:t] for s, t in batch_ind]\n",
    "batches_label = [labels[0:t] for s, t in batch_ind]\n",
    "batches_accuracy = []\n",
    "batches_lda_models = []\n",
    "for feats_batch, labels_batch in zip(batches_feats, batches_label):\n",
    "    X = feats_batch\n",
    "    y = labels_batch\n",
    "    print(X.shape, y.shape)\n",
    "    clf = LinearDiscriminantAnalysis()\n",
    "    batches_lda_models.append(clf)\n",
    "    clf.fit(X, y)\n",
    "    y_pred = clf.predict(X)\n",
    "    accuracy = (y_pred==y).astype(np.long).sum().item() / y.shape[0]\n",
    "    batches_accuracy.append(accuracy)\n",
    "print(\"=\"*80)\n",
    "for batch_i, accuracy in enumerate(batches_accuracy):\n",
    "    print(f\"Accuracy achieved by batch {batch_i}: {accuracy}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f3cb4c0b630>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# batches_lda_models[t] is trained on batches 0...t\n",
    "# so test on batch t+1=T\n",
    "batches = []\n",
    "accuracies = []\n",
    "for T in range(1, 10):\n",
    "    model = batches_lda_models[T-1]\n",
    "    \n",
    "    start, end = batch_ind[T]\n",
    "    X = features[start:end]\n",
    "    y = labels[start:end]\n",
    "\n",
    "    y_pred = model.predict(X)\n",
    "    accuracy = (y_pred==y).astype(np.long).sum().item() / y.shape[0]\n",
    "    \n",
    "    accuracies.append(accuracy)\n",
    "    batches.append(T + 1)\n",
    "\n",
    "fig, ax = plt.subplots(figsize=(6, 4))\n",
    "ax.plot(batches, accuracies, color=\"black\", marker=\"x\", markersize=10)\n",
    "ax.set_xticks(batches)\n",
    "ax.set_ylim([0.0, 1.0])\n",
    "ax.set_yticks(np.arange(0, 1.1, 0.1))\n",
    "ax.grid(axis='y', color='#F7BFBE')\n",
    "\n",
    "\n",
    "fig.savefig(\"writeup/lda_up_to.png\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Neural Network Classifiers"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training curves"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Early stopping pretraining"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# data_folder = \"output/backprop_context_earlystop_share2\"\n",
    "data_folder = \"output/backprop_context_patientearlystop_epoch1\"\n",
    "# data_folder = \"backprop_context_patientearlystop_share_epoch0\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Stop time for T=3: 0\n",
      "Stop time for T=4: 0\n",
      "Stop time for T=5: 0\n",
      "Stop time for T=6: 0\n",
      "Stop time for T=7: 0\n",
      "Stop time for T=8: 0\n",
      "Stop time for T=9: 0\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7fece4d80860>"
      ]
     },
     "execution_count": 84,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fece587a6d8>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import torch\n",
    "import os\n",
    "\n",
    "Ts = list(range(3, 10))\n",
    "samp_max = 20\n",
    "fig, axes = plt.subplots(len(Ts), 1, figsize=(12, 2*len(Ts)))\n",
    "for T, ax in zip(Ts, axes):\n",
    "    lloss = torch.load(os.path.join(data_folder, f\"val_lloss_{T}.pt\"))\n",
    "    lacc = torch.load(os.path.join(data_folder, f\"val_lacc_{T}.pt\"))\n",
    "    li = torch.load(os.path.join(data_folder, f\"val_li_{T}.pt\"))\n",
    "    stop_time = torch.load(os.path.join(data_folder, f\"val_stop_time_{T}.pt\"))\n",
    "    print(f\"Stop time for T={T}: {stop_time}\")\n",
    "    ax.set_ylabel(f\"Batch={T+1}\")\n",
    "    line_loss, = ax.plot(li[:samp_max], lloss[:samp_max])\n",
    "    line_acc, = ax.plot(li[:samp_max], lacc[:samp_max])\n",
    "    ax.hlines(1.0, 0, max(li[:samp_max]), linestyles=\"dashed\")\n",
    "    \n",
    "ax.set_xlabel(\"n samples seen\")\n",
    "    \n",
    "fig.legend([line_loss, line_acc], [\"Training Loss\", \"Testing Accuracy\"], loc=\"upper right\")\n",
    "\n",
    "# [Context is sequences ] [T=4]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0  0.7246666666666667\n",
      "1  0.83\n",
      "2  0.7806666666666666\n",
      "3  0.8153333333333334\n",
      "4  0.7873333333333333\n",
      "5  0.8513333333333334\n",
      "6  0.808\n",
      "7  0.7106666666666667\n",
      "8  0.8093333333333333\n",
      "9  0.7953333333333333\n",
      "10  0.8453333333333334\n",
      "better i=0\n",
      "better i=1\n",
      "better i=5\n",
      "5\n"
     ]
    }
   ],
   "source": [
    "for i, acc in enumerate(lacc[0:20]):\n",
    "    print(f\"{acc}\")\n",
    "\n",
    "\n",
    "patience = 10\n",
    "consec = 0\n",
    "last_best = float('-inf')\n",
    "last_best_i = 0\n",
    "for i, k in enumerate(lacc):\n",
    "    if k > last_best:\n",
    "        print(f\"better i={i}\")\n",
    "        last_best = k\n",
    "        consec = 0\n",
    "        last_best_i = i\n",
    "    else:\n",
    "        consec += 1\n",
    "        if consec >= patience:\n",
    "            print(f\"Violated patience at i={i}\")\n",
    "            consec = 0\n",
    "print(last_best_i)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Final training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "data_folder = \"output/nocontext_schedule1\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'data_folder' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-3-40749bf76785>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      6\u001b[0m \u001b[0mfig\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxes\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mplt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msubplots\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mTs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfigsize\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m16\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      7\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mT\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0max\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mzip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mTs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxes\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 8\u001b[0;31m     \u001b[0mlloss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mos\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpath\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mjoin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata_folder\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34mf\"lloss_{T}.pt\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      9\u001b[0m     \u001b[0mlacc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mos\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpath\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mjoin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata_folder\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34mf\"lacc_{T}.pt\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     10\u001b[0m     \u001b[0mli\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mos\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpath\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mjoin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata_folder\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34mf\"li_{T}.pt\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'data_folder' is not defined"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x1152 with 8 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Accuracy over time graph\n",
    "import torch\n",
    "import os\n",
    "\n",
    "Ts = list(range(2, 10))\n",
    "fig, axes = plt.subplots(len(Ts), 1, figsize=(4, 16))\n",
    "for T, ax in zip(Ts, axes):\n",
    "    lloss = torch.load(os.path.join(data_folder, f\"lloss_{T}.pt\"))\n",
    "    lacc = torch.load(os.path.join(data_folder, f\"lacc_{T}.pt\"))\n",
    "    li = torch.load(os.path.join(data_folder, f\"li_{T}.pt\"))\n",
    "    \n",
    "    ax.set_ylabel(f\"Batch={T+1}\")\n",
    "    line_loss, = ax.plot(li, lloss)\n",
    "    line_acc, = ax.plot(li, lacc)\n",
    "    ax.hlines(1.0, 0, max(li), linestyles=\"dashed\")\n",
    "    \n",
    "ax.set_xlabel(\"n samples seen\")\n",
    "    \n",
    "fig.legend([line_loss, line_acc], [\"Training Loss\", \"Testing Accuracy\"], loc=\"upper right\")\n",
    "\n",
    "# [Context is sequences ] [T=4]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7feb212da208>"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Final accuracies\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import torch\n",
    "import os\n",
    "\n",
    "vergara_eyeballed = [1.0, 0.74, 0.88, 0.93, 0.95, 0.70, 0.70, 0.92, 0.75, 0.65]\n",
    "\n",
    "Ts = list(range(2, 10))\n",
    "batches = [t+1 for t in Ts]\n",
    "accs = []\n",
    "for T, ax in zip(Ts, axes):\n",
    "    acc = torch.load(os.path.join(data_folder, f\"acc_{T}.pt\"))\n",
    "    accs.append(acc)\n",
    "    \n",
    "plt.plot(batches, accs, label=\"Choose-k neural network\")\n",
    "plt.plot(batches, vergara_eyeballed[2:], label=\"Weighted SVM ensemble\")\n",
    "\n",
    "plt.title(\"Generalization performance\")\n",
    "plt.xlabel(\"Batch\")\n",
    "plt.ylabel(\"Final testing accuracy\")\n",
    "plt.legend()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Adjustable accuracy plots"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# Default Parameters\n",
    "show_legend = True\n",
    "draw_errors = True\n",
    "draw_plot = False\n",
    "show_scatter = False\n",
    "draw_axis = False\n",
    "show_svm = False\n",
    "use_offset = False\n",
    "show_vergara = False\n",
    "save_figure = False\n",
    "\n",
    "\n",
    "#data_folders = ([f\"output/backprop{n}\" for n in range(10)],\n",
    "#                [f\"output/backprop_dropout{n}\" for n in range(10)])\n",
    "#dataset_names = (\"Backprop\", \"Backprop+Dropout\")\n",
    "\n",
    "# data_folders = ([f\"output/backprop{n}\" for n in range(10)],\n",
    "#                 [f\"output/evolve{n}\" for n in range(10)],\n",
    "#                 [f\"output/backprop_ensemble{n}\" for n in range(10)])\n",
    "# dataset_names = (\"With Context\", \"Evolve With Context\", \"Ensemble\")\n",
    "\n",
    "# data_folders = ([f\"output/backprop{n}\" for n in range(10)],\n",
    "#                 [f\"output/evolve{n}\" for n in range(10)],\n",
    "#                 [f\"output/evolve_uber{n}\" for n in range(10)])\n",
    "# dataset_names = (\"Backprop\", \"Evolve (old)\", \"Evolve (new)\")\n",
    "\n",
    "# data_folders = ([f\"output/backprop{n}\" for n in range(10)],)\n",
    "# dataset_names = (\"Context Model\",)\n",
    "\n",
    "\n",
    "# data_folders = ([f\"output/ensemble_long{n}\" for n in range(3)],\n",
    "#                 [f\"output/nocontext_long{n}\" for n in range(3)],\n",
    "#                 [f\"/media/jamie/PATRIOT/sensor-drift_bak/output/backprop{n}\" for n in range(10)],\n",
    "#                 [f\"/media/jamie/PATRIOT/sensor-drift_bak/output/backprop_context_earlystop_share{n}\" for n in range(3)],\n",
    "# )\n",
    "# dataset_names = (\"Ensemble\",\n",
    "#                 \"NoContext\",\n",
    "#                 \"Context\",\n",
    "#                 \"ContextShare\"\n",
    "#     )\n",
    "\n",
    "\n",
    "# FIGURE 2B parametrrs\n",
    "# data_folders = ([f\"output/nocontext_long{n}\" for n in range(3)],)\n",
    "# dataset_names = (\"Context Model\",)\n",
    "# figure_path = \"writeup/figure_sources/feedforward_accuracy.png\"\n",
    "\n",
    "# FIGURE 3 parametrrs\n",
    "# data_folders = ([f\"output/ensemble_long{n}\" for n in range(3)],)\n",
    "# dataset_names = (\"Ensemble Model\",)\n",
    "# figure_path = \"writeup/figure_sources/ensemble_accuracy.png\"\n",
    "# show_svm = True\n",
    "\n",
    "# FIGURE 4B parametrrs\n",
    "# data_folders = ([f\"output/context_long{n}\" for n in range(3)],)\n",
    "# dataset_names = (\"Context\",)\n",
    "# figure_path = \"writeup/figure_sources/context_accuracy.png\"\n",
    "\n",
    "# FIGURE 4C parametrrs\n",
    "# data_folders = ([f\"output/context_share_long{n}\" for n in range(3)],)\n",
    "# dataset_names = (\"Context Share\")\n",
    "# figure_path = \"writeup/figure_sources/context_share_accuracy.png\"\n",
    "\n",
    "# \n",
    "# n_trial = 5\n",
    "# data_folders = (\n",
    "#     [f\"output/nocontext_medium{n}\" for n in range(n_trial)],\n",
    "#     [f\"output/context_medium{n}\" for n in range(n_trial)]\n",
    "# )\n",
    "# dataset_names = [\n",
    "#     \"NoContext\",\n",
    "#     \"Context\",\n",
    "# ]\n",
    "\n",
    "\n",
    "n_trial = 30\n",
    "subselection = None #[2, -1]\n",
    "data_folders = ([f\"output/ensemble_harddecay{n}\" for n in range(n_trial)],\n",
    "                [f\"output/nocontext_harddecay{n}\" for n in range(n_trial)],\n",
    "                [f\"output/context_harddecay_k1{n}\" for n in range(n_trial)],\n",
    "                [f\"output/context_lstm_harddecay{n}\" for n in range(n_trial)],\n",
    "                [f\"output/context_harddecay_relu{n}\" for n in range(n_trial)],\n",
    "                [f\"output/nocontext_big_short_harddecay{n}\" for n in range(n_trial)]\n",
    ")\n",
    "# data_folders = ([f\"output/ensemble_short_harddecay{n}\" for n in range(n_trial)],\n",
    "#                 [f\"output/nocontext_short_harddecay{n}\" for n in range(n_trial)],\n",
    "#                 [f\"output/context_short_harddecay{n}\" for n in range(n_trial)],\n",
    "#                 [f\"output/context_lstm_short_harddecay{n}\" for n in range(n_trial)],\n",
    "#                 [f\"output/context_short_harddecay_relu{n}\" for n in range(n_trial)],\n",
    "# )\n",
    "\n",
    "dataset_names = [\"Feedforward NN Ensemble\",\n",
    "                \"Feedforward NN\",\n",
    "                \"Feedforward+Context NN\",\n",
    "                \"LSTM\",\n",
    "                \"Context relu\",\n",
    "                 \"Feedforward+Context NN Big\"\n",
    "    ]\n",
    "if subselection is not None:\n",
    "    data_folders = [data_folders[idx] for idx in subselection]\n",
    "    dataset_names = [dataset_names[idx] for idx in subselection]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "# Final accuracies\n",
    "\n",
    "all_points = []\n",
    "\n",
    "Ts = list(range(2, 10))\n",
    "batches = [t+1 for t in Ts]\n",
    "\n",
    "# For each method, for each batch, for each n, there is an accuracy\n",
    "df = pd.DataFrame(columns=[\"method\", \"batch\", \"n\", \"accuracy\"])\n",
    "for method_i, method_folders in enumerate(data_folders):\n",
    "    for T, batch in zip(Ts, batches):\n",
    "        for n, n_folder in enumerate(method_folders):\n",
    "            try:\n",
    "                acc = torch.load(os.path.join(n_folder, f\"acc_{T}.pt\"))\n",
    "            except FileNotFoundError:\n",
    "                print(\"not found\", dataset_names[method_i], \"T=\", T)\n",
    "            df = df.append({\n",
    "                \"method\": method_i,\n",
    "                \"batch\": batch,\n",
    "                \"n\": n,\n",
    "                \"accuracy\": acc\n",
    "            }, ignore_index=True)\n",
    "if show_svm:\n",
    "    method_i += 1\n",
    "    print(\"method_i\", method_i)\n",
    "    if \"SVM Ensemble\" not in dataset_names:\n",
    "        dataset_names.append(\"SVM Ensemble\")\n",
    "    dirpath = \"svm_ensemble_results/setting2\"\n",
    "    for trial in range(n_trial):\n",
    "        try:\n",
    "            with open(os.path.join(dirpath, f\"accuracies_trial{trial}.pkl\"), \"rb\") as f:\n",
    "                accs = pickle.load(f)\n",
    "        except FileNotFoundError:\n",
    "            print(\"not found\", dataset_names[method_i], \"T=\", T)\n",
    "        accs = accs[-len(batches):]\n",
    "        for i, batch in enumerate(batches):\n",
    "            acc = accs[i]\n",
    "            df = df.append({\n",
    "                \"method\": method_i,\n",
    "                \"batch\": batch,\n",
    "                \"n\": trial,\n",
    "                \"accuracy\": acc\n",
    "            }, ignore_index=True)\n",
    "\n",
    "# For each method, for each batch, calculate the mean and 95% confidence interval\n",
    "df2 = pd.DataFrame(columns=[\"method\", \"batch\", \"mu\", \"err\"])\n",
    "for method_index, method_name in enumerate(dataset_names):\n",
    "    for T, batch in zip(Ts, batches):\n",
    "        data = df[(df.method==method_index) & (df.batch==batch)].accuracy\n",
    "        try:\n",
    "            err = 1.96 * data.std() / math.sqrt(data.count())\n",
    "        except ZeroDivisionError:\n",
    "            print(method_name, data, \"T=\", T, data, method_index)\n",
    "            continue\n",
    "        df2 = df2.append({\n",
    "            \"method\": method_index,\n",
    "            \"batch\": batch,\n",
    "            \"mu\": data.mean().item(),\n",
    "            \"err\": err.item()\n",
    "        }, ignore_index=True)\n",
    "        \n",
    "if use_offset:\n",
    "    OFFSET = 0.15\n",
    "else:\n",
    "    OFFSET = 0.0\n",
    "\n",
    "colors = (\"black\", \"b\", \"r\", \"y\", \"c\", \"m\", \"k\")\n",
    "markers = (\"$\\u25EF$\", \"$\\u25EF$\", \"v\", \"v\", \"*\", \"*\")\n",
    "\n",
    "fig, ax = plt.subplots(1, 1)\n",
    "\n",
    "for i in sorted(df2.method.unique()):\n",
    "    i = int(i)\n",
    "    off = OFFSET if draw_errors or draw_plot else 0.0\n",
    "    m = df[df.method == i]\n",
    "    \n",
    "    # Scatter plot the accuracies\n",
    "    if show_scatter:\n",
    "        ax.scatter(\n",
    "            x=m.batch-off,\n",
    "            y=m.accuracy,\n",
    "            marker=markers[i],\n",
    "            s=100,\n",
    "            label=dataset_names[i],\n",
    "            c=colors[i])\n",
    "    \n",
    "    # Error bars\n",
    "    if draw_errors:\n",
    "        ax.errorbar(\n",
    "            x=df2[df2.method == i].batch,\n",
    "            y=df2[df2.method == i].mu,\n",
    "            yerr=df2[df2.method == i].err,\n",
    "            c=colors[i],\n",
    "            label=dataset_names[i],\n",
    "            capsize=10)\n",
    "    \n",
    "    # Line\n",
    "    if draw_plot:\n",
    "        ax.plot(\n",
    "            df2[df2.method == i].batch,\n",
    "            df2[df2.method == i].mu,\n",
    "            c=colors[i],\n",
    "            markersize=10)\n",
    "\n",
    "if show_vergara:\n",
    "    vergara_eyeballed = [1.0, 0.74, 0.88, 0.93, 0.95, 0.70, 0.70, 0.92, 0.75, 0.65]\n",
    "    ax.plot(batches, vergara_eyeballed[2:],\n",
    "            linestyle='--', c=\"g\", marker=\"$\\u25EF$\", label=\"Weighted SVM ensemble\")\n",
    "\n",
    "ax.set_ylim([0.0, 1.1])\n",
    "if draw_axis:\n",
    "    plt.xlabel(\"Batch\")\n",
    "    plt.ylabel(\"Test Accuracy\")\n",
    "    \n",
    "# Legend\n",
    "if show_legend:       \n",
    "    plt.legend()\n",
    "\n",
    "ax.set_xticks(np.arange(3, 11))\n",
    "    \n",
    "ax.set_ylim([0.7, 1.0])\n",
    "ax.set_yticks(np.arange(0.7, 1.0, 0.1))\n",
    "ax.grid(axis='y', color='#F7BFBE')\n",
    "\n",
    "plt.show()\n",
    "if save_figure:\n",
    "    fig.savefig(figure_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Results Figure"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "n_trial = 30\n",
    "data_folders = (\n",
    "    [f\"output/nocontext_harddecay{n}\" for n in range(n_trial)],\n",
    "    [f\"output/context_harddecay_k1{n}\" for n in range(n_trial)],\n",
    "    [f\"output/ensemble_harddecay{n}\" for n in range(n_trial)],            \n",
    "    [f\"output/context_lstm_harddecay{n}\" for n in range(n_trial)],\n",
    ")\n",
    "dataset_names = [\n",
    "    \"Feedforward NN\",\n",
    "    \"Feedforward+Context NN\",\n",
    "    \"Feedforward NN Ensemble\",\n",
    "    \"LSTM\",\n",
    "]\n",
    "# Default Parameters\n",
    "show_legend = True\n",
    "draw_errors = True\n",
    "draw_plot = False\n",
    "show_scatter = False\n",
    "draw_axis = False\n",
    "show_svm = True\n",
    "use_offset = False\n",
    "show_vergara = False\n",
    "save_figure = False\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "method_i 3\n",
      "SVM Ensemble Series([], Name: accuracy, dtype: float64) T= 2 Series([], Name: accuracy, dtype: float64) 4\n",
      "SVM Ensemble Series([], Name: accuracy, dtype: float64) T= 3 Series([], Name: accuracy, dtype: float64) 4\n",
      "SVM Ensemble Series([], Name: accuracy, dtype: float64) T= 4 Series([], Name: accuracy, dtype: float64) 4\n",
      "SVM Ensemble Series([], Name: accuracy, dtype: float64) T= 5 Series([], Name: accuracy, dtype: float64) 4\n",
      "SVM Ensemble Series([], Name: accuracy, dtype: float64) T= 6 Series([], Name: accuracy, dtype: float64) 4\n",
      "SVM Ensemble Series([], Name: accuracy, dtype: float64) T= 7 Series([], Name: accuracy, dtype: float64) 4\n",
      "SVM Ensemble Series([], Name: accuracy, dtype: float64) T= 8 Series([], Name: accuracy, dtype: float64) 4\n",
      "SVM Ensemble Series([], Name: accuracy, dtype: float64) T= 9 Series([], Name: accuracy, dtype: float64) 4\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 507.6x216 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "# Final accuracies\n",
    "\n",
    "all_points = []\n",
    "\n",
    "Ts = list(range(2, 10))\n",
    "batches = [t+1 for t in Ts]\n",
    "\n",
    "# For each method, for each batch, for each n, there is an accuracy\n",
    "df = pd.DataFrame(columns=[\"method\", \"batch\", \"n\", \"accuracy\"])\n",
    "for method_i, method_folders in enumerate(data_folders):\n",
    "    for T, batch in zip(Ts, batches):\n",
    "        for n, n_folder in enumerate(method_folders):\n",
    "            try:\n",
    "                acc = torch.load(os.path.join(n_folder, f\"acc_{T}.pt\"))\n",
    "            except FileNotFoundError:\n",
    "                print(\"not found\", dataset_names[method_i], \"T=\", T)\n",
    "            df = df.append({\n",
    "                \"method\": method_i,\n",
    "                \"batch\": batch,\n",
    "                \"n\": n,\n",
    "                \"accuracy\": acc\n",
    "            }, ignore_index=True)\n",
    "if show_svm:\n",
    "    method_i += 1\n",
    "    print(\"method_i\", method_i)\n",
    "    if \"SVM Ensemble\" not in dataset_names:\n",
    "        dataset_names.append(\"SVM Ensemble\")\n",
    "    dirpath = \"svm_ensemble_results/setting2\"\n",
    "    for trial in range(n_trial):\n",
    "        try:\n",
    "            with open(os.path.join(dirpath, f\"accuracies_trial{trial}.pkl\"), \"rb\") as f:\n",
    "                accs = pickle.load(f)\n",
    "        except FileNotFoundError:\n",
    "            print(\"not found\", dataset_names[method_i], \"T=\", T)\n",
    "        accs = accs[-len(batches):]\n",
    "        for i, batch in enumerate(batches):\n",
    "            acc = accs[i]\n",
    "            df = df.append({\n",
    "                \"method\": method_i,\n",
    "                \"batch\": batch,\n",
    "                \"n\": trial,\n",
    "                \"accuracy\": acc\n",
    "            }, ignore_index=True)\n",
    "\n",
    "# For each method, for each batch, calculate the mean and 95% confidence interval\n",
    "df2 = pd.DataFrame(columns=[\"method\", \"batch\", \"mu\", \"err\"])\n",
    "for method_index, method_name in enumerate(dataset_names):\n",
    "    for T, batch in zip(Ts, batches):\n",
    "        data = df[(df.method==method_index) & (df.batch==batch)].accuracy\n",
    "        try:\n",
    "            err = 1.96 * data.std() / math.sqrt(data.count())\n",
    "        except ZeroDivisionError:\n",
    "            print(method_name, data, \"T=\", T, data, method_index)\n",
    "            continue\n",
    "        df2 = df2.append({\n",
    "            \"method\": method_index,\n",
    "            \"batch\": batch,\n",
    "            \"mu\": data.mean().item(),\n",
    "            \"err\": err.item()\n",
    "        }, ignore_index=True)\n",
    "        \n",
    "if use_offset:\n",
    "    OFFSET = 0.15\n",
    "else:\n",
    "    OFFSET = 0.0\n",
    "\n",
    "colors = (\"#e41a1c\", \"#377eb8\", \"#4daf4a\", \"#984ea3\", \"#ff7f00\")\n",
    "markers = (\"$\\u25EF$\", \"$\\u25EF$\", \"v\", \"v\", \"*\", \"*\")\n",
    "\n",
    "fig, axes = plt.subplots(1, 2, figsize=(7.05, 3), sharey=True)\n",
    "highlights = [[0, 1], [2, 3]]\n",
    "for ax, highlight in zip(axes, highlights):\n",
    "\n",
    "    for i in sorted(df2.method.unique()):\n",
    "        i = int(i)\n",
    "        off = OFFSET if draw_errors or draw_plot else 0.0\n",
    "        m = df[df.method == i]\n",
    "\n",
    "        # Scatter plot the accuracies\n",
    "        if show_scatter:\n",
    "            ax.scatter(\n",
    "                x=m.batch-off,\n",
    "                y=m.accuracy,\n",
    "                marker=markers[i],\n",
    "                s=100,\n",
    "                label=dataset_names[i],\n",
    "                c=colors[i])\n",
    "\n",
    "        # Error bars\n",
    "        if i in highlight:\n",
    "            alpha = 1.0\n",
    "        else:\n",
    "            alpha = 0.2\n",
    "        if draw_errors:\n",
    "            ax.errorbar(\n",
    "                x=df2[df2.method == i].batch,\n",
    "                y=df2[df2.method == i].mu,\n",
    "                yerr=df2[df2.method == i].err,\n",
    "                c=colors[i],\n",
    "                label=dataset_names[i],\n",
    "                capsize=10,\n",
    "                linewidth=1,\n",
    "                elinewidth=1,\n",
    "                capthick=1,\n",
    "                alpha=alpha)\n",
    "\n",
    "        # Line\n",
    "        if draw_plot:\n",
    "            ax.plot(\n",
    "                df2[df2.method == i].batch,\n",
    "                df2[df2.method == i].mu,\n",
    "                c=colors[i],\n",
    "                markersize=10)\n",
    "\n",
    "if show_vergara:\n",
    "    vergara_eyeballed = [1.0, 0.74, 0.88, 0.93, 0.95, 0.70, 0.70, 0.92, 0.75, 0.65]\n",
    "    ax.plot(batches, vergara_eyeballed[2:],\n",
    "            linestyle='--', c=\"g\", marker=\"$\\u25EF$\", label=\"Weighted SVM ensemble\")\n",
    "\n",
    "ax.set_ylim([0.0, 1.1])\n",
    "if draw_axis:\n",
    "    plt.xlabel(\"Batch\")\n",
    "    plt.ylabel(\"Test Accuracy\")\n",
    "    \n",
    "# Legend\n",
    "import matplotlib.patches as patches\n",
    "legend_rectangles = []\n",
    "legend_lables = []\n",
    "for method_i in sorted(df2.method.unique()):\n",
    "    method_i = int(method_i)\n",
    "    legend_lables.append(dataset_names[method_i])\n",
    "    rect = patches.Rectangle((0, 0), 1, 1, facecolor=colors[method_i])\n",
    "    legend_rectangles.append(rect)\n",
    "axes[0].legend(legend_rectangles[:2], legend_lables[:2], loc=(0.02, 0.02))\n",
    "axes[1].legend(legend_rectangles[2:], legend_lables[2:], loc=(0.02, 0.02))\n",
    "\n",
    "# Axes and ticks\n",
    "axes[0].set_xticks(np.arange(3, 11))\n",
    "axes[1].set_xticks(np.arange(3, 11))\n",
    "    \n",
    "ax.set_ylim([0.4, 1.0])\n",
    "ax.set_yticks(np.arange(0.4, 1.05, 0.1))\n",
    "axes[0].set_ylabel(\"Accuracy\")\n",
    "axes[0].grid(axis='y', color='#bbbbbb')\n",
    "axes[1].grid(axis='y', color='#bbbbbb')\n",
    "axes[0].set_xlabel(\"Batch\")\n",
    "axes[1].set_xlabel(\"Batch\")\n",
    "\n",
    "plt.show()\n",
    "figure_path = \"writeup/figure_sources/fig_3.svg\"\n",
    "fig.savefig(figure_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'0.23.4'"
      ]
     },
     "execution_count": 88,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Statistical tests\n",
    "\n",
    "These require the \"df\" and \"df2\" variables to be present from the previous cells."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "################################################################################\n",
      "### Second test\n",
      "################################################################################\n",
      "################################################################################\n",
      "### Significance test for context, no context\n",
      "################################################################################\n",
      "ANOVA:\n",
      "           Source     SS   DF     MS          F         p-unc       np2\n",
      "0          method  0.021    1  0.021   5.294861  2.183088e-02  0.011283\n",
      "1           batch  2.334    7  0.333  84.069429  1.781386e-78  0.559139\n",
      "2  method * batch  0.042    7  0.006   1.529066  1.551741e-01  0.022548\n",
      "3        Residual  1.840  464  0.004        NaN           NaN       NaN\n",
      "Pairwise TTests:\n",
      "   Contrast    A    B  Paired  Parametric      T     dof       Tail     p-unc  \\\n",
      "0    method  0.0  1.0   False        True -2.922  464.78  two-sided  0.003644   \n",
      "1    method  0.0  2.0   False        True -4.443  442.42  two-sided  0.000011   \n",
      "2    method  0.0  3.0   False        True -3.276  438.39  two-sided  0.001138   \n",
      "3    method  0.0  4.0   False        True -4.644  454.88  two-sided  0.000004   \n",
      "4    method  0.0  5.0   False        True -3.613  461.17  two-sided  0.000336   \n",
      "5    method  1.0  2.0   False        True -1.529  471.13  two-sided  0.126912   \n",
      "6    method  1.0  3.0   False        True -0.214  468.94  two-sided  0.830592   \n",
      "7    method  1.0  4.0   False        True -1.820  476.34  two-sided  0.069352   \n",
      "8    method  1.0  5.0   False        True -0.728  477.74  two-sided  0.466948   \n",
      "9    method  2.0  3.0   False        True  1.417  477.84  two-sided  0.157073   \n",
      "10   method  2.0  4.0   False        True -0.355  476.16  two-sided  0.723015   \n",
      "11   method  2.0  5.0   False        True  0.777  473.47  two-sided  0.437516   \n",
      "12   method  3.0  4.0   False        True -1.729  474.92  two-sided  0.084386   \n",
      "13   method  3.0  5.0   False        True -0.561  471.63  two-sided  0.574822   \n",
      "14   method  4.0  5.0   False        True  1.093  477.38  two-sided  0.275076   \n",
      "\n",
      "        BF10  hedges  \n",
      "0      6.218  -0.266  \n",
      "1   1252.874  -0.405  \n",
      "2       17.7  -0.299  \n",
      "3   2947.857  -0.423  \n",
      "4     53.384  -0.329  \n",
      "5      0.315  -0.139  \n",
      "6      0.104  -0.020  \n",
      "7      0.504  -0.166  \n",
      "8      0.131  -0.066  \n",
      "9      0.268   0.129  \n",
      "10     0.108  -0.032  \n",
      "11     0.136   0.071  \n",
      "12     0.432  -0.158  \n",
      "13     0.118  -0.051  \n",
      "14     0.181   0.100  \n",
      "################################################################################\n",
      "### Significance test for context, LSTM\n",
      "################################################################################\n",
      "ANOVA:\n",
      "           Source     SS   DF     MS          F         p-unc       np2\n",
      "0          method  0.015    1  0.015   3.617390  5.779753e-02  0.007736\n",
      "1           batch  1.666    7  0.238  57.395916  5.330224e-59  0.464062\n",
      "2  method * batch  0.051    7  0.007   1.747027  9.618675e-02  0.025679\n",
      "3        Residual  1.924  464  0.004        NaN           NaN       NaN\n",
      "################################################################################\n",
      "### Significance test for context, big context\n",
      "################################################################################\n",
      "ANOVA:\n",
      "           Source     SS   DF     MS          F         p-unc       np2\n",
      "0          method  0.005    1  0.005   1.257770  2.626536e-01  0.002703\n",
      "1           batch  2.233    7  0.319  80.245739  6.665702e-76  0.547635\n",
      "2  method * batch  0.033    7  0.005   1.179373  3.130049e-01  0.017481\n",
      "3        Residual  1.845  464  0.004        NaN           NaN       NaN\n",
      "################################################################################\n",
      "### Significance test for SVM ensemble, NN ensemble\n",
      "################################################################################\n"
     ]
    },
    {
     "ename": "ValueError",
     "evalue": "'SVM Ensemble' is not in list",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-31-f3261822ebcc>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m    113\u001b[0m \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"#\"\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0;36m80\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    114\u001b[0m \u001b[0mid_1\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mdataset_names\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mindex\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"Feedforward NN Ensemble\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 115\u001b[0;31m \u001b[0mid_2\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mdataset_names\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mindex\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"SVM Ensemble\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    116\u001b[0m \u001b[0mdf_context\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mdf\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmethod\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0mid_1\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m|\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mdf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmethod\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0mid_2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    117\u001b[0m aov = pg.anova(\n",
      "\u001b[0;31mValueError\u001b[0m: 'SVM Ensemble' is not in list"
     ]
    }
   ],
   "source": [
    "import pingouin as pg\n",
    "\n",
    "# First test: For each batch, run an ANOVA for methods considered\n",
    "# This will be used to selectively bold results in the table\n",
    "if False:\n",
    "    print(\"#\" * 80)\n",
    "    print(\"### ANOVA, pairwise t-tests for each batch\")\n",
    "    print(\"#\" * 80)\n",
    "    for batch in range(3, 11):\n",
    "        df_batch = df[df.batch == batch]\n",
    "        print()\n",
    "        print(f\"### batch {batch}:\")\n",
    "        aov = pg.welch_anova(\n",
    "            dv='accuracy',\n",
    "            between='method',\n",
    "            data=df_batch\n",
    "        )\n",
    "        print(\"ANOVA:\")\n",
    "        print(aov)\n",
    "\n",
    "        pgs = pg.pairwise_ttests(\n",
    "            data=df_batch,\n",
    "            dv='accuracy',\n",
    "            between='method',\n",
    "            correction=True\n",
    "        )  \n",
    "        print(\"Pairwise TTests:\")\n",
    "        print(pgs)\n",
    "    print()\n",
    "    \n",
    "# Second test: Run an ANOVA for the grand means\n",
    "print(\"#\" * 80)\n",
    "print(\"### Second test\")\n",
    "print(\"#\" * 80)\n",
    "\n",
    "# Third test: Run an ANOVA for the grand means, blocked by batch\n",
    "print(\"#\" * 80)\n",
    "print(\"### Significance test for context, no context\")\n",
    "print(\"#\" * 80)\n",
    "context_index = dataset_names.index(\"Feedforward+Context NN\")\n",
    "nocontext_index = dataset_names.index(\"Feedforward NN\")\n",
    "df_context = df[(df.method == context_index) | (df.method == nocontext_index)]\n",
    "aov = pg.anova(\n",
    "    dv='accuracy',\n",
    "    between=['method', 'batch'],\n",
    "    data=df_context\n",
    ")\n",
    "print(\"ANOVA:\")\n",
    "print(aov)\n",
    "\n",
    "if False:\n",
    "    print(\"#\" * 80)\n",
    "    print(\"### Significance test for context, no context, by batch\")\n",
    "    print(\"#\" * 80)\n",
    "    for batch in range(3, 11):\n",
    "        print()\n",
    "        print(f\"### batch {batch}:\")\n",
    "        df_batch = df_context[df.batch == batch]\n",
    "        pgs = pg.pairwise_ttests(\n",
    "            data=df_batch,\n",
    "            dv='accuracy',\n",
    "            between='method',\n",
    "            correction=True\n",
    "        )  \n",
    "        print(\"Pairwise TTests:\")\n",
    "        print(pgs)\n",
    "    print()\n",
    "\n",
    "    \n",
    "pgs = pg.pairwise_ttests(\n",
    "    data=df,\n",
    "    dv='accuracy',\n",
    "    between=['method'],\n",
    "    correction=True\n",
    ")  \n",
    "print(\"Pairwise TTests:\")\n",
    "print(pgs)\n",
    "\n",
    "# Did LSTM outperform RNN?\n",
    "print(\"#\" * 80)\n",
    "print(\"### Significance test for context, LSTM\")\n",
    "print(\"#\" * 80)\n",
    "context_index = dataset_names.index(\"Feedforward+Context NN\")\n",
    "lstm_index = dataset_names.index(\"LSTM\")\n",
    "df_context = df[(df.method == context_index) | (df.method == lstm_index)]\n",
    "aov = pg.anova(\n",
    "    dv='accuracy',\n",
    "    between=['method', 'batch'],\n",
    "    data=df_context\n",
    ")\n",
    "print(\"ANOVA:\")\n",
    "print(aov)\n",
    "\n",
    "\n",
    "# Did the large RNN outperform the other one?\n",
    "print(\"#\" * 80)\n",
    "print(\"### Significance test for context, big context\")\n",
    "print(\"#\" * 80)\n",
    "context_index = dataset_names.index(\"Feedforward+Context NN\")\n",
    "large_context_index = dataset_names.index(\"Feedforward+Context NN Big\")\n",
    "df_context = df[(df.method == context_index) | (df.method == large_context_index)]\n",
    "aov = pg.anova(\n",
    "    dv='accuracy',\n",
    "    between=['method', 'batch'],\n",
    "    data=df_context\n",
    ")\n",
    "print(\"ANOVA:\")\n",
    "print(aov)\n",
    "\n",
    "\n",
    "print(\"#\" * 80)\n",
    "print(\"### Significance test for SVM ensemble, NN ensemble\")\n",
    "print(\"#\" * 80)\n",
    "id_1 = dataset_names.index(\"Feedforward NN Ensemble\")\n",
    "id_2 = dataset_names.index(\"SVM Ensemble\")\n",
    "df_context = df[(df.method == id_1) | (df.method == id_2)]\n",
    "aov = pg.anova(\n",
    "    dv='accuracy',\n",
    "    between=['method', 'batch'],\n",
    "    data=df_context\n",
    ")\n",
    "print(\"ANOVA:\")\n",
    "print(aov)\n",
    "\n",
    "\n",
    "# Third test: Run an ANOVA for the grand means, blocked by batch\n",
    "print(\"#\" * 80)\n",
    "print(\"### Significance test for context, no context\")\n",
    "print(\"#\" * 80)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Print Table 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\begin{tabular}{llllllllll}\n",
      "\\hline\n",
      " Batch                   & 3              & 4              & 5              & 6              & 7              & 8              & 9              & 10             & $\\mu$          \\\\\n",
      "\\hline\n",
      " Feedforward NN Ensemble & 0.921          & \\textbf{0.904} & 0.979          & 0.903          & 0.777          & 0.679          & 0.864          & 0.693          & 0.840          \\\\\n",
      " Feedforward NN          & 0.881          & 0.875          & 0.974          & 0.959          & 0.792          & 0.839          & 0.896          & 0.737          & 0.869          \\\\\n",
      " Feedforward+Context NN  & 0.882          & 0.869          & 0.975          & 0.947          & \\textbf{0.820} & 0.864          & \\textbf{0.939} & 0.763          & 0.882          \\\\\n",
      " LSTM                    & 0.891          & 0.877          & 0.923          & 0.913          & 0.809          & 0.849          & 0.932          & \\textbf{0.773} & 0.871          \\\\\n",
      " Context relu            & \\textbf{0.924} & 0.878          & 0.949          & 0.955          & 0.786          & \\textbf{0.880} & 0.939          & 0.769          & \\textbf{0.885} \\\\\n",
      " No Context              & 0.864          & 0.885          & \\textbf{0.981} & \\textbf{0.960} & 0.790          & 0.851          & 0.913          & 0.761          & 0.876          \\\\\n",
      "\\hline\n",
      "\\end{tabular}\n",
      "\n",
      "Average accuracy over all batches:\n",
      "Feedforward NN Ensemble: 0.8399901791138339\n",
      "Feedforward NN: 0.8690827237475173\n",
      "Feedforward+Context NN: 0.8821937662788277\n",
      "LSTM: 0.8709034082551901\n",
      "Context relu: 0.8851380506704699\n"
     ]
    }
   ],
   "source": [
    "# Final accuracies\n",
    "\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import torch\n",
    "import os\n",
    "import math\n",
    "from tabulate import tabulate\n",
    "\n",
    "all_points = []\n",
    "\n",
    "Ts = list(range(2, 10))\n",
    "batches = [t+1 for t in Ts]\n",
    "\n",
    "# For each method, for each batch, for each n, there is an accuracy\n",
    "df = pd.DataFrame(columns=[\"method\", \"batch\", \"n\", \"accuracy\"])\n",
    "for method_i, method_folders in enumerate(data_folders):\n",
    "    for T, batch in zip(Ts, batches):\n",
    "        for n, n_folder in enumerate(method_folders):\n",
    "            try:\n",
    "                acc = torch.load(os.path.join(n_folder, f\"acc_{T}.pt\"))\n",
    "            except FileNotFoundError:\n",
    "                print(\"not found\", dataset_names[method_i], \"T=\", T)\n",
    "            df = df.append({\n",
    "                \"method\": method_i,\n",
    "                \"batch\": batch,\n",
    "                \"n\": n,\n",
    "                \"accuracy\": acc\n",
    "            }, ignore_index=True)\n",
    "if show_svm:\n",
    "    method_i += 1\n",
    "    if \"SVM Ensemble\" not in dataset_names:\n",
    "        dataset_names.append(\"SVM Ensemble\")\n",
    "    dirpath = \"svm_ensemble_results/setting2\"\n",
    "    for trial in range(n_trial):\n",
    "        try:\n",
    "            with open(os.path.join(dirpath, f\"accuracies_trial{trial}.pkl\"), \"rb\") as f:\n",
    "                accs = pickle.load(f)\n",
    "        except FileNotFoundError:\n",
    "            print(\"not found\", dataset_names[method_i], \"T=\", T)\n",
    "        accs = accs[-len(batches):]\n",
    "        for i, batch in enumerate(batches):\n",
    "            acc = accs[i]\n",
    "            df = df.append({\n",
    "                \"method\": method_i,\n",
    "                \"batch\": batch,\n",
    "                \"n\": trial,\n",
    "                \"accuracy\": acc\n",
    "            }, ignore_index=True)\n",
    "\n",
    "# For each method, for each batch, calculate the mean and 95% confidence interval\n",
    "df2 = pd.DataFrame(columns=[\"method\", \"batch\", \"mu\", \"err\"])\n",
    "for method_index, method_name in enumerate(dataset_names):\n",
    "    for T, batch in zip(Ts, batches):\n",
    "        data = df[(df.method==method_index) & (df.batch==batch)].accuracy\n",
    "        try:\n",
    "            err = 1.96 * data.std() / math.sqrt(data.count())\n",
    "        except ZeroDivisionError:\n",
    "            print(method_name, data, \"T=\", T)\n",
    "            continue\n",
    "        df2 = df2.append({\n",
    "            \"method\": method_index,\n",
    "            \"batch\": batch,\n",
    "            \"mu\": data.mean().item(),\n",
    "            \"err\": err.item()\n",
    "        }, ignore_index=True)\n",
    "        \n",
    "# Compute a grand total mean and 95% CI\n",
    "for method_index, method_name in enumerate(dataset_names):\n",
    "    data = df[(df.method==method_index) & (df.batch>=3)].accuracy\n",
    "    try:\n",
    "        err = 1.96 * data.std() / math.sqrt(data.count())\n",
    "    except ZeroDivisionError:\n",
    "        print(method_name, data)\n",
    "        continue\n",
    "    df2 = df2.append({\n",
    "        \"method\": method_index,\n",
    "        \"batch\": \"$\\mu\",\n",
    "        \"mu\": data.mean().item(),\n",
    "        \"err\": err.item()\n",
    "    }, ignore_index=True)\n",
    "\n",
    "\n",
    "draw_errors = False\n",
    "headers = list(range(3, 11)) + [\"$\\mu$\"]\n",
    "table = []\n",
    "for i in sorted(df2.method.unique()):\n",
    "    i = int(i)\n",
    "    off = OFFSET if draw_errors else 0.0\n",
    "    means = df2[df2.method == i].mu.tolist()\n",
    "    errs = df2[df2.method == i].err.tolist()\n",
    "#     means.append(np.mean(means))\n",
    "    if draw_errors:\n",
    "        row = [f\"{means[j]:.3f} $\\pm$ {errs[j]:.3f}\" for j in range(len(means))]\n",
    "    else:\n",
    "        row = [f\"{means[j]:.3f}\" for j in range(len(means))]\n",
    "    row.insert(0, dataset_names[i])\n",
    "    table.append(row)\n",
    "headers.insert(0, \"Batch\")\n",
    "\n",
    "\n",
    "# bold the largest value in each column\n",
    "for column_id, _ in enumerate(table[0]):\n",
    "    if column_id == 0:\n",
    "        continue\n",
    "    max_row_id = -1\n",
    "    max_row_val = -np.inf\n",
    "    for row_id, _ in enumerate(table):\n",
    "        val = float(table[row_id][column_id])\n",
    "        if val > max_row_val:\n",
    "            max_row_id = row_id\n",
    "            max_row_val = val\n",
    "    table[max_row_id][column_id] = f\"\\\\textbf{{{max_row_val:.3f}}}\"\n",
    "\n",
    "print(tabulate(table, headers, tablefmt=\"latex_raw\", floatfmt=\".3f\"))\n",
    "\n",
    "\n",
    "# How much improvement did the Context model have over NoContext?\n",
    "print()\n",
    "print(\"Average accuracy over all batches:\")\n",
    "indices = [0, 1, 2, 3, 4]\n",
    "for i in indices:\n",
    "    means = df2[df2.method == i].mu.tolist()\n",
    "    avg = np.mean(means)\n",
    "    print(f\"{dataset_names[i]}: {avg}\")\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Print Table 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of parameters:\n",
      "Context Model: 14315\n",
      "No Context Model: 7575\n"
     ]
    }
   ],
   "source": [
    "from main_backprop_context import ContextModel\n",
    "from main_backprop_nocontext import NoContextModel\n",
    "\n",
    "def count_parameters(model):\n",
    "    return sum(p.numel() for p in model.parameters() if p.requires_grad)\n",
    "\n",
    "print(\"Number of parameters:\")\n",
    "    \n",
    "context_model = ContextModel(k=1)\n",
    "print(f\"Context Model: {count_parameters(context_model)}\")\n",
    "\n",
    "no_context_model = NoContextModel()\n",
    "print(f\"No Context Model: {count_parameters(no_context_model)}\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "NoContext skill_size=96: 14429 parameters\n",
      "NoContext skill_size=95: 14280 parameters\n"
     ]
    }
   ],
   "source": [
    "# Find the size of a No Context model with equal # parameters to the Context Modle\n",
    "n_context_parameters = count_parameters(ContextModel(k=1))\n",
    "\n",
    "n_nocontext_parameters = 0\n",
    "skill_size = 20\n",
    "while n_nocontext_parameters < n_context_parameters:\n",
    "    skill_size += 1\n",
    "    n_nocontext_parameters = count_parameters(NoContextModel(skill_size=skill_size))\n",
    "print(f\"NoContext skill_size={skill_size}: {n_nocontext_parameters} parameters\")\n",
    "skill_size -= 1\n",
    "n_nocontext_parameters = count_parameters(NoContextModel(skill_size=skill_size))\n",
    "print(f\"NoContext skill_size={skill_size}: {n_nocontext_parameters} parameters\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Single-Network Models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# Evaluate the accuracy of the single-network models on all batches\n",
    "\n",
    "from main_backprop_ensemble import NoContextModel, test_network\n",
    "from batches import split_all\n",
    "import pickle\n",
    "import numpy as np\n",
    "\n",
    "n_trials = 30\n",
    "\n",
    "def evaluate_accuracies(T: int, trial: int):\n",
    "    \"\"\"Load the network trained using batch T, and evaluate it on every batch 0..9\n",
    "    :return: Array with 10 entries\"\"\"\n",
    "    data_folder = f\"output/ensemble_harddecay{trial}\"\n",
    "    \n",
    "    net = NoContextModel()\n",
    "    net.load_state_dict(torch.load(os.path.join(data_folder, f\"model_{T}.pt\")))\n",
    "\n",
    "    accuracies = [] \n",
    "    for t in range(10):\n",
    "        acc = test_network(net, t, split_all)\n",
    "        accuracies.append(acc)\n",
    "    accuracies = np.array(accuracies)\n",
    "    \n",
    "    return accuracies\n",
    "\n",
    "def evaluate_accuracies_all_trials(T: int):\n",
    "    \"\"\":return: Accuracies shape (trials, batches)\"\"\"\n",
    "    trials = []\n",
    "    for trial in range(n_trials):\n",
    "        accuracies = evaluate_accuracies(T, trial)\n",
    "        trials.append(accuracies)\n",
    "    return np.stack(trials)\n",
    "\n",
    "all_accuracies = []\n",
    "for batch in range(10):\n",
    "    accuracies = evaluate_accuracies_all_trials(batch) # batches 0-indexed\n",
    "    all_accuracies.append(accuracies)\n",
    "with open(\"output/single_network_accuracies.pkl\", \"wb\") as f:\n",
    "    pickle.dump(all_accuracies, f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 507.6x216 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "with open(\"output/single_network_accuracies.pkl\", \"rb\") as f:\n",
    "    all_accuracies = pickle.load(f)\n",
    "\n",
    "colors = ['#d7191c', '#d98330', '#7bad74', '#2b83ba']\n",
    "    \n",
    "# Left plot\n",
    "selected_batches = [1, 3, 5, 7]\n",
    "fig, axes = plt.subplots(1, 2, figsize=(7.05, 3), sharey=True)\n",
    "ax = axes[0]\n",
    "for i, train_batch in enumerate(selected_batches):\n",
    "    accuracies = all_accuracies[train_batch]\n",
    "    errs = []\n",
    "    means = []\n",
    "    for batch in range(10):\n",
    "        data = accuracies[:, batch]\n",
    "        err = 1.96 * data.std() / math.sqrt(data.shape[0])\n",
    "        mean = data.mean()\n",
    "        errs.append(err)\n",
    "        means.append(mean)\n",
    "        \n",
    "    ax.axvline(train_batch+1, c=colors[i], linestyle=\"--\", linewidth=1)\n",
    "\n",
    "    ax.errorbar(\n",
    "        x=list(range(1, 11)),\n",
    "        y=means,\n",
    "        yerr=errs,\n",
    "        c=colors[i],\n",
    "#         label=dataset_names[i],\n",
    "        capsize=10,\n",
    "        linewidth=1,\n",
    "        elinewidth=1,\n",
    "        capthick=1,\n",
    "    )\n",
    "\n",
    "ax.set_ylabel(\"Accuracy\")\n",
    "ax.set_xlabel(\"Batch\")\n",
    "ax.set_ylim([0.0, 1.0])\n",
    "ax.set_yticks(np.arange(0, 1.1, 0.1))\n",
    "ax.set_xticks(np.arange(1, 11))\n",
    "ax.grid(axis='y', color='#bbbbbb')\n",
    "\n",
    "def flatten_and_index(arr):\n",
    "    flat = []\n",
    "    ind = []\n",
    "    for i, sub in enumerate(arr):\n",
    "        flat += sub\n",
    "        ind += [i] * len(sub)\n",
    "    return ind, flat\n",
    "ax = axes[1]\n",
    "accuracies_by_difference = [[] for _ in range(10)]\n",
    "for train_batch in range(10):\n",
    "    accuracies = all_accuracies[train_batch]\n",
    "    for test_batch in range(10):\n",
    "        data = accuracies[:, test_batch]\n",
    "        difference = abs(train_batch-test_batch)\n",
    "        accuracies_by_difference[difference] += data.tolist()\n",
    "errs = []\n",
    "means = []\n",
    "for difference, accuracies in enumerate(accuracies_by_difference):\n",
    "    err = 1.96 * np.std(accuracies) / math.sqrt(len(accuracies))\n",
    "    mean = np.mean(accuracies)\n",
    "    errs.append(err)\n",
    "    means.append(mean)\n",
    "ax.errorbar(\n",
    "    x=list(range(10)),\n",
    "    y=means,\n",
    "    yerr=errs,\n",
    "    c='black',\n",
    "#         label=dataset_names[i],\n",
    "    capsize=10,\n",
    "    linewidth=1,\n",
    "    elinewidth=1,\n",
    "    capthick=1,\n",
    ")\n",
    "\n",
    "# x, y = flatten_and_index(accuracies_by_difference)\n",
    "# ax.scatter(x, y, alpha=0.2)\n",
    "\n",
    "ax.set_xticks(np.arange(10))\n",
    "ax.grid(axis='y', color='#bbbbbb')\n",
    "ax.set_xlabel(\"Difference of Batches\")\n",
    "\n",
    "fig.savefig(\"writeup/figure_sources/feedforward_centers.svg\")\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}