doc/source/examples/od_aegmm_kddcup.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# AEGMM and VAEGMM outlier detection on KDD Cup '99 dataset\n",
    "\n",
    "## Method\n",
    "\n",
    "The *AEGMM* method follows the [Deep Autoencoding Gaussian Mixture Model for Unsupervised Anomaly Detection](https://openreview.net/forum?id=BJJLHbb0-) ICLR 2018 paper. The encoder compresses the data while the reconstructed instances generated by the decoder are used to create additional features based on the reconstruction error between the input and the reconstructions. These features are combined with encodings and fed into a Gaussian Mixture Model (GMM). Training of the *AEGMM* model is unsupervised on *normal* (inlier) data. The sample energy of the GMM can then be used to determine whether an instance is an outlier (*high sample energy*) or not (*low sample energy*). *VAEGMM* on the other hand uses a [variational autoencoder](https://arxiv.org/abs/1312.6114) instead of a plain autoencoder.\n",
    "\n",
    "## Dataset\n",
    "\n",
    "The outlier detector needs to detect computer network intrusions using TCP dump data for a local-area network (LAN) simulating a typical U.S. Air Force LAN. A connection is a sequence of TCP packets starting and ending at some well defined times, between which data flows to and from a source IP address to a target IP address under some well defined protocol. Each connection is labeled as either normal, or as an attack.\n",
    "\n",
    "There are 4 types of attacks in the dataset:\n",
    "\n",
    "- DOS: denial-of-service, e.g. syn flood;\n",
    "- R2L: unauthorized access from a remote machine, e.g. guessing password;\n",
    "- U2R: unauthorized access to local superuser (root) privileges;\n",
    "- probing: surveillance and other probing, e.g., port scanning.\n",
    "\n",
    "The dataset contains about 5 million connection records.\n",
    "\n",
    "There are 3 types of features:\n",
    "\n",
    "- basic features of individual connections, e.g. duration of connection\n",
    "- content features within a connection, e.g. number of failed log in attempts\n",
    "- traffic features within a 2 second window, e.g. number of connections to the same host as the current connection"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This notebook requires the `seaborn` package for visualization which can be installed via `pip`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install seaborn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import logging\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "from sklearn.metrics import confusion_matrix, f1_score\n",
    "import tensorflow as tf\n",
    "tf.keras.backend.clear_session()\n",
    "from tensorflow.keras.layers import Dense, InputLayer\n",
    "\n",
    "from alibi_detect.datasets import fetch_kdd\n",
    "from alibi_detect.models.tensorflow import eucl_cosim_features\n",
    "from alibi_detect.od import OutlierAEGMM, OutlierVAEGMM\n",
    "from alibi_detect.utils.data import create_outlier_batch\n",
    "from alibi_detect.utils.fetching import fetch_detector\n",
    "from alibi_detect.saving import save_detector, load_detector\n",
    "from alibi_detect.utils.visualize import plot_instance_score, plot_feature_outlier_tabular, plot_roc\n",
    "\n",
    "logger = tf.get_logger()\n",
    "logger.setLevel(logging.ERROR)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load dataset\n",
    "\n",
    "We only keep a number of continuous (18 out of 41) features."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(494021, 18) (494021,)\n"
     ]
    }
   ],
   "source": [
    "kddcup = fetch_kdd(percent10=True)  # only load 10% of the dataset\n",
    "print(kddcup.data.shape, kddcup.target.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Assume that a model is trained on *normal* instances of the dataset (not outliers) and standardization is applied:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(400000, 18) (400000,)\n",
      "0.0% outliers\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(0)\n",
    "normal_batch = create_outlier_batch(kddcup.data, kddcup.target, n_samples=400000, perc_outlier=0)\n",
    "X_train, y_train = normal_batch.data.astype('float32'), normal_batch.target\n",
    "print(X_train.shape, y_train.shape)\n",
    "print('{}% outliers'.format(100 * y_train.mean()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "mean, stdev = X_train.mean(axis=0), X_train.std(axis=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Apply standardization:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train = (X_train - mean) / stdev"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load or define AEGMM outlier detector\n",
    "\n",
    "The pretrained outlier and adversarial detectors used in the example notebooks can be found [here](https://console.cloud.google.com/storage/browser/seldon-models/alibi-detect). You can use the built-in ```fetch_detector``` function which saves the pre-trained models in a local directory ```filepath``` and loads the detector. Alternatively, you can train a detector from scratch:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "load_outlier_detector = True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No threshold level set. Need to infer threshold using `infer_threshold`.\n"
     ]
    }
   ],
   "source": [
    "filepath = 'my_path'  # change to directory (absolute path) where model is downloaded\n",
    "detector_type = 'outlier'\n",
    "dataset = 'kddcup'\n",
    "detector_name = 'OutlierAEGMM'\n",
    "filepath = os.path.join(filepath, detector_name)\n",
    "if load_outlier_detector:  # load pretrained outlier detector\n",
    "    od = fetch_detector(filepath, detector_type, dataset, detector_name)\n",
    "else:  # define model, initialize, train and save outlier detector\n",
    "    # the model defined here is similar to the one defined in the original paper\n",
    "    n_features = X_train.shape[1]\n",
    "    latent_dim = 1\n",
    "    n_gmm = 2  # nb of components in GMM\n",
    "\n",
    "    encoder_net = tf.keras.Sequential(\n",
    "    [\n",
    "        InputLayer(input_shape=(n_features,)),\n",
    "        Dense(60, activation=tf.nn.tanh),\n",
    "        Dense(30, activation=tf.nn.tanh),\n",
    "        Dense(10, activation=tf.nn.tanh),\n",
    "        Dense(latent_dim, activation=None)\n",
    "    ])\n",
    "\n",
    "    decoder_net = tf.keras.Sequential(\n",
    "    [\n",
    "        InputLayer(input_shape=(latent_dim,)),\n",
    "        Dense(10, activation=tf.nn.tanh),\n",
    "        Dense(30, activation=tf.nn.tanh),\n",
    "        Dense(60, activation=tf.nn.tanh),\n",
    "        Dense(n_features, activation=None)\n",
    "    ])\n",
    "\n",
    "    gmm_density_net = tf.keras.Sequential(\n",
    "    [\n",
    "        InputLayer(input_shape=(latent_dim + 2,)),\n",
    "        Dense(10, activation=tf.nn.tanh),\n",
    "        Dense(n_gmm, activation=tf.nn.softmax)\n",
    "    ])\n",
    "    \n",
    "    # initialize outlier detector\n",
    "    od = OutlierAEGMM(threshold=None,  # threshold for outlier score\n",
    "                      encoder_net=encoder_net,         # can also pass AEGMM model instead\n",
    "                      decoder_net=decoder_net,         # of separate encoder, decoder\n",
    "                      gmm_density_net=gmm_density_net, # and gmm density net\n",
    "                      n_gmm=n_gmm,\n",
    "                      recon_features=eucl_cosim_features)  # fn used to derive features\n",
    "                                                           # from the reconstructed\n",
    "                                                           # instances based on cosine \n",
    "                                                           # similarity and Eucl distance \n",
    "    \n",
    "    # train\n",
    "    od.fit(X_train,\n",
    "           epochs=50,\n",
    "           batch_size=1024,\n",
    "           verbose=True)\n",
    "    \n",
    "    # save the trained outlier detector\n",
    "    save_detector(od, filepath)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The warning tells us we still need to set the outlier threshold. This can be done with the `infer_threshold` method. We need to pass a batch of instances and specify what percentage of those we consider to be normal via `threshold_perc`. Let's assume we have some data which we know contains around 5% outliers. The percentage of outliers can be set with `perc_outlier` in the `create_outlier_batch` function."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5.0% outliers\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(0)\n",
    "perc_outlier = 5\n",
    "threshold_batch = create_outlier_batch(kddcup.data, kddcup.target, n_samples=1000, perc_outlier=perc_outlier)\n",
    "X_threshold, y_threshold = threshold_batch.data.astype('float32'), threshold_batch.target\n",
    "X_threshold = (X_threshold - mean) / stdev\n",
    "print('{}% outliers'.format(100 * y_threshold.mean()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "New threshold: 12.8471097946167\n"
     ]
    }
   ],
   "source": [
    "od.infer_threshold(X_threshold, threshold_perc=100-perc_outlier)\n",
    "print('New threshold: {}'.format(od.threshold))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Save outlier detector with updated threshold:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "save_detector(od, filepath)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Detect outliers\n",
    "\n",
    "We now generate a batch of data with 10% outliers and detect the outliers in the batch. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1000, 18) (1000,)\n",
      "10.0% outliers\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(1)\n",
    "outlier_batch = create_outlier_batch(kddcup.data, kddcup.target, n_samples=1000, perc_outlier=10)\n",
    "X_outlier, y_outlier = outlier_batch.data.astype('float32'), outlier_batch.target\n",
    "X_outlier = (X_outlier - mean) / stdev\n",
    "print(X_outlier.shape, y_outlier.shape)\n",
    "print('{}% outliers'.format(100 * y_outlier.mean()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Predict outliers:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "od_preds = od.predict(X_outlier, return_instance_score=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Display results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "F1 score and confusion matrix:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.8889\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "labels = outlier_batch.target_names\n",
    "y_pred = od_preds['data']['is_outlier']\n",
    "f1 = f1_score(y_outlier, y_pred)\n",
    "print('F1 score: {:.4f}'.format(f1))\n",
    "cm = confusion_matrix(y_outlier, y_pred)\n",
    "df_cm = pd.DataFrame(cm, index=labels, columns=labels)\n",
    "sns.heatmap(df_cm, annot=True, cbar=True, linewidths=.5)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plot instance level outlier scores vs. the outlier threshold:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "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": [
    "plot_instance_score(od_preds, y_outlier, labels, od.threshold, ylim=(None, None))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can also plot the ROC curve for the outlier scores of the detector:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "roc_data = {'AEGMM': {'scores': od_preds['data']['instance_score'], 'labels': y_outlier}}\n",
    "plot_roc(roc_data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Investigate results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can visualize the encodings of the instances in the latent space and the features derived from the instance reconstructions by the decoder. The encodings and features are then fed into the GMM density network."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "enc = od.aegmm.encoder(X_outlier)  # encoding\n",
    "X_recon = od.aegmm.decoder(enc)  # reconstructed instances\n",
    "recon_features = od.aegmm.recon_features(X_outlier, X_recon)  # reconstructed features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "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": [
    "df = pd.DataFrame(dict(enc=enc[:, 0].numpy(), \n",
    "                       cos=recon_features[:, 0].numpy(), \n",
    "                       eucl=recon_features[:, 1].numpy(), \n",
    "                       label=y_outlier))\n",
    "\n",
    "groups = df.groupby('label')\n",
    "fig, ax = plt.subplots()\n",
    "for name, group in groups:\n",
    "    ax.plot(group.enc, group.cos, marker='o', \n",
    "            linestyle='', ms=6, label=labels[name])\n",
    "plt.title('Encoding vs. Cosine Similarity')\n",
    "plt.xlabel('Encoding')\n",
    "plt.ylabel('Cosine Similarity')\n",
    "ax.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "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": [
    "fig, ax = plt.subplots()\n",
    "for name, group in groups:\n",
    "    ax.plot(group.enc, group.eucl, marker='o', \n",
    "            linestyle='', ms=6, label=labels[name])\n",
    "plt.title('Encoding vs. Relative Euclidean Distance')\n",
    "plt.xlabel('Encoding')\n",
    "plt.ylabel('Relative Euclidean Distance')\n",
    "ax.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A lot of the outliers are already separated well in the latent space."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Use VAEGMM outlier detector\n",
    "\n",
    "We can again instantiate the pretrained VAEGMM detector from the [Google Cloud Bucket](https://console.cloud.google.com/storage/browser/seldon-models/alibi-detect). You can use the built-in ```fetch_detector``` function which saves the pre-trained models in a local directory ```filepath``` and loads the detector. Alternatively, you can train a detector from scratch:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "load_outlier_detector = True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No threshold level set. Need to infer threshold using `infer_threshold`.\n"
     ]
    }
   ],
   "source": [
    "filepath = 'my_path'  # change to directory (absolute path) where model is downloaded\n",
    "detector_type = 'outlier'\n",
    "dataset = 'kddcup'\n",
    "detector_name = 'OutlierVAEGMM'\n",
    "filepath = os.path.join(filepath, detector_name)\n",
    "if load_outlier_detector:  # load pretrained outlier detector\n",
    "    od = fetch_detector(filepath, detector_type, dataset, detector_name)\n",
    "else:  # define model, initialize, train and save outlier detector\n",
    "    # the model defined here is similar to the one defined in\n",
    "    # the OutlierVAE notebook\n",
    "    n_features = X_train.shape[1]\n",
    "    latent_dim = 2\n",
    "    n_gmm = 2\n",
    "\n",
    "    encoder_net = tf.keras.Sequential(\n",
    "    [\n",
    "        InputLayer(input_shape=(n_features,)),\n",
    "        Dense(20, activation=tf.nn.relu),\n",
    "        Dense(15, activation=tf.nn.relu),\n",
    "        Dense(7, activation=tf.nn.relu)\n",
    "    ])\n",
    "\n",
    "    decoder_net = tf.keras.Sequential(\n",
    "    [\n",
    "        InputLayer(input_shape=(latent_dim,)),\n",
    "        Dense(7, activation=tf.nn.relu),\n",
    "        Dense(15, activation=tf.nn.relu),\n",
    "        Dense(20, activation=tf.nn.relu),\n",
    "        Dense(n_features, activation=None)\n",
    "    ])\n",
    "\n",
    "    gmm_density_net = tf.keras.Sequential(\n",
    "    [\n",
    "        InputLayer(input_shape=(latent_dim + 2,)),\n",
    "        Dense(10, activation=tf.nn.relu),\n",
    "        Dense(n_gmm, activation=tf.nn.softmax)\n",
    "    ])\n",
    "    \n",
    "    \n",
    "    # initialize outlier detector\n",
    "    od = OutlierVAEGMM(threshold=None,\n",
    "                       encoder_net=encoder_net,\n",
    "                       decoder_net=decoder_net,\n",
    "                       gmm_density_net=gmm_density_net,\n",
    "                       n_gmm=n_gmm,\n",
    "                       latent_dim=latent_dim,\n",
    "                       samples=10,\n",
    "                       recon_features=eucl_cosim_features)\n",
    "    \n",
    "    # train\n",
    "    od.fit(X_train,\n",
    "           epochs=50,\n",
    "           batch_size=1024,\n",
    "           cov_elbo=dict(sim=.0025),  # standard deviation assumption\n",
    "           verbose=True)           # for elbo training\n",
    "    \n",
    "    # save the trained outlier detector\n",
    "    save_detector(od, filepath)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Need to infer the threshold again:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "New threshold: 8.69978871345519\n"
     ]
    }
   ],
   "source": [
    "od.infer_threshold(X_threshold, threshold_perc=100-perc_outlier)\n",
    "print('New threshold: {}'.format(od.threshold))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Save outlier detector with updated threshold:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "save_detector(od, filepath)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Detect outliers and display results\n",
    "\n",
    "Predict:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "od_preds = od.predict(X_outlier, return_instance_score=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "F1 score and confusion matrix:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.9463\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "labels = outlier_batch.target_names\n",
    "y_pred = od_preds['data']['is_outlier']\n",
    "f1 = f1_score(y_outlier, y_pred)\n",
    "print('F1 score: {:.4f}'.format(f1))\n",
    "cm = confusion_matrix(y_outlier, y_pred)\n",
    "df_cm = pd.DataFrame(cm, index=labels, columns=labels)\n",
    "sns.heatmap(df_cm, annot=True, cbar=True, linewidths=.5)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plot instance level outlier scores vs. the outlier threshold:"
   ]
  },
  {
   "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": [
    "plot_instance_score(od_preds, y_outlier, labels, od.threshold, ylim=(None, None))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You can zoom in by adjusting the min and max values in `ylim`. We can also compare the VAEGMM ROC curve with AEGMM:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAmkAAAFNCAYAAABbpPhvAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/MnkTPAAAACXBIWXMAAAsTAAALEwEAmpwYAABJbElEQVR4nO3dd3xUxfrH8c8DBJCiFFGvoJIQiNSgIEgRAkhTULkiRa+oWLliAxG7XK9eTUQRrBQVUelNFAUsRBFFeg/SkSpNIAVICPP7Y5f8QksWyGZ3k+/79doXe86Zc86zOyQ8zMyZMeccIiIiIhJcCgQ6ABERERE5mZI0ERERkSCkJE1EREQkCClJExEREQlCStJEREREgpCSNBEREZEgpCRNREREJAgpSRORoGFmG83soJklmdkOMxtuZiVOKNPQzH40s0Qz229mX5lZtRPKnG9mb5vZn95rrfNuX5i7n0hE5OwpSRORYNPeOVcCqA1cBTxz7ICZNQBmAF8ClwLhwBJgtplFeMsUBn4AqgNtgPOBBsAeoJ6/gjazQv66tojkT0rSRCQoOed2ANPxJGvHxAEjnHMDnXOJzrm9zrnngTlAP2+ZbsDlQAfn3Ern3FHn3E7n3H+dc9+c6l5mVt3MvjOzvWb2l5k9690/3MxeyVQuxsy2ZNreaGZ9zWwpkOx9P/6Eaw80s0He9xeY2Udmtt3MtprZK2ZW8Ny+KRHJq5SkiUhQMrMKQFtgrXe7GNAQGHeK4mOBlt731wPTnHNJPt6nJPA9MA1P61wknpY4X3UFbgRKAaOBG7zXxJuAdQJGessOB45473EV0Aq47wzuJSL5iJI0EQk2k80sEdgM7ARe8u4vg+d31vZTnLMdODberOxpypxOO2CHc+5N59whbwvd72dw/iDn3Gbn3EHn3CZgIdDBe6w5kOKcm2NmFwM3AI8755KdczuBAUCXM7iXiOQjStJEJNjc4pwrCcQAV/L/ydffwFHgH6c45x/Abu/7PacpczqXAevOKlKPzSdsj8TTugZwO//finYFEAZsN7N9ZrYPGAxcdA73FpE8TEmaiAQl59xPeLoH+3u3k4HfgNtOUbwT/99F+T3Q2syK+3irzUDEaY4lA8UybV9yqlBP2B4HxHi7azvw/0naZuAwcKFzrpT3db5zrrqPcYpIPqMkTUSC2dtASzOL9m4/DdxlZo+aWUkzK+0d2N8A+I+3zGd4EqIJZnalmRUws7Jm9qyZ3XCKe3wN/MPMHjezIt7r1vceW4xnjFkZM7sEeDy7gJ1zu4B44BNgg3Muwbt/O54nU9/0ThFSwMwqmVnTM/1SRCR/UJImIkHLm/CMAF70bv8CtAb+iWfc2SY8A/AbO+fWeMscxvPwwCrgO+AAMBdPt+lJY82cc4l4HjpoD+wA1gDNvIc/wzPFx0Y8CdYYH0Mf6Y1h5An7uwGFgZV4um/Hc2ZdsyKSj5hzJ7bUi4iIiEigqSVNREREJAgpSRMREREJQkrSRERERIKQkjQRERGRIKQkTURERCQIFQp0AGeqVKlSLjIyMtBhyFlKTk6meHFf5xiVYKK6C22qv9ClugttCxYs2O2cK3c254ZcknbxxRczf/78QIchZyk+Pp6YmJhAhyFnQXUX2lR/oUt1F9rMbNPZnqvuThEREZEgpCRNREREJAgpSRMREREJQkrSRERERIKQkjQRERGRIKQkTURERCQIKUkTERERCUJ+S9LM7GMz22lmy09z3MxskJmtNbOlZna1v2IRERERCTX+bEkbDrTJ4nhboLL39QDwgR9jEREREQkpfkvSnHM/A3uzKHIzMMJ5zAFKmdk//BWPiIiISCgJ5Ji08sDmTNtbvPtEREREQt7q1avP6fyQWLvTzB7A0yVKuXLliI+Pzzj28A/JJKed2fUWF7mfUpacgxGKr2IA4gMbg5ydGFDdhbAYUP2FqBhQ3YWQlDTHJ4vSuLZCQcJLn1tbWCCTtK3AZZm2K3j3ncQ5NwQYAhAVFeUyLzSbPG0qG1+/8czu3C8Z+u0/s3MkR2ih4NClugttqr/QpboLDcnJybz11lu8+8G7NGjQnKaPvkKZGjUgzs76moFM0qYAPc1sNFAf2O+c2x7AeERERETOyJ9//sm+ffuoUqUKu3btIj4+nqpVq+bItf2WpJnZKDyttBea2RbgJSAMwDn3IfANcAOwFkgB7jnTe0T/ZwYXnBeWdaHXr4BD+47fV7TUmd5KREREJMPy5cuJi4tj6tSp/O9//6NWrVoMGjQoR+/htyTNOdc1m+MOePhc7rH/YFr2XZ2H9qlrU0RERM6Zc449e/ZQtmxZHn30UVq2bMmgQYMoVaqUX+4XEg8O+ORULWagVjMRERE5J0ePHmXKlCnExsZSpkwZpk6dyo8//uj3++adJE0tZiIiIpKDnHOYGXfffTcJCQn07duXDh065Nr9806SJiIiIpID9u/fz+DBgxkzZgxz5sxh4MCBlCpVCrOzf1LzbOSNBdZfv0LdmiIiInLOvvjiCyIiIliyZAkff/wxYWFhlC5dOtcTNMgrSdqhffD0pkBHISIiIiFo9erV/Pvf/2bPnj3Uq1eP+fPn88UXXxAdHR3QuEI6SVtc5H7od4Fa0UREROSMrV69mltvvZXGjRtz0UUXUahQISpXrkx4eHigQwNCfExaKdPKASIiIuI75xzTp0+nWrVqFChQgKZNmzJixAiKFy8e6NBOEtItaSIiIiK+SEtL44svvqB27do89dRT7Nixg8jISB599NGgTNAglFvSXr+Cfa44pQIdh4iIiASt5ORkDhw4QKFChRgxYgSvv/46bdq0CciDAGcqdFvSDu2j9uGhgY5CREREgtDu3bvp168f4eHhjBo1inLlyjF9+nTatm0bEgkahHJLmoiIiMgJ0tPTAWjYsCFNmzZl1qxZREVFBTiqs6MkTURERELekiVLiIuLIyUlhUmTJrFs2TKKFCkS6LDOSeh2d4qIiIgADz30EG3btiU6Oprhw4cDhHyCBkrSREREJMSkp6czYcIE7rzzTpxzPPbYY2zYsIGnnnqKCy64INDh5RglaSIiIhIypkyZQtWqVXnjjTf45z//iXOOqlWr5omWsxMpSRMREZGgtm/fPgYMGEBqaiplypRh6NCh/Pbbb3To0IECBfJuKpN3P5mIiIiEtO3bt9OnTx8qVarEwoULOXDgAI0bN6Zp06YhM43GudDTnSIiIhJUVq1axT/+8Q/Wr19PamoqCxcu5Iorrgh0WLlOSZqIiIgEhd9++43Y2Fh+++03Jk2aRKNGjWjUqFGgwwoYdXeKiIhIwDjnOHjwIBs2bOBf//oXLVu2ZMOGDTRs2DDQoQWckjQRERHJdWlpaYwYMYJatWrxzjvvEB4ezpo1a3j44YcpVqxYoMMLCuruFBERkVyVlpZG9erVueyyy3jzzTdp2bIlQJ5+UvNsKEkTERERv9u5cyfvvPMOBw4cYODAgXz//fdcfvnlgQ4rqCllFREREb969tlnufLKK9m1axePPPIIgBI0HyhJExERkRy3aNEiXn/9dQCaNWvGypUr+fDDD4mMjAxwZKFDSZqIiIjkmF9++YVWrVrRvn17wsLCcM7RsmVLLrnkkkCHFnKUpImIiMg5SU9P55tvvsE5x+bNm7n99ttZv349vXv3zhcrA/iLHhwQERGRs3Lw4EGGDx9O//79ueSSS7j22mvp2rVroMPKM5SkiYiIyBn5+++/KVGiBF999RXffvstI0aMyNcrA/iLujtFRETEJ5s3b6Z3795UqlSJ+fPn06lTJ6ZMmaIEzU+UpImIiEiWnHOsXLmS6OhoAJYsWUKDBg0CHFXep+5OEREROaVffvmFuLg4Wrduzb///W/Wr19PqVKlAh1WvqEkTURERI5z8OBBWrZsyY4dO3jyySe56667MDMlaLlMSZqIiIiQmprKyJEjSU5O5uGHH+aVV17huuuuo2DBgoEOLd/SmDQREZF8btCgQVSqVImRI0dSvXp1AGJiYpSgBZha0kRERPKhv/76i9mzZ/PPf/6TwoUL8+WXX3L11VcHOizJRC1pIiIi+cjatWt56KGHqFq1KrNmzcI5x0MPPaQELQipJU1ERCQfWLt2LZGRkYwePZpy5cqxatUqLrrookCHJVlQkiYiIpJHOeeYMWMGsbGxrFu3juXLl/P8888HOizxkbo7RURE8pgjR47gnOOjjz6id+/e3H333axdu5aSJUsGOjQ5A0rSRERE8oiUlBTeffddKleuzMKFC+nWrRvLli2jW7duhIWFBTo8OUNK0kRERPKApUuXEh4ezg8//MDIkSOpU6cOhQsXxswCHZqcJY1JExERCVGbNm1iwIABNG/enDZt2vDTTz9x5ZVXBjosySF+bUkzszZm9oeZrTWzp09x/HIzm2lmi8xsqZnd4M94RERE8oLExETuvPNOrr76asLCwqhbty6FCxdWgpbH+K0lzcwKAu8BLYEtwDwzm+KcW5mp2PPAWOfcB2ZWDfgGqOivmEREREKVc46ff/6Z5ORk2rZtS6NGjXjnnXe0nmYe5s+WtHrAWufceudcKjAauPmEMg443/v+AmCbH+MREREJOc45Jk2aRIMGDbjvvvtITk7GzHjooYeUoOVx/hyTVh7YnGl7C1D/hDL9gBlm9ghQHLjej/GIiIiEjMOHD7Ny5UrMjPj4ePr06cMtt9yi9TTzkUA/ONAVGO6ce9PMGgCfmVkN59zRzIXM7AHgAYBy5coRHx9PjPdYfHx8bsYr5ygpKUl1FqJUd6FN9Rc6kpKS+Oqrr5g4cSLR0dE8/vjjdOjQAYBZs2YFODrJTf5M0rYCl2XaruDdl9m9QBsA59xvZlYUuBDYmbmQc24IMAQgKirKxcTEQLznWExMTI4HLv4THx+vOgtRqrvQpvoLfsnJyRQvXpz777+flJQUvv/+e6Kjo1V3+Zg/k7R5QGUzC8eTnHUBbj+hzJ9AC2C4mVUFigK7/BiTiIhIUPnjjz/o378/33zzDWvXrmXw4MEUKKBpTMWPDw44544APYHpQAKepzhXmNnLZnaTt1hv4H4zWwKMAu52zjl/xSQiIhJMPvroI6677jrKly/PkiVLOO+885SgSQa/jklzzn2DZ1qNzPtezPR+JdDInzGIiIgEC+cc06ZN44033mDw4MF06NCBLl26ULx48UCHJkEo0A8OiIiI5AvLli3jjjvuAKBv375UrFhR62lKlpSkiYiI+ElycjIfffQRDRs2JCIigtjYWNq0aaP1NMUn6vgWERHJYUlJSbz00kuEh4fz008/UaxYMcqUKUPbtm2VoInP1JImIiKSQzZu3Mj+/fuJiopi//79zJo1i6ioqECHJSFKLWkiIiLnaMmSJdxxxx3UqVOHuXPnUrRoUd5++20laHJOlKSJiIicBeccO3fuxDlH7969qV27NuvXr+f+++8PdGiSR6i7U0RE5Aykp6czadIk4uLiuPjii/nqq6/4/vvvAx2W5EFK0kRERHzgnMPM6NatG+vXr+fZZ5/lpptuyv5EkbOkJE1ERCQL+/bt44MPPmDs2LHMnTuX999/n/PPP19PaYrfaUyaiIjIaXz++edUqlSJhIQERowYQVhYGBdccIESNMkVStJEREQySUhI4MEHH2Tv3r1ce+21LFq0iBEjRlCzZs1Ahyb5jJI0ERER4I8//uCWW24hJiaGChUqUKhQISIjI7n88ssDHZrkUxqTJiIi+dbRo0f55ptvqFmzJgULFqRly5aMHDmSYsWKBTo0ESVpIiKS/6SlpTFq1Cji4uIICwtj6NCh1K1bl8jIyECHJpJBSZqIiOQbSUlJHDhwgLCwMEaPHs1bb71Fy5Yt9SCABCWNSRMRkTxv586dvPDCC4SHhzNu3DjKlSvHN998Q6tWrZSgSdBSS5qIiORZR44cwcy47rrraNasGb/99pu6NCVkKEkTEZE8Z+HChcTFxXH48GEmTZrEsmXLKFy4cKDDEjkj6u4UEZE85f777+emm27immuuYcSIEQBK0CQkKUkTEZGQduTIEcaMGcPtt9+Oc44nn3yS9evX07t3b0qWLBno8ETOmpI0EREJWZMnTyYqKopBgwbRpUsXAKKiotRyJnmCkjQREQkpe/fupX///qSmplKuXDlGjBjB7Nmzuemmm/SkpuQpPidpZqbpl0VEJGC2bdtGr169iIyMZMWKFSQmJtKoUSMaNWoU6NBE/CLbpzvNrCEwDCgBXG5m0cCDzrl/+zs4ERGRFStWUKFCBTZs2ICZsXTpUipUqBDosET8zpeWtAFAa2APgHNuCdDEn0GJiIj88ssvtG/fnhYtWrBy5UoaNWrEm2++qQRN8g2fujudc5tP2JXuh1hERCSfO3r0KCkpKaxfv5577rmHG2+8kQ0bNtCgQYNAhyaS63xJ0jZ7uzydmYWZ2ZNAgp/jEhGRfCQ1NZVPPvmEGjVq8P777xMREcEff/zBQw89xHnnnRfo8EQCwpcVBx4CBgLlga3ADEDj0UREJEekpaVRrVo1IiIieOedd2jevDkABQpoAgLJ33xJ0qKcc3dk3mFmjYDZ/glJRETyur/++ouBAweSlJTEoEGDiI+P11gzkRP48t+Ud3zcJyIikq2nn36aqlWrsn//fh5//HEAJWgip3DaljQzawA0BMqZWa9Mh84HCvo7MBERyTvmz5/PjBkzePbZZ7n++uvp1asXF110UaDDEglqWbWkFcYzN1ohoGSm1wGgo/9DExGRUPfzzz/TokUL/vnPf1K8eHGcc1x//fVK0ER8cNqWNOfcT8BPZjbcObcpF2MSEZEQduTIEb799lvatWvHtm3buOuuu+jatSthYWGBDk0kpPjy4ECKmb0BVAeKHtvpnGvut6hERCTkpKSk8PHHH/Pmm29y2WWX0ahRo4xFz0XkzPny4MAXwCogHPgPsBGY58eYREQkhOzZs4e0tDS+/vprvv/+e0aOHMnPP/9MmTJlAh2aSEjzJUkr65z7CEhzzv3knOsOqBVNRCSf27RpE4899hiVK1dmwYIFdOrUicmTJ2t1AJEc4kuSlub9c7uZ3WhmVwH675GISD7lnGPFihVcffXVFClShOXLl3PttdcGOiyRPMeXMWmvmNkFQG8886OdDzzuz6BERCS4OOf4+eefiY2N5cYbb+Tf//4369ev54ILLgh0aCJ5VrZJmnPua+/b/UAzyFhxQERE8oGUlBRatGjBnj176NOnD3feeSdmpgRNxM+ymsy2INAJz5qd05xzy82sHfAscB5wVe6EKCIiue3w4cN89tlnHDp0iJ49exIbG0ujRo0oWFBzmYvklqzGpH0E3AeUBQaZ2edAfyDOOacETUQkjxowYADh4eFMmDCB6OhoAJo0aaIETSSXZdXdWReo5Zw7amZFgR1AJefcntwJTUREcsv27duZPXs2HTt2pFixYnz77bcZCZqIBEZWLWmpzrmjAM65Q8B6JWgiInnLH3/8wf3330/16tWZM2cOAA8++KASNJEgkFVL2pVmttT73oBK3m0DnHOuVnYXN7M2wEA8C7IPc869fooynYB+gAOWOOduP7OPICIiZ2r16tVUqVKF8ePHU758eVavXs2FF14Y6LBEJJOskrSq53Jh74MH7wEtgS3APDOb4pxbmalMZeAZoJFz7m8z04q7IiJ+4pzj22+/JS4ujk2bNrF06VKee+65QIclIqeR1QLr57qoej1grXNuPYCZjQZuBlZmKnM/8J5z7m/vPXee4z1FROQER44cwTnHsGHDeOedd+jbty+dOnXSguciQc6XyWzPVnlgc6btLUD9E8pUATCz2Xi6RPs556adeCEzewB4AKBcuXLEx8cT4z0WHx+fs1GLXyUlJanOQpTqLvQcPHiQqVOnMn78eJ555hmqV6/OwIEDMTNmz54d6PDER/rZy7/8maT5ev/KQAxQAfjZzGo65/ZlLuScGwIMAYiKinIxMTEQ7zkWExOTW7FKDoiPj1edhSjVXWhZsmQJt912G02aNGHKlCmkpKSo/kKUfvbyL1/W7sTMzjOzqDO89lbgskzbFbz7MtsCTHHOpTnnNgCr8SRtIiJyhjZs2MAjjzzCl19+SdWqVZk9ezYTJkygXr16gQ5NRM5CtkmambUHFgPTvNu1zWyKD9eeB1Q2s3AzKwx0AU48bzKeVjTM7EI83Z/rfYxdRESAAwcOcPvtt1O3bl2KFy9OvXr1KFy4MFWqVAl0aCJyDnzp7uyH5yGAeADn3GIzC8/uJOfcETPrCUzHM97sY+fcCjN7GZjvnJviPdbKzFYC6UAfzcUmIpI95xzx8fGkpKRwww03EBMTwwcffKD1NEXyEF+StDTn3H4zy7zP+XJx59w3wDcn7Hsx03sH9PK+REQkG845JkyYQGxsLImJibz66quYGQ888ECgQxORHOZLkrbCzG4HCnrnNXsU+NW/YYmISGaHDh1ixYoV1KlTh9mzZ/Pcc89x0003UaCAT0OLRSQE+fLT/QhQHTgMjAT2A4/7MSYREfHat28fr732GuHh4QwcOBDwLIB+yy23KEETyeN8aUm70jn3HKBpqUVEcklSUhIlSpTgySefJDU1lRkzZlCzZs1AhyUiuciXJO1NM7sEGA+Mcc4t93NMIiL5VkJCAm+88QbTp09n3bp1DB06lBPGBItIPpFtW7lzrhnQDNgFDDazZWb2vN8jExHJZ4YOHUpMTAzh4eEsXbqUokWLKkETycd8GtDgnNvhnBsEPIRnzrQXsz5DRESyc/ToUb7++muaNm3KmjVr6NixIxs2bOCFF16gbNmygQ5PRAIs2+5OM6sKdAZuBfYAY4Defo5LRCRPW7JkCXfccQdhYWH07duX8PBwChUK9Ep9IhJMfPmN8DGexKy1c26bn+MREcmzkpKSGDp0KNdddx0RERG89dZbtGzZUl2aInJKvoxJa+Cce1sJmojI2UlMTOT5558nPDycOXPmUKJECcqUKUOrVq2UoInIaZ22Jc3MxjrnOpnZMo5fYcDwLBZQy+/RiYiEsPXr13PgwAGuvPJKDh06xG+//UZkZGSgwxKREJFVd+dj3j/b5UYgIiJ5xcKFC4mLi+P7778nNjaW2rVr079//0CHJSIh5rTdnc657d63/3bObcr8Av6dO+GJiIQG5xx//fUXzjmefvpp6tWrx4YNG7j33nsDHZqIhChfpuBoeYp9bXM6EBGRUHTkyBHGjBlD3bp1eeCBBzAzZsyYQa9evShZsmSgwxOREJbVmLQeeFrMIsxsaaZDJYHZ/g5MRCSYOecwM+688042b95Mv379uPHGGwMdlojkIVmNSRsJfAu8BjydaX+ic26vX6MSEQlSe/fu5f3332f8+PHMmzePwYMHc/755wc6LBHJg7Lq7nTOuY3Aw0BiphdmVsb/oYmIBJcRI0YQGRnJunXrGDlyJGFhYUrQRMRvsmtJawcswDMFR+bJfBwQ4ce4RESCwooVKxgwYABxcXE0atSIpUuXUqFChUCHJSL5QFZPd7bz/hnunIvw/nnspQRNRPK0VatW0b59e1q0aEFERASFChWiUqVKStBEJNf4snZnI2Cxcy7ZzP4FXA287Zz70+/RiYjkomMLnkdHRxMWFka7du0YO3Ys5513XqBDE5F8yJcpOD4AUswsGs/C6uuAz/walYhILkpNTeWTTz6hRo0avPzyy+zevZtKlSrx4IMPKkETkYDxZYH1I845Z2Y3A+865z4yM83OKCIh78CBAyQmJlK4cGEmTpzIO++8Q/PmzbWepogEBV9a0hLN7BngTmCqmRUAwvwbloiI/+zYsYNnn32WiIgIJk6cSLly5fjqq69o0aKFEjQRCRq+JGmdgcNAd+fcDqAC8IZfoxIR8YO0tDTS09Np2rQpBw4cYN68eTzyyCOBDktE5JSy7e50zu0wsy+Aa8ysHTDXOTfC/6GJiOSMefPmERsby9GjR5k4cSLLly8nLEwdAiIS3LJtSTOzTsBc4DagE/C7mXX0d2AiIjnh3nvvpWPHjlx33XWMGOH5/6USNBEJBb48OPAccI1zbieAmZUDvgfG+zMwEZGzceTIEcaOHcuUKVMYNWoUTz/9NB9++KESMxEJOb6MSStwLEHz2uPjeSIiuWrixIlUrlyZwYMHc+eddwJQuXJlJWgiEpJ8aUmbZmbTgVHe7c7AN/4LSUTEd3v27GHYsGE88cQTXHLJJYwaNYprr7020GGJiJyzbFvEnHN9gMFALe9riHOur78DExHJypYtW3jssceoXLkya9asISkpiYYNGypBE5E847QtaWZWGegPVAKWAU8657bmVmAiIqeydOlSrrjiCv7880+KFCnC8uXLufTSSwMdlohIjsuqJe1j4GvgVmAB8E6uRCQicgLnHD/99BM33HADrVu3JiEhgYYNGxIXF6cETUTyrKzGpJV0zg31vv/DzBbmRkAiIsekp6dz6NAhduzYwYMPPkjv3r2ZOHEiRYsWDXRoIiJ+l1WSVtTMrgKOrZFyXuZt55ySNhHxi8OHDzNixAj69+/Pgw8+SK9evUhISNCSTSKSr2SVpG0H3sq0vSPTtgOa+ysoEcm/UlNTqVq1KlFRUQwePJimTZsCKEETkXzntEmac65ZbgYiIvnXtm3bGDhwIMnJybz77rv88ssvGmsmIvmeJqUVkYDq06cPNWrU4ODBgzz55JMAStBERFCSJiIBMGfOHP773/8C0LZtW1avXs2gQYOoWLFiYAMTEQkiStJEJNfEx8fTtGlTunTpQqlSpXDO0bx5cy688MJAhyYiEnSyTdLM419m9qJ3+3Izq+f/0EQkL0hLS2Py5Mk45/jrr7948MEHWbNmDY888ogeBhARyYIva3e+DxzF8zTny0AiMAG4xo9xiUiIS05OZtiwYbz11ltERETQtGlTOnfuHOiwRERChi9JWn3n3NVmtgjAOfe3mRX2c1wiEqJ27dpF6dKlmTp1KrNmzWLcuHHUq6fGdxGRM+XLmLQ0MyuIZ240zKwcnpY1EZEMGzZsoGfPnlSpUoUFCxbQqVMnxo8frwRNROQs+ZKkDQImAReZ2avAL8D/fLm4mbUxsz/MbK2ZPZ1FuVvNzJlZXZ+iFpGg4Zxj+fLlXHPNNZQsWZKVK1dSv379QIclIhLysu3udM59YWYLgBZ4loS6xTmXkN153ta394CWwBZgnplNcc6tPKFcSeAx4PeziF9EAsA5x8yZM4mNjeXmm2+mR48ebNiwgZIlSwY6NBGRPCPbJM3MLgdSgK8y73PO/ZnNqfWAtc659d5zRgM3AytPKPdfIBbocwZxi0iApKSkEBMTQ2JiIn369OGOO+7AzJSgiYjkMF8eHJiKZzyaAUWBcOAPoHo255UHNmfa3gIc1wdiZlcDlznnppqZkjSRIHXo0CGmTJnC0qVLefTRRxkwYAANGjSgQAFNtSgi4i++dHfWzLztTaz+fa43NrMCeBZsv9uHsg8ADwCUK1eO+Ph4YrzH4uPjzzUUyUVJSUmqsxAzZswYxo4dS0REBBERERn19/PPPwc2MDkj+tkLXaq7/MuXlrTjOOcWmpkvo4K3Apdl2q7g3XdMSaAGEO+d0PISYIqZ3eScm3/CPYcAQwCioqJcTEwMxHuOxcTEnOlHkACKj49XnYWArVu3Mnv2bDp16sSaNWt4+OGH2b17t+ouhOlnL3Sp7vIvX1Yc6JXp9aSZjQS2+XDteUBlMwv3zqvWBZhy7KBzbr9z7kLnXEXnXEVgDnBSgiYiuSchIYHu3btTs2ZNFi5cCMD9999PjRo1AhyZiEj+40tLWubRwEfwjFGbkN1JzrkjZtYTmA4UBD52zq0ws5eB+c65KVlfQURyS0JCAlWrVmXy5MmEh4ezdu1aypQpE+iwRETytSyTNO80GiWdc0+ezcWdc98A35yw78XTlI05m3uIyNk5evQoU6dOJTY2lh07drB48WKeeeaZQIclIiJep+3uNLNCzrl0oFEuxiMifpaWloZzjmHDhvHSSy/xyCOPsGrVKkqUKBHo0EREJJOsxqTN9f652MymmNmdZvbPY6/cCE5Eck5iYmLGYudLlizhnnvuYcGCBXTu3JlChc74GSIREfEzX34zFwX2AM35//nSHDDRj3GJSA5atGgRrVq1onnz5kyePJnatWsHOiQREclGVknaRWbWC1jO/ydnxzi/RiUi52zdunW8+eabtGrVihtuuIHffvuNyMjIQIclIiI+yqq7syBQwvsqmen9sZeIBKEDBw7QuXNn6tevT5kyZWjYsCGFCxdWgiYiEmKyaknb7px7OdciEZGz5pzj+++/59ChQ7Rr145WrVoxbNgwracpIhLCsmpJsyyOiUgQcM4xZswY6taty+OPP056ejpmxr333qsETUQkxGXVktYi16IQkTNy8OBBli9fzjXXXMPcuXPp168fN954oxY8FxHJQ077G905tzc3AxGR7O3du5dXXnmF8PBwPvjgAwDefPNN2rdvrwRNRCSP0eRIIiEgMTGRkiVL0rdvX9LT0/nxxx+pVq1aoMMSERE/UpImEsRWrFhBXFwcP/zwA2vXrmXIkCGYabioiEh+oP4RkSA1ePBgWrRoQVRUFMuWLaNo0aJK0ERE8hElaSJB4ujRo3z55Zc0btyYtWvX0qlTJzZs2MCzzz5L6dKlAx2eiIjkMnV3igSBxYsX07VrV4oXL07fvn0JDw+nYMGCgQ5LREQCSEmaSIAcOHCAIUOG0LRpUyIjI3nvvfdo1qyZujRFRARQd6dIrjtw4ADPPPMMERERLFy4kPPPP5/SpUvTvHlzJWgiIpJBLWkiuWTNmjUkJSVRtWpV0tPTmTdvHuHh4YEOS0REgpRa0kT8bN68edx22200atSIJUuWULRoUeLi4pSgiYhIlpSkifiBc47t27fjnOP555+ncePGrF+/nrvvvjvQoYmISIgI6e7OC84LC3QIIsc5cuQIY8eOJS4ujoiICCZOnMj06dMDHZaIiISgkE7SlrzUKtAhiACeljMz41//+hfbt2/nf//7H23btg10WCIiEsJCMkmL/s8MlgQ6CBFgz549vPvuu0yYMIGFCxcydOhQSpYsGeiwREQkDwjJMWn7D6YFOgQRhg8fTuXKldm8eTPjxo2jUKFCStBERCTHhGSSJhIoS5cu5Z577mHv3r00adKE5cuXM2zYMKKiogIdmoiI5DEhmaQtLnI/FC0V6DAkH0lISOCGG26gTZs2XHnllRQuXJiIiAguvfTSQIcmIiJ5VEiOSStlyfD0/kCHIXlceno6X375JXXq1KFw4cJ06NCBiRMnUrRo0UCHJiIi+UBIJmki/nT48GFGjBhB//79KVWqFB9++CFXXXUVlSpVCnRoIiKSjyhJE/Hav38/SUlJFC5cmK+//prBgwfTtGlTracpIiIBEZJj0kRy0rZt23jqqaeIiIhg8uTJlCtXji+//JKYmBglaCIiEjBqSZN8Ky0tjQIFChATE0Pbtm1ZuHAhV1xxRaDDEhERAZSkST40Z84c4uLiAJg4cSIrVqwgLExLjImISHAJue7OEkkb2OeKBzoMCVF33303Xbt2pXnz5nz22WcAStBERCQohVySZi6d2oeHBjoMCRFpaWl89tlndOrUiaNHj/L888+zZs0aevbsSfHiSvZFRCR4hVySJuKrcePGUalSJT755BO6d++OmREZGUmhQurlFxGR4KckTfKUXbt28dprr5GamkqFChUYP348P/74I23atNGTmiIiElKUpEme8Oeff9KzZ0+ioqLYuHEjycnJNGjQgHr16gU6NBERkbOifh8JaYsWLSIiIoItW7ZQsmRJVq5cySWXXBLosERERM6ZkjQJOc45fvzxR+Li4li+fDmTJk2iYcOGNGzYMNChiYiI5Bh1d0rISE9PJzk5mXXr1vHII4/QuXNn1q9fry5NERHJk5SkSdA7ePAgH374IVFRUQwdOpTIyEhWrFhB9+7dKVKkSKDDExER8Qt1d0pQS01NpWrVqtSoUYNPPvmExo0bA+hJTRERyfOUpEnQ2bJlC2+//TYHDx7kvffeY86cOXoYQERE8h11d0pQ6dWrF7Vq1SI9PZ2+ffsCKEETEZF8ya9Jmpm1MbM/zGytmT19iuO9zGylmS01sx/M7Ap/xiPB6ddff+Xll18GoH379qxdu5YBAwZw+eWXBzgyERGRwPFbkmZmBYH3gLZANaCrmVU7odgioK5zrhYwHojzVzwSfH744QcaN27Mv/71L8qVK4dzjmbNmlGmTJlAhyYiIhJw/mxJqwesdc6td86lAqOBmzMXcM7NdM6leDfnABX8GI8EgbS0NCZOnIhzjj179tCzZ09Wr15Njx499DCAiIhIJv58cKA8sDnT9hagfhbl7wW+9WM8EkCJiYkMHTqU119/nejoaJo1a0anTp0CHZaIiEjQCoqnO83sX0BdoOlpjj8APABQ5x+exr/4+Phcik7Oxd9//03JkiX55Zdf+Omnn3j22WepXbs2S5YsCXRocoaSkpL0cxfCVH+hS3WXf/kzSdsKXJZpu4J333HM7HrgOaCpc+7wqS7knBsCDAGoe2lBBxATE5PD4UpOWrduHf3792fMmDFMmzaNfv36AZ7kWnUXmlR3oU31F7pUd/mXP8ekzQMqm1m4mRUGugBTMhcws6uAwcBNzrmdfoxFcolzjmXLllG/fn3Kli3LqlWrtGyTiIjIWfBbS5pz7oiZ9QSmAwWBj51zK8zsZWC+c24K8AZQAhjnHTT+p3PuJn/FJP7hnOP7778nLi6ODh060KNHDzZu3EiJEiUCHZqIiEjI8uuYNOfcN8A3J+x7MdP76/15f/G/lJQUrrvuOg4dOsRTTz1F165dMTMlaCIiIucoKB4ckNBy8OBBPvnkE9LS0njsscd49913qV+/PgUKaAELERGRnKJ/VeWMxMbGUrFiRaZPn079+p4ZVRo0aKAETUREJIepJU2ytXnzZn799Vc6d+7MhRdeSHx8PFWrVg10WCIiInmamj/ktJYvX85dd9113Lxm9957rxI0ERGRXKAkTY7jnGPFihUAfP3111x55ZWsXbuW//3vfwGOTEREJH9Rd6cAcPToUaZMmUJcXBy7d+9m4cKFPP3004EOS0REJN9SS1o+l5qainOOIUOG8Oqrr9KrVy8SEhI0hYaIiEiAKUnLpw4cOMAbb7xBeHg4S5cu5b777mPu3Ll07NiRggULBjo8ERGRfE9JWj60cOFCIiIiWLRoEVOnTiU6OppChQrhXfVBREREgoDGpOUTq1evpn///rRt25Ybb7yRefPmER4eHuiwRERyXVpaGlu2bOHQoUOBDsUnF1xwAQkJCYEOQ7JRtGhRKlSoQFhYWI5dU0laHrd//37uvfdefv75Z3r06EHjxo0pXLiwEjQRybe2bNlCyZIlqVixYkj0ICQmJlKyZMlAhyFZcM6xZ88etmzZkqP/vipJy4Occ0yfPp3U1FTat29Pu3bt+PTTTylevHigQxMRCbhDhw6FTIImocHMKFu2LLt27crR62pMWh7inOOLL76gdu3aPPXUU4DnL87dd9+tBE1EJBMlaJLT/PF3SklaHpCcnMzvv/+OmbF48WJee+01lixZwk033RTo0ERE5DQmT56MmbFq1aqMfRs3buS8886jdu3aGa+RI0cCkJSURI8ePahUqRJXX301derUYejQoRnnmRnPP/98xrV2795NWFgYPXv2BKBfv36YGWvXrs0o8/bbb2NmzJ8/36eYa9euTZcuXY7bFxMTc9z5GzdupEaNGhnbc+fOpUmTJkRFRXHVVVdx3333kZKS4tP9pk2bRlRUFJGRkbz++uunLLNp0yZatGhBrVq1iImJYcuWLRnH+vbtS40aNahRowZjxozJ2H/33XcTHh6e8R0vXrwY8AwRat++PdHR0VSvXp1PPvnEpzj9RUlaCNu9ezf9+vUjPDw84wf1jTfe4IYbbtD/EkVEgtyoUaNo3Lgxo0aNOm5/pUqVWLx4ccbr9ttvB+C+++6jdOnSrFmzhoULFzJt2jT27t2bcV54eDhTp07N2B43bhzVq1c/7to1a9Zk9OjRWZY5nYSEBNLT05k1axbJyck+nfPXX39x2223ERsbyx9//MGiRYto06YNiYmJ2Z6bnp7Oww8/zLfffsvKlSsZNWoUK1euPKnck08+Sbdu3Vi6dCkvvvgizzzzDABTp05l4cKFLF68mN9//53+/ftz4MCBjPPeeOONjO+4du3aALz33ntUq1aNJUuWEB8fT+/evUlNTfXps/qDkrQQtH//fgCeeeYZtm7dyqxZsxg2bFiAoxIREV8lJSXxyy+/8NFHHx2XNJ3OunXrmDt3Lq+88goFCnj+6S5Xrhx9+/bNKFOsWDGqVq2a0ao1ZswYOnXqdNx1brnlFr788suMa15wwQVceOGFPsU8atQo7rzzTlq1apVxjey899573HXXXTRo0CBjX8eOHbn44ouzPXfu3LlERkYSERFB4cKF6dKlyynvu3LlSpo3bw5As2bNMsqsXLmSJk2aUKhQIYoXL06tWrWYNm1alvc0MxITE3HOkZSURJkyZShUKHDD95WkhZAlS5Zwxx13UL16dQ4dOsSQIUMYOnQoUVFRgQ5NRETOwJdffkmbNm2oUqUKZcuWZcGCBRnH1q1bd1x356+//sqKFSuIjo7OSNBOp0uXLowePZrNmzdTsGBBLr300uOOn3/++Vx22WUsX76c0aNH07lz5+OO33fffaft+hwzZgxdunSha9euJ7X+nc7y5cupU6fOKY/NnDnzuM957NWwYUMAtm7dymWXXZZRvkKFCmzduvWk60RHRzNx4kQAJk2aRGJiInv27CE6Oppp06aRkpLC7t27mTlzJps3b84477nnnqNWrVo88cQTHD58GICePXuSkJDApZdeSs2aNRk4cGC237k/6enOEPHBBx/w3//+l8cee4z333+fokWLBjokEZE8oeLTU7MvdIY2vn5jlsdHjRrFY489BngSq1GjRmUkM8e6O49JTExk5syZx53/6quvMm7cOHbu3Mm2bdsy9rdp04YXXniBiy+++KQE7Jhjidz06dP54Ycfjht3dbpemfnz53PhhRdy+eWXU758ebp3787evXspU6bMKYfX+DLkplmzZsd9zrPVv39/evbsyfDhw2nSpAnly5enYMGCtGrVinnz5tGwYUPKlStHgwYNMlbUee2117jkkktITU3lgQceIDY2lhdffJHp06dTu3ZtfvzxR9atW0fLli257rrrOP/88885zrOhJC1IpaenM3nyZPr3789nn31G165d6d69O0WKFAl0aCIieUp2CVVO27t3Lz/++CPLli3DzEhPT8fMeOONN057zrFxUkePHqVAgQI899xzPPfccyets1y4cGHq1KnDm2++ycqVK5kyZcpJ12rXrh19+vShbt26Picfo0aNYtWqVVSsWBHwLC04YcIE7r//fsqWLcvff/993Oc71oVavXp1FixYwM0333zSNWfOnMkTTzxx0v5ixYrx66+/Ur58+eNavrZs2UL58uVPKn/ppZdmtKQlJSUxYcIESpUqBZDxPQHcfvvtVKlSBYB//OMfABQpUoR77rmH/v37A/DJJ5/w9NNPY2ZERkYSHh7OqlWrqFevnk/fU05Td2cQWrRoEVWrViUuLo4+ffoQHh5OqVKllKCJiOQB48eP584772TTpk1s3LiRzZs3Ex4ezqxZs057TmRkJHXr1uX5558nPT0d8Mz35pw7qWzv3r2JjY2lTJkyp7xWsWLFiI2NzUhesnP06FHGjh3LsmXL2LhxIxs3buTLL7/M6PKMiYnh888/z4jl008/pVmzZoCn+/DTTz/l999/z7jexIkT+euvvzJa0k58/frrrwBcc801rFmzhg0bNpCamsro0aNPOWvB7t27OXr0KOBpIevevTvgaezYs2cPAEuXLmXp0qW0atUKgO3btwOeqasmT56c8TTq5Zdfzg8//AB4Hnr4448/iIiI8Ol78ge1pAWJffv28eGHH9KiRQsiIyMZOnQoTZo00VOaIiJ5zKhRo44b8A9w6623Zuw/NibtmNtvv52nnnqKYcOG0adPHyIjIylbtiznnXcecXFxJ12/evXq2T6xeeI0Gsfcd999PPTQQ9StWzdj36xZsyhfvvxx49uaNGnCypUr2b59Ow888ACrVq0iOjoaM6Nu3bq89tprAFx88cWMHj2aJ598kp07d1KgQAGaNGlCmzZtsv2eChUqxLvvvkvr1q1JT0+ne/fuGZ/rxRdfpG7dutx0003Ex8fzzDPPYGY0adKE9957D/As/3XdddcBnrF4n3/+ecZDAHfccQe7du3COUft2rX58MMPAXjhhRe4++67qVmzJs45YmNjfX6wwh/sVFl4MKt7aUG3u9uUXG+e9pcDBw7wyiuv8NFHH9G2bVteeuklKleuHOiw/CY+Pp6YmJhAhyFnQXUX2lR//y8hIYGqVasGOgyfaVmo0HGqv1tmtsA5V/c0p2RJLWkBsmrVKlJSUqhWrRoFCxZk4cKFXHHFFYEOS0RERIKExqTlst9++40OHTrQpEkTli9fTtGiRXnttdeUoImIiMhxlKTlAuccW7du5ejRo7z88su0aNGCjRs30q1bt0CHJiIiIkFK3Z1+lJaWxujRo4mLiyMqKorx48fz7bffBjosERERCQFK0vzg2Dw2d9xxB7t376Z///4Zj/2KiIiI+EJJWg7atWsXgwYNYtKkSSxevJhPPvmE4sWLBzosERERCUEak5ZDPv74Y6pUqcLOnTuZNGlSxoKuIiIimTVr1ozp06cft+/tt9+mR48egGdy1rCwsIy5u46pWLEiNWvWzFjj8tFHH8049tZbb3HllVdSs2ZNoqOj6dWrF2lpaRnnHZsv7JjatWtnTOAaHx+PmR23JNTixYsxs4yZ+LPz+OOPU758+YxJZQH69et30vkVK1Zk9+7dAOzYsYMuXbpQqVIl6tSpww033MDq1at9ut+GDRuoX78+kZGRdO7cmdTU1JPKpKamcs8992R8J/Hx8RnHxowZQ61atahevfpxc9Y98cQTGd9vlSpVMlYuAM8kvZUrV6Zy5cp8+umnPsV5rpSknYNFixbRrVs39uzZQ7NmzUhISGDw4MF5ep4zERE5N127dmX06NHH7Rs9ejRdu3YFYNy4cVx77bWnXMR85syZGTPzDxo0CIAPP/yQGTNmMGfOHJYtW8a8efO46KKLOHjwYMZ5iYmJGUssJSQknHTdGjVqMHbs2IztUaNGER0d7dPnOXr0KJMmTeKyyy7jp59+8ukc5xwdOnQgJiaGdevWsWDBAl577TX++usvn87v27cvTzzxBGvXrqV06dJ89NFHJ5UZOnQoAMuWLeO7776jd+/eHD16lD179tCnTx9++OEHVqxYwY4dOzJWGRgwYEDG9/vII4/wz3/+E/AsdfWf//yH33//nblz5/Kf//znuKWw/EVJ2llYsWIFrVq1on379tSqVYsiRYoQHh7OJZdcEujQREQkyHXs2JGpU6dmtP5s3LiRbdu2ZbR2jRo1ijfffJOtW7eyZcuWbK/36quv8sEHH2S0+hQuXJinn376uHU5O3XqxJgxYzKufywhPOaKK67g0KFD/PXXXzjnmDZtGm3btvXp88THx1O9enV69OhxysTyVGbOnElYWBgPPfRQxr7o6OiTWvxOxTnHjz/+SMeOHQG46667mDx58knlVq5cSfPmzQG46KKLKFWqFPPnz2f9+vVUrlyZcuXKAXD99dczYcKEk87P/D1Nnz6dli1bUqZMGUqXLk3Lli2ZNm2aT5/1XChJ81F6ejpjx45l06ZNFC1alC5durBu3TqefPLJkxa4FREROZ0yZcpQr169jKf9R48eTadOnTAzNm/ezPbt26lXr95xidUxzZo1y+iOGzBgAAcOHCApKYnw8PAs73nrrbdmLEL+1Vdf0b59+5PKdOzYkXHjxvHrr79y9dVXH7de9IcffnhS9+sxx5KZDh06MHXq1Ixu1qwsX76cOnXqnPJYYmJixmc88bVy5Ur27NlDqVKlMpZ4qlChAlu3bj3pOtHR0UyZMoUjR46wYcMGFixYwObNm4mMjOSPP/5g48aNHDlyhMmTJx+3kDvApk2b2LBhQ0aSt3XrVi677LKM46e7Z07TgwPZOHToEMOHD6d///5cdNFFvP/++9SuXZtKlSoFOjQREckJ/S7wwzX3Z3n4WJfnzTffzOjRozO668aMGUOnTp0Az/qa3bt354EHHsg4b+bMmcetJXngwIHjrjt9+nT69u3Lvn37GDlyJA0bNgSgbNmylC5dmtGjR1O1alWKFSt2UkydOnWic+fOrFq1iq5du2YsdA4c1+KVWWpqKt988w1vvfUWJUuWpH79+kyfPp127dqddu3p7NakLlmyJIsXLz7t8WNj2rLTvXt3EhISqFu3LldccQUNGzakYMGClC5dmg8++IDOnTtToEABGjZsyLp16447d/To0XTs2JGCBQv6dC9/UZJ2Gn///TfJyckUKVKEGTNmMHz4cBo3bhzosEREJKdlk1D5w80338wTTzzBwoULSUlJyWhVGjVqFDt27OCLL74AYNu2baxdu5arrrrqlNc5//zzKVGiBBs2bCA8PJzWrVvTunVr2rVrd9Jg+s6dO/Pwww8zfPjwU17rkksuISwsjO+++46BAwcel6SdzvTp09m3bx81a9YEICUlhfPOO4927dpRtmxZtm/fflz5xMRESpUqRfXq1Rk/fvwpr5mYmHjabs+RI0dStWpV9u3bx5EjRyhUqBBbtmyhfPnyJ5UtVKgQAwYMyNhu2LAhVapUAaB9+/YZrYlDhgw5KRkbPXp0xkLtAOXLlz/uwYMtW7bkylq46u48wZYtW+jduzeVKlXiq6++oly5ckycOFEJmoiI5JgSJUrQrFkzunfvnjHuafXq1SQlJbF161Y2btzIxo0beeaZZ06bzBzzzDPP0KNHD/bt2wd4xmwdOnTopHIdOnTgqaeeonXr1qe91ssvv0xsbKzPLUijRo1i2LBhGfFu2LCB7777jpSUFJo0acKUKVNITEwEYOLEiURHR1OwYEGaN2/O4cOHGTJkSMa1li5dyqxZszJa0k71qlatGmZGs2bNMr6XTz/9lJtvvvmk2FJSUkhOTgbgu+++o1ChQlSrVg2AnTt3Ap4Gmffff5/77rsv47xVq1bx999/06BBg4x9rVu3ZsaMGfz999/8/fffzJgxI8vvMaeoJc0rNTWVAgUK0KJFC2688UaWLFlyXP+ziIhITjo2juvYk56jRo2iQ4cOx5W59dZbue2223j11VcBz5i0YwlUrVq1GDFiBD169CA5OZn69etTpEgRSpQoQaNGjU5qfStZsuRx002cyrHu0RMdG4+WudszJSWFadOmHTdWrXjx4jRu3JivvvqKzp0707NnTxo3boyZcdFFF2VM82FmTJo0iccff5zY2FiKFi1KxYoVefvtt7P72gCIjY2lS5cuPP/881x11VXce++9AEyZMoX58+fz8ssvs3PnTlq3bk2BAgUoX748n332Wcb5jz32GEuWLAHgxRdfzGhhA08rWpcuXY7rli1TpgwvvPAC11xzTcY5ZcqU8SnWc2HOOb/fJCfVvbSg291tChtfvzFHrjd79mxiY2MJCwtjwoQJGc2n4h/x8fG50kQsOU91F9pUf/8vISGBqlWrBjoMnyUmJlKyZMlAhyE+ONXfLTNb4JyrezbXy9fdnXfeeSfdunWjbdu2fP755wBK0ERERCQo5KuMJDU1lZEjR/L1118zduxY/vOf/3D55ZcrMRMREZGgk29a0saOHUulSpX44osveOihhzAzIiIilKCJiIhIUMrTSdpff/3FK6+8wuHDh7n88suZPHky3333Hddff32287SIiEjeFWrjsSX4+ePvVJ5M0jZt2kSPHj2oWrUq27ZtIyUlhWuvvfa0sxuLiEj+UbRoUfbs2aNETXKMc449e/ZQtGjRHL1unurrW7BgAZGRkWzbto2yZcuyatUqLrrookCHJSIiQaRChQps2bKFXbt2BToUnxw6dCjH//GXnFe0aFEqVKiQo9f0a5JmZm2AgUBBYJhz7vUTjhcBRgB1gD1AZ+fcxjO5h3OO7777jtjYWFavXs3EiRNp0KDBcZPQiYiIHBMWFpbtWpfBJD4+/rQrDkje5rfuTjMrCLwHtAWqAV3NrNoJxe4F/nbORQIDgFhfr3/kyBGSkpIyFjm/6667WLduXcZEcyIiIiKhzJ9j0uoBa51z651zqcBo4MR1G24GPvW+Hw+0sGxG9B91kLhwKlWqVOHjjz8mMjKSJUuW0K1bNwoXLpzjH0JEREQkEPzZ3Vke2JxpewtQ/3RlnHNHzGw/UBY47RL3K3YdpeCGhXz1+ecZy1foSU0RERHJa0LiwQEzewB4wLt5mLW/L2/UqFEgQ5KzdyFZJOES1FR3oU31F7pUd6Et6mxP9GeSthXIvEJ5Be++U5XZYmaFgAvwPEBwHOfcEGAIgJnNP9s1sCTwVH+hS3UX2lR/oUt1F9rMbP7ZnuvPMWnzgMpmFm5mhYEuwJQTykwB7vK+7wj86DRxjYiIiIj/WtK8Y8x6AtPxTMHxsXNuhZm9DMx3zk0BPgI+M7O1wF48iZyIiIhIvufXMWnOuW+Ab07Y92Km94eA287wskNyIDQJHNVf6FLdhTbVX+hS3YW2s64/U++iiIiISPDJk2t3ioiIiIS6oE3SzKyNmf1hZmvN7OlTHC9iZmO8x383s4oBCFNOwYe662VmK81sqZn9YGZXBCJOObXs6i9TuVvNzJmZnjoLIr7Un5l18v4MrjCzkbkdo5yaD787LzezmWa2yPv784ZAxCknM7OPzWynmS0/zXEzs0Heul1qZlf7ct2gTNL8vaSU+I+PdbcIqOucq4VnpYm43I1STsfH+sPMSgKPAb/nboSSFV/qz8wqA88AjZxz1YHHcztOOZmPP3vPA2Odc1fhedDu/dyNUrIwHGiTxfG2QGXv6wHgA18uGpRJGn5aUkpyRbZ155yb6ZxL8W7OwTOHngQHX372AP6L5z9Gh3IzOMmWL/V3P/Cec+5vAOfczlyOUU7Nl7pzwPne9xcA23IxPsmCc+5nPLNUnM7NwAjnMQcoZWb/yO66wZqknWpJqfKnK+OcOwIcW1JKAsuXusvsXuBbv0YkZyLb+vM201/mnJuam4GJT3z5+asCVDGz2WY2x8yy+t+/5B5f6q4f8C8z24Jn5oRHcic0yQFn+m8jECLLQkneZGb/AuoCTQMdi/jGzAoAbwF3BzgUOXuF8HS5xOBpxf7ZzGo65/YFMijxSVdguHPuTTNrgGee0RrOuaOBDkz8I1hb0s5kSSmyWlJKcp0vdYeZXQ88B9zknDucS7FJ9rKrv5JADSDezDYC1wJT9PBA0PDl528LMMU5l+ac2wCsxpO0SWD5Unf3AmMBnHO/AUXxrOspwc+nfxtPFKxJmpaUCl3Z1p2ZXQUMxpOgaTxMcMmy/pxz+51zFzrnKjrnKuIZU3iTc+6s16aTHOXL787JeFrRMLML8XR/rs/FGOXUfKm7P4EWAGZWFU+StitXo5SzNQXo5n3K81pgv3Nue3YnBWV3p5aUCl0+1t0bQAlgnPdZjz+dczcFLGjJ4GP9SZDysf6mA63MbCWQDvRxzqkXIsB8rLvewFAzewLPQwR3q3EiOJjZKDz/+bnQO2bwJSAMwDn3IZ4xhDcAa4EU4B6frqv6FREREQk+wdrdKSIiIpKvKUkTERERCUJK0kRERESCkJI0ERERkSCkJE1EREQkCClJE5EcZ2bpZrY406tiFmWTcuB+w81sg/deC72zsZ/pNYYdW9DazJ494div5xqj9zrHvpflZvaVmZXKpnxtM7shJ+4tIqFHU3CISI4zsyTnXImcLpvFNYYDXzvnxptZK6C/c67WOVzvnGPK7rpm9imw2jn3ahbl7wbqOud65nQsIhL81JImIn5nZiXM7AdvK9cyM7v5FGX+YWY/Z2ppus67v5WZ/eY9d5yZZZc8/QxEes/t5b3WcjN73LuvuJlNNbMl3v2dvfvjzayumb0OnOeN4wvvsSTvn6PN7MZMMQ83s45mVtDM3jCzeWa21Mwe9OFr+Q3vAstmVs/7GReZ2a9mFuWddf5loLM3ls7e2D82s7nesid9jyKSdwTligMiEvLOM7PF3vcbgNuADs65A96liOaY2ZQTZku/HZjunHvVzAoCxbxlnweud84lm1lfoBee5OV02gPLzKwOnlm96wMG/G5mPwERwDbn3I0AZnZB5pOdc0+bWU/nXO1TXHsM0AmY6k2iWgA98KypuN85d42ZFQFmm9kM79qYJ/F+vhZ4Vk4BWAVc5511/nrgf865W83sRTK1pJnZ//Asgdfd21U618y+d84lZ/F9iEiIUpImIv5wMHOSY2ZhwP/MrAlwFE8L0sXAjkznzAM+9pad7JxbbGZNgWp4kh6AwnhaoE7lDTN7Hs9ahvfiSYImHUtgzGwicB0wDXjTzGLxdJHOOoPP9S0w0JuItQF+ds4d9Hax1jKzjt5yF+BZtPzEJO1Y8loeSAC+y1T+UzOrjGe5n7DT3L8VcJOZPendLgpc7r2WiOQxStJEJDfcAZQD6jjn0sxsI54EI4Nz7mdvEncjMNzM3gL+Br5zznX14R59nHPjj22YWYtTFXLOrTazq/Gso/eKmf3gnMuqZS7zuYfMLB5oDXQGRh+7HfCIc256Npc46JyrbWbF8KzR+DAwCPgvMNM518H7kEX8ac434Fbn3B++xCsioU1j0kQkN1wA7PQmaM2AK04sYGZXAH8554YCw4CrgTlAIzM7NsasuJlV8fGes4BbzKyYmRUHOgCzzOxSIMU59znwhvc+J0rztuidyhg83ajHWuXAk3D1OHaOmVXx3vOUnHMpwKNAbzMrhOf72eo9fHemoolAyUzb04FHzNusaGZXne4eIhL6lKSJSG74AqhrZsuAbnjGYJ0oBlhiZovwtFINdM7twpO0jDKzpXi6Oq/05YbOuYXAcGAu8DswzDm3CKiJZyzXYuAl4JVTnD4EWHrswYETzACaAt8751K9+4YBK4GFZrYcGEw2PRXeWJYCXYE44DXvZ8983kyg2rEHB/C0uIV5Y1vh3RaRPEpTcIiIiIgEIbWkiYiIiAQhJWkiIiIiQUhJmoiIiEgQUpImIiIiEoSUpImIiIgEISVpIiIiIkFISZqIiIhIEFKSJiIiIhKE/g8zrfFFythDCgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "roc_data['VAEGMM'] = {'scores': od_preds['data']['instance_score'], 'labels': y_outlier}\n",
    "plot_roc(roc_data)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}