PFN_DCTR_Hunting.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"0\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "# energyflow imports\n",
    "import energyflow as ef\n",
    "from energyflow.archs import PFN\n",
    "from energyflow.utils import data_split, to_categorical\n",
    "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Utilities"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def pad_events(events, val=0, max_length=None):\n",
    "    event_lengths = [event.shape[0] for event in events]\n",
    "    if max_length is None:\n",
    "        max_length = max(event_lengths)\n",
    "    return np.asarray([np.vstack((event, val*np.ones((max_length - ev_len, event.shape[1]))))\n",
    "                     for event,ev_len in zip(events, event_lengths)])\n",
    "\n",
    "# events: a numpy object array of numpy arrays of events\n",
    "def pad_events_generator(events, targets,  batch_size, weights = None):\n",
    "    num_events = len(events)\n",
    "    \n",
    "#     events = np.asarray([event if len(event) >= initial_pad else \n",
    "#                              np.vstack((event, np.zeros((initial_pad - len(event), event.shape[1]))))\n",
    "#                              for event in events])\n",
    "    \n",
    "    while True:\n",
    "        start = 0\n",
    "        while start < num_events:\n",
    "            end = min(start + batch_size, num_events)\n",
    "            if weights is None:\n",
    "                yield pad_events(events[start:end]), targets[start:end]\n",
    "            else:\n",
    "                yield pad_events(events[start:end]), targets[start:end], weights[start:end]\n",
    "            start = end"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "# base_dir = '/data0/users/aandreassen/DCTRHunting/'\n",
    "# sample0 = np.load(base_dir + 'herwig_full_event_with_mass_preprocessed.npy', allow_pickle=True)\n",
    "# sample1 = np.load(base_dir + 'pythia_full_event_with_mass_preprocessed.npy', allow_pickle=True)\n",
    "# signal  = np.load(base_dir + 'signal_full_event_with_mass_preprocessed.npy', allow_pickle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# print(max([len(s) for s in sample0]))\n",
    "# print(max([len(s) for s in sample1]))\n",
    "# print(max([len(s) for s in signal]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "#sample0_padded = pad_events(sample0, max_length = 551)\n",
    "#sample1_padded = pad_events(sample1, max_length = 551)\n",
    "#signal_padded = pad_events(signal, max_length = 551)\n",
    "\n",
    "#np.save(base_dir + 'herwig_full_event_with_mass_preprocessed_padded.npy', sample0_padded)\n",
    "#np.save(base_dir + 'pythia_full_event_with_mass_preprocessed_padded.npy', sample1_padded)\n",
    "#np.save(base_dir + 'signal_full_event_with_mass_preprocessed_padded.npy', signal_padded)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "base_dir = '/data0/users/aandreassen/DCTRHunting/'\n",
    "sample0 = np.load(base_dir + 'herwig_leading_two_jets_with_mass_preprocessed_padded.npy', allow_pickle=True)\n",
    "sample1 = np.load(base_dir + 'pythia_leading_two_jets_with_mass_preprocessed_padded.npy', allow_pickle=True)[:990000]\n",
    "signal  = np.load(base_dir + 'signal_leading_two_jets_with_mass_preprocessed_padded.npy', allow_pickle=True)[:10000]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "SR_low  = 3250/1000\n",
    "SR_high = 3750/1000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [],
   "source": [
    "sample0_SR = [event for event in sample0 if event[0,-1]>SR_low and event[0,-1]<SR_high]\n",
    "sample0_BR = [event for event in sample0 if event[0,-1]<SR_low or  event[0,-1]>SR_high]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "sample1_SR = [event for event in sample1 if event[0,-1]>SR_low and event[0,-1]<SR_high]\n",
    "sample1_BR = [event for event in sample1 if event[0,-1]<SR_low or  event[0,-1]>SR_high]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "signal_SR = [event for event in signal if event[0,-1]>SR_low and event[0,-1]<SR_high]\n",
    "signal_BR = [event for event in signal if event[0,-1]<SR_low or  event[0,-1]>SR_high]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.0547044069277248"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(signal_SR)/len(sample1_SR)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "y0 = np.zeros(len(sample0_BR))\n",
    "y1 = np.ones(len(sample1_BR))\n",
    "ys = np.ones(len(signal_BR))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = np.concatenate([sample0_BR, sample1_BR, signal_BR], axis=0)\n",
    "y = np.concatenate([y0, y1, ys], axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "Y = to_categorical(y, num_classes=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "# do train/val/test split \n",
    "(X_train, X_val, X_test,\n",
    " Y_train, Y_val, Y_test) = data_split(X, Y, val=0.1, test=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_train = len(X_train)\n",
    "n_val = len(X_val)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SR"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "y0_SR = np.zeros(len(sample0_SR))\n",
    "y1_SR = np.ones(len(sample1_SR))\n",
    "ys_SR = np.ones(len(signal_SR))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_SR = np.concatenate([sample0_SR, sample1_SR, signal_SR], axis=0)\n",
    "y_SR = np.concatenate([y0_SR, y1_SR, ys_SR], axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "Y_SR = to_categorical(y_SR, num_classes=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [],
   "source": [
    "# do train/val/test split \n",
    "(X_SR_train, X_SR_val, X_SR_test,\n",
    " Y_SR_train, Y_SR_val, Y_SR_test) = data_split(X_SR, Y_SR, val=0.1, test=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_SR_train = len(X_SR_train)\n",
    "n_SR_val = len(X_SR_val)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Build Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_2\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input (InputLayer)              (None, None, 4)      0                                            \n",
      "__________________________________________________________________________________________________\n",
      "tdist_0 (TimeDistributed)       (None, None, 100)    500         input[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "activation_8 (Activation)       (None, None, 100)    0           tdist_0[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "tdist_1 (TimeDistributed)       (None, None, 100)    10100       activation_8[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "activation_9 (Activation)       (None, None, 100)    0           tdist_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "tdist_2 (TimeDistributed)       (None, None, 128)    12928       activation_9[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "mask (Lambda)                   (None, None)         0           input[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "activation_10 (Activation)      (None, None, 128)    0           tdist_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "sum (Dot)                       (None, 128)          0           mask[0][0]                       \n",
      "                                                                 activation_10[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_0 (Dense)                 (None, 100)          12900       sum[0][0]                        \n",
      "__________________________________________________________________________________________________\n",
      "activation_11 (Activation)      (None, 100)          0           dense_0[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 100)          10100       activation_11[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "activation_12 (Activation)      (None, 100)          0           dense_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_2 (Dense)                 (None, 100)          10100       activation_12[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "activation_13 (Activation)      (None, 100)          0           dense_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "output (Dense)                  (None, 2)            202         activation_13[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "activation_14 (Activation)      (None, 2)            0           output[0][0]                     \n",
      "==================================================================================================\n",
      "Total params: 56,830\n",
      "Trainable params: 56,830\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "Phi_sizes, F_sizes = (100, 100, 128), (100, 100, 100)\n",
    "# build architecture\n",
    "pfn = PFN(input_dim=4, Phi_sizes=Phi_sizes, F_sizes=F_sizes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 1000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [],
   "source": [
    "earlystopping = EarlyStopping(patience=10, restore_best_weights=True)\n",
    "checkpoint = ModelCheckpoint('./saved_models/pythia_pythia-with-signal.model', save_best_only=True)\n",
    "\n",
    "callbacks = [earlystopping, checkpoint]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 1518229 samples, validate on 168692 samples\n",
      "Epoch 1/20\n",
      "    192/1518229 [..............................] - ETA: 4:24:44 - loss: 90.5241 - acc: 0.5312 "
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/aandreassen/python-envs/py36/lib/python3.6/site-packages/keras/callbacks/callbacks.py:95: RuntimeWarning: Method (on_train_batch_end) is slow compared to the batch update (0.128360). Check your callbacks.\n",
      "  % (hook_name, delta_t_median), RuntimeWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  34112/1518229 [..............................] - ETA: 18:57 - loss: 2.6969 - acc: 0.5132"
     ]
    }
   ],
   "source": [
    "# train model\n",
    "pfn.fit(X_train, Y_train,\n",
    "                  epochs=20,\n",
    "                  validation_data=(X_val, Y_val),\n",
    "                  verbose=1, \n",
    "                  callbacks = callbacks)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Get Weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "168/167 [==============================] - 4s 21ms/step\n"
     ]
    }
   ],
   "source": [
    "train_preds = pfn.predict_generator(pad_events_generator(X_SR_train, Y_SR_train, batch_size),\n",
    "                  steps=n_SR_train/batch_size,\n",
    "                  verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "19/18 [==============================] - 0s 23ms/step\n"
     ]
    }
   ],
   "source": [
    "val_preds = pfn.predict_generator(pad_events_generator(X_SR_val, Y_SR_val, batch_size),\n",
    "                  steps=n_SR_val/batch_size,\n",
    "                  verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "weights_SR_train = train_preds[:,1]/train_preds[:,0]\n",
    "weights_SR_val = val_preds[:,1]/val_preds[:,0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "for i,(y, _) in enumerate(Y_SR_train):\n",
    "  # Don't reweigth elements from class 1 y==0 <-> target ==[0, 1]\n",
    "  if y == 0:\n",
    "    weights_SR_train[i] = 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "for i,(y, _) in enumerate(Y_SR_val):\n",
    "  # Don't reweigth elements from class 1 y==0 <-> target ==[0, 1]\n",
    "  if y == 0:\n",
    "    weights_SR_val[i] = 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_3\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input (InputLayer)              (None, None, 4)      0                                            \n",
      "__________________________________________________________________________________________________\n",
      "tdist_0 (TimeDistributed)       (None, None, 100)    500         input[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "activation_15 (Activation)      (None, None, 100)    0           tdist_0[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "tdist_1 (TimeDistributed)       (None, None, 100)    10100       activation_15[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "activation_16 (Activation)      (None, None, 100)    0           tdist_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "tdist_2 (TimeDistributed)       (None, None, 128)    12928       activation_16[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "mask (Lambda)                   (None, None)         0           input[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "activation_17 (Activation)      (None, None, 128)    0           tdist_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "sum (Dot)                       (None, 128)          0           mask[0][0]                       \n",
      "                                                                 activation_17[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_0 (Dense)                 (None, 100)          12900       sum[0][0]                        \n",
      "__________________________________________________________________________________________________\n",
      "activation_18 (Activation)      (None, 100)          0           dense_0[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 100)          10100       activation_18[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "activation_19 (Activation)      (None, 100)          0           dense_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_2 (Dense)                 (None, 100)          10100       activation_19[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "activation_20 (Activation)      (None, 100)          0           dense_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "output (Dense)                  (None, 2)            202         activation_20[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "activation_21 (Activation)      (None, 2)            0           output[0][0]                     \n",
      "==================================================================================================\n",
      "Total params: 56,830\n",
      "Trainable params: 56,830\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "Phi_sizes, F_sizes = (100, 100, 128), (100, 100, 100)\n",
    "# build architecture\n",
    "pfn2 = PFN(input_dim=4, Phi_sizes=Phi_sizes, F_sizes=F_sizes)\n",
    "\n",
    "earlystopping = EarlyStopping(patience=10, restore_best_weights=True)\n",
    "checkpoint = ModelCheckpoint('pythia_pythia-with-signal_SR.model', save_best_only=True)\n",
    "\n",
    "callbacks = [earlystopping, checkpoint]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 1000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/5\n",
      "168/167 [==============================] - 6s 34ms/step - loss: 12.3268 - acc: 0.4991 - val_loss: 1.9443 - val_acc: 0.5154\n",
      "Epoch 2/5\n",
      "168/167 [==============================] - 5s 29ms/step - loss: 1.4450 - acc: 0.5001 - val_loss: 1.2400 - val_acc: 0.5177\n",
      "Epoch 3/5\n",
      "168/167 [==============================] - 5s 30ms/step - loss: 1.0286 - acc: 0.4988 - val_loss: 1.0986 - val_acc: 0.5166\n",
      "Epoch 4/5\n",
      "168/167 [==============================] - 5s 30ms/step - loss: 0.9330 - acc: 0.5000 - val_loss: 0.8247 - val_acc: 0.4857\n",
      "Epoch 5/5\n",
      "168/167 [==============================] - 5s 30ms/step - loss: 1.0040 - acc: 0.4990 - val_loss: 0.8383 - val_acc: 0.4854\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.callbacks.History at 0x7f816aaaf1d0>"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# train model\n",
    "pfn2.fit_generator(pad_events_generator(X_SR_train, Y_SR_train, batch_size, weights_SR_train),\n",
    "                  epochs=5,\n",
    "                  steps_per_epoch=n_SR_train/batch_size,\n",
    "                  validation_data=pad_events_generator(X_SR_val, Y_SR_val, batch_size, weights_SR_val),\n",
    "                  validation_steps=n_SR_val/batch_size,\n",
    "                  verbose=1, \n",
    "                  callbacks = callbacks)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ROC Curves"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import roc_curve\n",
    "from matplotlib import pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "19/18 [==============================] - 0s 22ms/step\n"
     ]
    }
   ],
   "source": [
    "final_preds = pfn2.predict_generator(pad_events_generator(X_SR_val, Y_SR_val, batch_size),\n",
    "                  steps=n_SR_val/batch_size,\n",
    "                  verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "roc = roc_curve(Y_SR_val.argmax(axis = -1), final_preds[:,1], weights_SR_val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "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": [
    "plt.plot(roc[0], roc[1], label='pythia vs pythia-with-signal', color='red')\n",
    "plt.plot([0,1],[0,1], label='random', ls=\"--\", color='k')\n",
    "plt.xlabel(\"True Positive\")\n",
    "plt.ylabel(\"False Negative\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10/10 [==============================] - 0s 21ms/step\n",
      "10/10 [==============================] - 0s 26ms/step\n"
     ]
    }
   ],
   "source": [
    "y_signal = np.ones(len(signal))\n",
    "signal_preds = pfn2.predict_generator(pad_events_generator(signal, to_categorical(y_signal, 2), batch_size),\n",
    "                  steps=len(signal)/batch_size,\n",
    "                  verbose=1)\n",
    "\n",
    "signal_weights = pfn.predict_generator(pad_events_generator(signal, to_categorical(y_signal, 2), batch_size),\n",
    "                  steps=len(signal)/batch_size,\n",
    "                  verbose=1)\n",
    "\n",
    "signal_weights = signal_weights[:,1]/signal_weights[:,0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "990/990 [==============================] - 19s 19ms/step\n",
      "990/990 [==============================] - 20s 20ms/step\n"
     ]
    }
   ],
   "source": [
    "y_pythia = np.zeros(len(pythia1))\n",
    "pythia_preds = pfn2.predict_generator(pad_events_generator(pythia1, to_categorical(y_pythia, 2), batch_size),\n",
    "                  steps=len(pythia1)/batch_size,\n",
    "                  verbose=1)\n",
    "\n",
    "pythia_weights = pfn.predict_generator(pad_events_generator(pythia1, to_categorical(y_pythia, 2), batch_size),\n",
    "                  steps=len(pythia1)/batch_size,\n",
    "                  verbose=1)\n",
    "\n",
    "pythia_weights = pythia_weights[:,1]/pythia_weights[:,0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "990000"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(pythia1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_roc = np.concatenate([y_signal, y_pythia])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds_roc = np.concatenate([signal_preds, pythia_preds])[:,1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "weights_roc = np.concatenate([signal_weights, pythia_weights])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "sig_roc = roc_curve(y_roc, preds_roc, sample_weight = weights_roc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "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": [
    "plt.plot(sig_roc[0], sig_roc[1], label='signal vs pythia', color='red')\n",
    "plt.plot([0,1],[0,1], label='random', ls=\"--\", color='k')\n",
    "plt.xlabel(\"True Positive\")\n",
    "plt.ylabel(\"False Negative\")\n",
    "plt.legend(loc='best')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/venv/lib/python3.6/site-packages/ipykernel_launcher.py:1: MatplotlibDeprecationWarning: \n",
      "The 'normed' kwarg was deprecated in Matplotlib 2.1 and will be removed in 3.1. Use 'density' instead.\n",
      "  \"\"\"Entry point for launching an IPython kernel.\n",
      "/opt/venv/lib/python3.6/site-packages/ipykernel_launcher.py:2: MatplotlibDeprecationWarning: \n",
      "The 'normed' kwarg was deprecated in Matplotlib 2.1 and will be removed in 3.1. Use 'density' instead.\n",
      "  \n"
     ]
    },
    {
     "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": [
    "plt.hist(signal_preds[:,1], bins = np.linspace(0,1,101), alpha = 0.5, normed=True, label = 'signal', weights=signal_weights)\n",
    "plt.hist(pythia_preds[:,1], bins = np.linspace(0,1,101), alpha = 0.5, normed=True, label='pythia', weights=pythia_weights)\n",
    "plt.legend(loc='best')\n",
    "plt.xlabel(\"P_{Signal+Pythia}\")\n",
    "plt.ylabel(\"Normalized probability\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/venv/lib/python3.6/site-packages/ipykernel_launcher.py:1: RuntimeWarning: divide by zero encountered in true_divide\n",
      "  \"\"\"Entry point for launching an IPython kernel.\n",
      "/opt/venv/lib/python3.6/site-packages/ipykernel_launcher.py:1: RuntimeWarning: invalid value encountered in true_divide\n",
      "  \"\"\"Entry point for launching an IPython kernel.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f8168a18080>]"
      ]
     },
     "execution_count": 69,
     "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": [
    "plt.plot((roc[1]-roc[0])/roc[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "roc_now = roc_curve(Y_SR_val.argmax(axis = -1), val_preds[:,1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "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": [
    "plt.plot(roc_now[0], roc_now[1], label='pythia vs herwig', color='red')\n",
    "plt.plot([0,1],[0,1], label='random', ls=\"--\", color='k')\n",
    "plt.xlabel(\"True Positive\")\n",
    "plt.ylabel(\"False Negative\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "ename": "ValueError",
     "evalue": "operands could not be broadcast together with shapes (2,) (16504,) ",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-40-5fe7cc76e731>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mroc_now\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0mroc\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mValueError\u001b[0m: operands could not be broadcast together with shapes (2,) (16504,) "
     ]
    }
   ],
   "source": [
    "roc_now[1]-roc[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "roc_now"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "val_preds[:,1].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}