Evaluation.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from util.metrics import calc_mc_accuracy, calc_mc_set_accuracy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load experiment data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded experiment log with id 1131\n"
     ]
    }
   ],
   "source": [
    "# e_name = 'mnist_gan2_mcc_experiment_2574'\n",
    "# e_name = 'experiment_8944'\n",
    "e_name = 'test'\n",
    "\n",
    "# Import results\n",
    "# with open('experiments/experiment_' + str(eid) + '.json') as f:\n",
    "with open('experiments/' + str(e_name) + '.json') as f:\n",
    "    log = json.loads(f.read())\n",
    "\n",
    "print('Loaded experiment log with id ' + str(log['id']))\n",
    "# print('Performed attacks: {0}'.format(log['attacks']))\n",
    "# print('Repetitions: {0}'.format(len(log[log['attacks'][0]]['results'])))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "attacks = ['legacy_pca_mc_category_attack_1000', 'legacy_pca_mc_category_attack_10000']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "attacks = log['attacks']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Collect and aggregate result values\n",
    "\n",
    "values = dict()\n",
    "\n",
    "for attack in attacks: #log['attacks']:\n",
    "    \n",
    "    values[attack] = dict()\n",
    "    \n",
    "    results = log[attack]['results']\n",
    "    # print(results)\n",
    "    \n",
    "    config_name = log[attack]['base_config']\n",
    "    # config_name = 'pca_mc_category_attack'\n",
    "    \n",
    "    with open('configs/' + str(config_name) + '.json') as f:\n",
    "        attack_config = json.loads(f.read())\n",
    "    with open('configs/' + str(log['model']) + '.json') as f:\n",
    "        model_config = json.loads(f.read())\n",
    "    \n",
    "    # print(attack_config)\n",
    "    # print(model_config)\n",
    "    # print(log[attack]['attack_type'])\n",
    "    \n",
    "    if 'mc_attack' in log[attack]['attack_type'] or 'mc_category_attack' in log[attack]['attack_type']:\n",
    "    \n",
    "        # Heuristics\n",
    "        heuristics = results[0].keys()\n",
    "        \n",
    "        for heuristic in heuristics:\n",
    "            \n",
    "            values[attack][heuristic] = dict()\n",
    "            values[attack][heuristic]['single'] = list()\n",
    "            values[attack][heuristic]['set'] = list()\n",
    "            \n",
    "            # Single MI Accuracy\n",
    "            single_mi_acc = calc_mc_accuracy(heuristic, 'mc_attack_log_acc', results)\n",
    "            values[attack][heuristic]['single'].append(single_mi_acc)\n",
    "            # print(heuristic, single_mi_acc)\n",
    "            \n",
    "            # Set MI Accuracy\n",
    "            set_mi_acc = calc_mc_set_accuracy(heuristic, 'mc_attack_log_acc', results)\n",
    "            values[attack][heuristic]['set'].append(set_mi_acc)\n",
    "            # print(set_mi_acc)\n",
    "    \n",
    "    elif log[attack]['attack_type'] == 'reconstruction_attack':\n",
    "        values[attack]['reconstruction_accuracy_single'] = np.mean([x['reconstruction_accuracy'] for x in results])\n",
    "        values[attack]['reconstruction_accuracy_set'] = (sum([x['successful_set_attack'] for x in results]) / len(results))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'legacy_pca_mc_category_attack_1000': {'median': {'single': [0.53],\n",
       "   'set': [0.8]},\n",
       "  '0.01': {'single': [0.5200000000000001], 'set': [0.6]},\n",
       "  '0.1': {'single': [0.496], 'set': [0.2]},\n",
       "  '1': {'single': [0.534], 'set': [0.7]}},\n",
       " 'legacy_pca_mc_category_attack_10000': {'median': {'single': [0.53],\n",
       "   'set': [0.8]},\n",
       "  '0.01': {'single': [0.48600000000000004], 'set': [0.4]},\n",
       "  '0.1': {'single': [0.5199999999999999], 'set': [0.6]},\n",
       "  '1': {'single': [0.534], 'set': [0.8]}}}"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "values"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plots for Monte Carlo Attacks"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create Plot for different Sample Sizes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "# Show single membership inference results\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "heuristics = results[0].keys()\n",
    "attacks = values.keys()\n",
    "# xlabels = ['10\\u00b3', '10\\u2074', '10\\u2075', '10\\u2076']\n",
    "xlabels = ['10\\u00b3', '10\\u2074']\n",
    "plt.xticks(range(len(xlabels)), xlabels)\n",
    "#plt.figure(figsize=(10, 6))\n",
    "\n",
    "for heuristic in heuristics:\n",
    "    \n",
    "    # Aggregate results\n",
    "    mean_single_accs = [np.array(values[attack][heuristic]['single']).mean() for attack in attacks]\n",
    "    \n",
    "    if heuristic != 'median':\n",
    "        label = '{:3g}% percentile'.format(float(heuristic) * 10)\n",
    "    else:\n",
    "        label = 'Median Heuristic'\n",
    "    \n",
    "    plt.plot(range(len(xlabels)), mean_single_accs, marker='o', label=label)\n",
    "    \n",
    "plt.xlabel('Sample Size')\n",
    "plt.ylabel('Single MI Accuracy')\n",
    "plt.legend(loc='upper right', bbox_to_anchor=(1.01, 1.2), ncol=2);\n",
    "plt.tight_layout()\n",
    "#plt.savefig('figures/mnist_gan_mc_category_single_accs.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "# Show set membership inference results\n",
    "\n",
    "heuristics = results[0].keys()\n",
    "attacks = values.keys()\n",
    "x = ['10\\u00b3', '10\\u2074', '10\\u2075', '10\\u2076']\n",
    "#print(heuristics)\n",
    "#print(attacks)\n",
    "\n",
    "#plt.xlim(100, 1000000)\n",
    "plt.xticks(range(len(x)), x)\n",
    "plt.xlabel('Sample Size')\n",
    "plt.ylabel('Set MI Accuracy')\n",
    "\n",
    "for heuristic in heuristics:\n",
    "    \n",
    "    # Aggregate results\n",
    "    mean_set_accs = [np.array(values[attack][heuristic]['set']).mean() for attack in attacks]\n",
    "    \n",
    "    if heuristic != 'median':\n",
    "        label = '{:3g}% percentile'.format(float(heuristic) * 10)\n",
    "    else:\n",
    "        label = 'Median Heuristic'\n",
    "    \n",
    "    plt.plot(range(len(x)), mean_set_accs, marker='o', label=label)\n",
    "\n",
    "# plt.figure(figsize=(10, 6))\n",
    "plt.legend(loc='upper right', bbox_to_anchor=(1.01, 1.2), ncol=2);\n",
    "plt.tight_layout()\n",
    "#plt.savefig('figures/mnist_gan_mc_category_set_accs.png')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create Boxplot for different Metrics (over multiple Runs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "# Show single membership inference results\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "heuristics = results[0].keys()\n",
    "attacks = values.keys()\n",
    "#xlabels = ['10\\u00b3', '10\\u2074', '10\\u2075', '10\\u2076']\n",
    "#xlabels = [1000, 10000, 100000, 1000000]\n",
    "\n",
    "\n",
    "x = list()\n",
    "\n",
    "for heuristic in heuristics:\n",
    "    x.append([e[heuristic]['mc_attack_log_acc'] for e in log['pca_mc_category_attack_1000']['results']])\n",
    "\n",
    "\n",
    "plt.boxplot(x)\n",
    "plt.ylabel('Accuracy')\n",
    "plt.xlabel('Metric')\n",
    "plt.xticks(range(1, len(heuristics) + 1), heuristics)\n",
    "# plt.show()\n",
    "# plt.savefig('test.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyError",
     "evalue": "'results'",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mKeyError\u001b[0m                                  Traceback (most recent call last)",
      "\u001b[1;32m<ipython-input-66-37d2f63e2087>\u001b[0m in \u001b[0;36m<module>\u001b[1;34m\u001b[0m\n\u001b[0;32m      4\u001b[0m \u001b[1;32mimport\u001b[0m \u001b[0mnumpy\u001b[0m \u001b[1;32mas\u001b[0m \u001b[0mnp\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      5\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m----> 6\u001b[1;33m \u001b[0mheuristics\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mlog\u001b[0m\u001b[1;33m[\u001b[0m\u001b[1;34m'results'\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m[\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mkeys\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m      7\u001b[0m \u001b[0mx\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mlist\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      8\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;31mKeyError\u001b[0m: 'results'"
     ]
    }
   ],
   "source": [
    "# Show single membership inference results\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "heuristics = log['results'][0].keys()\n",
    "x = list()\n",
    "\n",
    "for heuristic in heuristics:\n",
    "    x.append([e[heuristic]['mc_attack_log_acc'] for e in log['results']])\n",
    "\n",
    "plt.figure(figsize=(10, 6))\n",
    "plt.boxplot(x)\n",
    "plt.ylabel('Accuracy')\n",
    "plt.xlabel('Metric')\n",
    "plt.xticks(range(1, len(heuristics) + 1), heuristics)\n",
    "plt.show()\n",
    "\n",
    "# plt.savefig('test.png')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Barplot for Set MI (over multiple Runs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "heuristics = log['results'][0].keys()\n",
    "x = list()\n",
    "\n",
    "for heuristic in heuristics:\n",
    "    x.append(sum([e[heuristic]['successful_set_attack_log'] for e in log['results']]) / len(log['results']))\n",
    "\n",
    "plt.figure(figsize=(10, 6))\n",
    "plt.bar(height=x, x=heuristics)\n",
    "plt.ylabel('Set Accuracy')\n",
    "plt.xlabel('Metric')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plots for Reconstruction Attack"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "x = [int(x.replace('recon_attack_', '')) for x in log['attacks']]\n",
    "y = [v['reconstruction_accuracy_single'] for v in values.values()]\n",
    "plt.xticks(range(len(x)), x)\n",
    "plt.ylim(0.4, 0.8)\n",
    "plt.plot(range(len(x)), y, marker='o')\n",
    "plt.xlabel('No. of Repetitions of Reconstructions')\n",
    "plt.ylabel('Single MI Reconstruction Accuracy')\n",
    "plt.title('Reconstruction Attack on mnist_vae3')\n",
    "# plt.show()\n",
    "plt.savefig('mnist_vae3_recon_attack_single_mi')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "x = [int(x.replace('recon_attack_', '')) for x in log['attacks']]\n",
    "y = [v['reconstruction_accuracy_set'] for v in values.values()]\n",
    "plt.xticks(range(len(x)), x)\n",
    "plt.ylim(0, 1.1)\n",
    "plt.plot(range(len(x)), y, marker='o')\n",
    "plt.xlabel('No. of Repetitions of Reconstructions')\n",
    "plt.ylabel('Set MI Reconstruction Accuracy')\n",
    "plt.title('Reconstruction Attack on mnist_vae3')\n",
    "# plt.show()\n",
    "plt.savefig('mnist_vae3_recon_attack_set_mi')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compare multiple experiments"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded experiment log with id 9382\n",
      "Performed attacks: ['pca_mc_attack_1000', 'pca_mc_attack_10000', 'pca_mc_attack_100000', 'pca_mc_attack_1000000']\n",
      "Repetitions: 10\n"
     ]
    }
   ],
   "source": [
    "experiment_ids = [9382]\n",
    "logs = list()\n",
    "\n",
    "values = dict()\n",
    "\n",
    "for eid in experiment_ids:\n",
    "\n",
    "    # Import results\n",
    "    with open('experiments/experiment_' + str(eid) + '.json') as f:\n",
    "        log = json.loads(f.read())\n",
    "    #logs.append(log)\n",
    "    \n",
    "    print('Loaded experiment log with id ' + str(eid))\n",
    "    print('Performed attacks: {0}'.format(log['attacks']))\n",
    "    print('Repetitions: {0}'.format(len(log[log['attacks'][0]]['results'])))\n",
    "    \n",
    "    attack = list(log.keys())[-1] # does this always work?\n",
    "    \n",
    "    results = log[attack]['results']\n",
    "    config_name = log[attack]['base_config']\n",
    "    \n",
    "    with open('configs/' + str(config_name) + '.json') as f:\n",
    "        attack_config = json.loads(f.read())\n",
    "    with open('configs/' + str(attack_config['model_config']) + '.json') as f:\n",
    "        model_config = json.loads(f.read())\n",
    "    \n",
    "    model_type = model_config['type']\n",
    "    base_attack_name = attack_config['attack_type']\n",
    "    \n",
    "    if base_attack_name not in values.keys():\n",
    "        values[base_attack_name] = dict()\n",
    "    \n",
    "    if 'mc_attack' in log[attack]['attack_type']:\n",
    "        \n",
    "        #Only use median heuristic\n",
    "        heuristic = 'median_perc'\n",
    "            \n",
    "        values[base_attack_name][model_type] = dict()\n",
    "        values[base_attack_name][model_type]['single'] = dict()\n",
    "        values[base_attack_name][model_type]['set'] = dict()\n",
    "\n",
    "        # Single MI Accuracy\n",
    "        single_mi_acc = calc_mc_accuracy(heuristic, '50_perc_mc_attack_log', results)\n",
    "        values[base_attack_name][model_type]['single']['mean'] = single_mi_acc\n",
    "        values[base_attack_name][model_type]['single']['std'] = np.std([x[heuristic]['50_perc_mc_attack_log'] for x in results])\n",
    "        # print(heuristic, single_mi_acc)\n",
    "\n",
    "        # Set MI Accuracy\n",
    "        set_mi_acc = calc_mc_set_accuracy(heuristic, '50_perc_mc_attack_log', results)\n",
    "        values[base_attack_name][model_type]['set']['mean'] = set_mi_acc\n",
    "        accuracies = np.array([x[heuristic]['50_perc_mc_attack_log'] for x in results])\n",
    "        advantages = calc_advantage(accuracies)\n",
    "        probabilities = np.array(list(map(calc_probability, advantages)))\n",
    "        values[base_attack_name][model_type]['set']['std'] = np.std(probabilities)\n",
    "        # print(set_mi_acc)\n",
    "    \n",
    "    else:\n",
    "        # reconstruction attack\n",
    "        pass\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_probability(advantage):\n",
    "    if advantage > 0:\n",
    "        prob = 1\n",
    "    elif advantage == 0:\n",
    "        prob = 0.5\n",
    "    elif advantage < 0:\n",
    "        prob = 0\n",
    "    return prob\n",
    "\n",
    "def calc_advantage(accuracies):\n",
    "    advantages = (accuracies - 0.5) * 2\n",
    "    return advantages"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'pca_mc_attack': {'GAN': {'single': {'mean': 0.49800000000000005,\n",
       "    'std': 0.041182520563948014},\n",
       "   'set': {'mean': 0.55, 'std': 0.4716990566028302}}}}"
      ]
     },
     "execution_count": 107,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "N = 5\n",
    "menMeans = (20, 35, 30, 35, 27)\n",
    "menStd =   (2, 3, 4, 1, 2)\n",
    "\n",
    "ind = np.arange(N)  # the x locations for the groups\n",
    "width = 0.35       # the width of the bars\n",
    "\n",
    "fig = plt.figure()\n",
    "ax = fig.add_subplot(111)\n",
    "rects1 = ax.bar(ind, menMeans, width, color='royalblue', yerr=menStd)\n",
    "\n",
    "womenMeans = (25, 32, 34, 20, 25)\n",
    "womenStd =   (3, 5, 2, 3, 3)\n",
    "rects2 = ax.bar(ind+width, womenMeans, width, color='seagreen', yerr=womenStd)\n",
    "\n",
    "# add some\n",
    "ax.set_ylabel('Scores')\n",
    "ax.set_title('Scores by group and gender')\n",
    "ax.set_xticks(ind + width / 2)\n",
    "ax.set_xticklabels( ('G1', 'G2', 'G3', 'G4', 'G5') )\n",
    "\n",
    "ax.legend( (rects1[0], rects2[0]), ('Men', 'Women') )\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'id': 1270,\n",
       " 'attacks': ['pca_mc_attack'],\n",
       " 'pca_mc_attack_1000': {'attack_type': 'pca_mc_attack',\n",
       "  'results': [{'1': {'50_perc_mc_attack_log': 0.52,\n",
       "     '50_perc_mc_attack_eps': 0.52,\n",
       "     '50_perc_mc_attack_frac': 0.54,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.11352838501691924, 0.11328629051553953],\n",
       "     'set_attack_eps_vals': [1.0018, 0.9982000000000001],\n",
       "     'set_attack_frac_vals': [1.1280467083215517, 1.1241757500702776]},\n",
       "    '0.1': {'50_perc_mc_attack_log': 0.49,\n",
       "     '50_perc_mc_attack_eps': 0.51,\n",
       "     '50_perc_mc_attack_frac': 0.49,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': False,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.007143509850873135, 0.007119893457452385],\n",
       "     'set_attack_eps_vals': [0.10000000000000003, 0.10000000000000003],\n",
       "     'set_attack_frac_vals': [0.10749184922373442, 0.107466227011662]},\n",
       "    '0.01': {'50_perc_mc_attack_log': 0.52,\n",
       "     '50_perc_mc_attack_eps': 0.51,\n",
       "     '50_perc_mc_attack_frac': 0.52,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': False,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.0003510071865144189, 0.00034652785786472214],\n",
       "     'set_attack_eps_vals': [0.01, 0.01],\n",
       "     'set_attack_frac_vals': [0.010357264863064285, 0.01035262297346588]},\n",
       "    'median_perc': {'50_perc_mc_attack_log': 0.52,\n",
       "     '50_perc_mc_attack_eps': 0.52,\n",
       "     '50_perc_mc_attack_frac': 0.52,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [1.0644552825301907e-05, 7.811569845665582e-06],\n",
       "     'set_attack_eps_vals': [0.0052000000000000015, 0.004800000000000001],\n",
       "     'set_attack_frac_vals': [0.005210660964458833, 0.00480782073344274]}}]},\n",
       " 'pca_mc_attack_10000': {'attack_type': 'pca_mc_attack',\n",
       "  'results': [{'1': {'50_perc_mc_attack_log': 0.49,\n",
       "     '50_perc_mc_attack_eps': 0.54,\n",
       "     '50_perc_mc_attack_frac': 0.55,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.13114803413176718, 0.1309231796455954],\n",
       "     'set_attack_eps_vals': [1.00156, 0.99844],\n",
       "     'set_attack_frac_vals': [1.1495140601742946, 1.146138474971862]},\n",
       "    '0.1': {'50_perc_mc_attack_log': 0.52,\n",
       "     '50_perc_mc_attack_eps': 0.53,\n",
       "     '50_perc_mc_attack_frac': 0.53,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.005577018360058325, 0.005553212684088307],\n",
       "     'set_attack_eps_vals': [0.10021999999999999, 0.09978000000000001],\n",
       "     'set_attack_frac_vals': [0.10604776160205423, 0.10558260490728069]},\n",
       "    '0.01': {'50_perc_mc_attack_log': 0.5,\n",
       "     '50_perc_mc_attack_eps': 0.5,\n",
       "     '50_perc_mc_attack_frac': 0.5,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.0002030017005243138, 0.00020076606640459413],\n",
       "     'set_attack_eps_vals': [0.01006, 0.00994],\n",
       "     'set_attack_frac_vals': [0.010266536293076339, 0.010144251071015065]},\n",
       "    'median_perc': {'50_perc_mc_attack_log': 0.52,\n",
       "     '50_perc_mc_attack_eps': 0.52,\n",
       "     '50_perc_mc_attack_frac': 0.52,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [1.081627207619541e-06, 8.708115997043038e-07],\n",
       "     'set_attack_eps_vals': [0.0005200000000000001, 0.0004800000000000001],\n",
       "     'set_attack_frac_vals': [0.0005210833633651186,\n",
       "      0.00048087194710070283]}}]},\n",
       " 'pca_mc_attack_100000': {'attack_type': 'pca_mc_attack',\n",
       "  'results': [{'1': {'50_perc_mc_attack_log': 0.5,\n",
       "     '50_perc_mc_attack_eps': 0.54,\n",
       "     '50_perc_mc_attack_frac': 0.54,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.1300023266833848, 0.12977557972029555],\n",
       "     'set_attack_eps_vals': [1.001227, 0.9987729999999999],\n",
       "     'set_attack_frac_vals': [1.147545815577, 1.1448330005288188]},\n",
       "    '0.1': {'50_perc_mc_attack_log': 0.51,\n",
       "     '50_perc_mc_attack_eps': 0.54,\n",
       "     '50_perc_mc_attack_frac': 0.53,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.005787678830255301, 0.005763432194453321],\n",
       "     'set_attack_eps_vals': [0.100143, 0.099857],\n",
       "     'set_attack_frac_vals': [0.10621581456119333, 0.10590404481433827]},\n",
       "    '0.01': {'50_perc_mc_attack_log': 0.5,\n",
       "     '50_perc_mc_attack_eps': 0.48,\n",
       "     '50_perc_mc_attack_frac': 0.48,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.00026589555636952867, 0.0002631743698429484],\n",
       "     'set_attack_eps_vals': [0.010032000000000001, 0.009968],\n",
       "     'set_attack_frac_vals': [0.010304181532408358, 0.010237384431233811]},\n",
       "    'median_perc': {'50_perc_mc_attack_log': 0.48,\n",
       "     '50_perc_mc_attack_eps': 0.48,\n",
       "     '50_perc_mc_attack_frac': 0.48,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': False,\n",
       "     'successful_set_attack_frac': False,\n",
       "     'set_attack_log_vals': [1.116720284745999e-07, 8.746963765537193e-08],\n",
       "     'set_attack_eps_vals': [4.799999999999999e-05, 5.199999999999999e-05],\n",
       "     'set_attack_frac_vals': [4.8111860200378844e-05,\n",
       "      5.2087583621971386e-05]}}]},\n",
       " 'pca_mc_attack_1000000': {'attack_type': 'pca_mc_attack',\n",
       "  'results': [{'1': {'50_perc_mc_attack_log': 0.5,\n",
       "     '50_perc_mc_attack_eps': 0.54,\n",
       "     '50_perc_mc_attack_frac': 0.54,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.1305854023217908, 0.1303617640305691],\n",
       "     'set_attack_eps_vals': [1.0012031, 0.9987968999999999],\n",
       "     'set_attack_frac_vals': [1.1483084217960422, 1.1456468729050686]},\n",
       "    '0.1': {'50_perc_mc_attack_log': 0.51,\n",
       "     '50_perc_mc_attack_eps': 0.55,\n",
       "     '50_perc_mc_attack_frac': 0.54,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.005805107559290715, 0.005781065451351226],\n",
       "     'set_attack_eps_vals': [0.10016390000000001, 0.0998361],\n",
       "     'set_attack_frac_vals': [0.10625216758649965, 0.10589882767028522]},\n",
       "    '0.01': {'50_perc_mc_attack_log': 0.5,\n",
       "     '50_perc_mc_attack_eps': 0.5,\n",
       "     '50_perc_mc_attack_frac': 0.5,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [0.00027645053327235723, 0.00027375707425381936],\n",
       "     'set_attack_eps_vals': [0.0100355, 0.0099645],\n",
       "     'set_attack_frac_vals': [0.010318310249278052, 0.01024453791585194]},\n",
       "    'median_perc': {'50_perc_mc_attack_log': 0.51,\n",
       "     '50_perc_mc_attack_eps': 0.51,\n",
       "     '50_perc_mc_attack_frac': 0.51,\n",
       "     'successful_set_attack_log': True,\n",
       "     'successful_set_attack_eps': True,\n",
       "     'successful_set_attack_frac': True,\n",
       "     'set_attack_log_vals': [1.1593378277697396e-08, 9.279790474537345e-09],\n",
       "     'set_attack_eps_vals': [5.399999999999999e-06, 4.9999999999999996e-06],\n",
       "     'set_attack_frac_vals': [5.411613061290288e-06, 5.00929251777181e-06]}}]}}"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "log"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}