Keras Ch 6 Sequence with Convnets.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1D Convnet on IMDB Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "loading data\n",
      "Pad Sequences (samples x time)\n",
      "('x_train.shape:', (25000, 500))\n",
      "('x_test.shape:', (25000, 500))\n"
     ]
    }
   ],
   "source": [
    "from keras.datasets import imdb\n",
    "from keras.preprocessing import sequence\n",
    "\n",
    "max_features = 10000\n",
    "max_len = 500\n",
    "\n",
    "print('loading data')\n",
    "(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=max_features)\n",
    "\n",
    "print('Pad Sequences (samples x time)')\n",
    "x_train = sequence.pad_sequences(x_train, maxlen=max_len)\n",
    "x_test = sequence.pad_sequences(x_test, maxlen=max_len)\n",
    "print('x_train.shape:', x_train.shape)\n",
    "print('x_test.shape:', x_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 500, 128)          1280000   \n",
      "_________________________________________________________________\n",
      "conv1d_1 (Conv1D)            (None, 494, 32)           28704     \n",
      "_________________________________________________________________\n",
      "max_pooling1d_1 (MaxPooling1 (None, 98, 32)            0         \n",
      "_________________________________________________________________\n",
      "conv1d_2 (Conv1D)            (None, 85, 32)            14368     \n",
      "_________________________________________________________________\n",
      "global_max_pooling1d_1 (Glob (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 1)                 33        \n",
      "=================================================================\n",
      "Total params: 1,323,105\n",
      "Trainable params: 1,323,105\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "from keras import models\n",
    "from keras import layers\n",
    "from keras.optimizers import RMSprop\n",
    "\n",
    "from util import print_curves\n",
    "\n",
    "from IPython.display import SVG\n",
    "from keras.utils.vis_utils import model_to_dot\n",
    "\n",
    "\n",
    "def plot_model_jupyter(model):\n",
    "    SVG(model_to_dot(model).create(prog='dot', format='svg'))\n",
    "\n",
    "\n",
    "model = models.Sequential()\n",
    "model.add(layers.Embedding(max_features, 128, input_length=max_len))\n",
    "model.add(layers.Conv1D(32, 7, activation='relu'))\n",
    "model.add(layers.MaxPooling1D(5))\n",
    "model.add(layers.Conv1D(32, 14, activation='relu'))\n",
    "# default pool size 2\n",
    "model.add(layers.GlobalMaxPooling1D())\n",
    "model.add(layers.Dense(1, activation='sigmoid'))\n",
    "\n",
    "model.summary()\n",
    "plot_model_jupyter(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "plot_model_jupyter(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 20000 samples, validate on 5000 samples\n",
      "Epoch 1/10\n",
      "20000/20000 [==============================] - 3s 151us/step - loss: 0.4946 - acc: 0.7414 - val_loss: 0.3218 - val_acc: 0.8626\n",
      "Epoch 2/10\n",
      "20000/20000 [==============================] - 2s 77us/step - loss: 0.2538 - acc: 0.8973 - val_loss: 0.3102 - val_acc: 0.8682\n",
      "Epoch 3/10\n",
      "20000/20000 [==============================] - 2s 77us/step - loss: 0.1870 - acc: 0.9268 - val_loss: 0.3986 - val_acc: 0.8468\n",
      "Epoch 4/10\n",
      "20000/20000 [==============================] - 2s 78us/step - loss: 0.1380 - acc: 0.9484 - val_loss: 0.3312 - val_acc: 0.8782\n",
      "Epoch 5/10\n",
      "20000/20000 [==============================] - 2s 78us/step - loss: 0.0963 - acc: 0.9662 - val_loss: 0.3757 - val_acc: 0.8770\n",
      "Epoch 6/10\n",
      "20000/20000 [==============================] - 2s 77us/step - loss: 0.0632 - acc: 0.9802 - val_loss: 0.4155 - val_acc: 0.8770\n",
      "Epoch 7/10\n",
      "20000/20000 [==============================] - 2s 77us/step - loss: 0.0356 - acc: 0.9901 - val_loss: 0.5837 - val_acc: 0.8564\n",
      "Epoch 8/10\n",
      "20000/20000 [==============================] - 2s 78us/step - loss: 0.0192 - acc: 0.9956 - val_loss: 0.6197 - val_acc: 0.8588\n",
      "Epoch 9/10\n",
      "20000/20000 [==============================] - 2s 76us/step - loss: 0.0085 - acc: 0.9982 - val_loss: 0.6266 - val_acc: 0.8724\n",
      "Epoch 10/10\n",
      "20000/20000 [==============================] - 2s 79us/step - loss: 0.0045 - acc: 0.9990 - val_loss: 0.6757 - val_acc: 0.8706\n",
      "Max Val Acc: 0.8782\n",
      "Min Val Loss: 0.310200035405\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fed720c17d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fed725c7ad0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model.compile(optimizer=RMSprop(), loss='binary_crossentropy', metrics=['acc'])\n",
    "history = model.fit(\n",
    "    x_train,\n",
    "    y_train,\n",
    "    epochs=10,\n",
    "    batch_size=128,\n",
    "    validation_split=0.2)\n",
    "\n",
    "print_curves(history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "plot_model_jupyter(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/svg+xml": [
       "<svg height=\"483pt\" viewBox=\"0.00 0.00 359.00 483.00\" width=\"359pt\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g class=\"graph\" id=\"graph0\" transform=\"scale(1 1) rotate(0) translate(4 479)\">\n",
       "<title>G</title>\n",
       "<polygon fill=\"white\" points=\"-4,4 -4,-479 355,-479 355,4 -4,4\" stroke=\"none\"/>\n",
       "<!-- 140657803786320 -->\n",
       "<g class=\"node\" id=\"node1\"><title>140657803786320</title>\n",
       "<polygon fill=\"none\" points=\"52,-438.5 52,-474.5 299,-474.5 299,-438.5 52,-438.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"175.5\" y=\"-452.8\">embedding_1_input: InputLayer</text>\n",
       "</g>\n",
       "<!-- 140657803785872 -->\n",
       "<g class=\"node\" id=\"node2\"><title>140657803785872</title>\n",
       "<polygon fill=\"none\" points=\"74,-365.5 74,-401.5 277,-401.5 277,-365.5 74,-365.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"175.5\" y=\"-379.8\">embedding_1: Embedding</text>\n",
       "</g>\n",
       "<!-- 140657803786320&#45;&gt;140657803785872 -->\n",
       "<g class=\"edge\" id=\"edge1\"><title>140657803786320-&gt;140657803785872</title>\n",
       "<path d=\"M175.5,-438.313C175.5,-430.289 175.5,-420.547 175.5,-411.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"179,-411.529 175.5,-401.529 172,-411.529 179,-411.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140657803726032 -->\n",
       "<g class=\"node\" id=\"node3\"><title>140657803726032</title>\n",
       "<polygon fill=\"none\" points=\"100,-292.5 100,-328.5 251,-328.5 251,-292.5 100,-292.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"175.5\" y=\"-306.8\">conv1d_1: Conv1D</text>\n",
       "</g>\n",
       "<!-- 140657803785872&#45;&gt;140657803726032 -->\n",
       "<g class=\"edge\" id=\"edge2\"><title>140657803785872-&gt;140657803726032</title>\n",
       "<path d=\"M175.5,-365.313C175.5,-357.289 175.5,-347.547 175.5,-338.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"179,-338.529 175.5,-328.529 172,-338.529 179,-338.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140657803786640 -->\n",
       "<g class=\"node\" id=\"node4\"><title>140657803786640</title>\n",
       "<polygon fill=\"none\" points=\"48.5,-219.5 48.5,-255.5 302.5,-255.5 302.5,-219.5 48.5,-219.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"175.5\" y=\"-233.8\">max_pooling1d_1: MaxPooling1D</text>\n",
       "</g>\n",
       "<!-- 140657803726032&#45;&gt;140657803786640 -->\n",
       "<g class=\"edge\" id=\"edge3\"><title>140657803726032-&gt;140657803786640</title>\n",
       "<path d=\"M175.5,-292.313C175.5,-284.289 175.5,-274.547 175.5,-265.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"179,-265.529 175.5,-255.529 172,-265.529 179,-265.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140657803786960 -->\n",
       "<g class=\"node\" id=\"node5\"><title>140657803786960</title>\n",
       "<polygon fill=\"none\" points=\"100,-146.5 100,-182.5 251,-182.5 251,-146.5 100,-146.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"175.5\" y=\"-160.8\">conv1d_2: Conv1D</text>\n",
       "</g>\n",
       "<!-- 140657803786640&#45;&gt;140657803786960 -->\n",
       "<g class=\"edge\" id=\"edge4\"><title>140657803786640-&gt;140657803786960</title>\n",
       "<path d=\"M175.5,-219.313C175.5,-211.289 175.5,-201.547 175.5,-192.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"179,-192.529 175.5,-182.529 172,-192.529 179,-192.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140657803519248 -->\n",
       "<g class=\"node\" id=\"node6\"><title>140657803519248</title>\n",
       "<polygon fill=\"none\" points=\"0,-73.5 0,-109.5 351,-109.5 351,-73.5 0,-73.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"175.5\" y=\"-87.8\">global_max_pooling1d_1: GlobalMaxPooling1D</text>\n",
       "</g>\n",
       "<!-- 140657803786960&#45;&gt;140657803519248 -->\n",
       "<g class=\"edge\" id=\"edge5\"><title>140657803786960-&gt;140657803519248</title>\n",
       "<path d=\"M175.5,-146.313C175.5,-138.289 175.5,-128.547 175.5,-119.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"179,-119.529 175.5,-109.529 172,-119.529 179,-119.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140657803725392 -->\n",
       "<g class=\"node\" id=\"node7\"><title>140657803725392</title>\n",
       "<polygon fill=\"none\" points=\"111.5,-0.5 111.5,-36.5 239.5,-36.5 239.5,-0.5 111.5,-0.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"175.5\" y=\"-14.8\">dense_1: Dense</text>\n",
       "</g>\n",
       "<!-- 140657803519248&#45;&gt;140657803725392 -->\n",
       "<g class=\"edge\" id=\"edge6\"><title>140657803519248-&gt;140657803725392</title>\n",
       "<path d=\"M175.5,-73.3129C175.5,-65.2895 175.5,-55.5475 175.5,-46.5691\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"179,-46.5288 175.5,-36.5288 172,-46.5289 179,-46.5288\" stroke=\"black\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>"
      ],
      "text/plain": [
       "<IPython.core.display.SVG object>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "SVG(model_to_dot(model).create(prog='dot', format='svg'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "25000/25000 [==============================] - 1s 50us/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.74228758404254913, 0.85128000000000004]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(x_test, y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Try on Temperature Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import pandas as pd\n",
    "\n",
    "data_dir = './Downloads/'\n",
    "fname = os.path.join(data_dir, 'jena_climate_2009_2016.csv')\n",
    "df = pd.read_csv(fname)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_time = df\n",
    "df = df.drop('Date Time', axis=1)\n",
    "mean = df.mean(0)\n",
    "std = df.std(0)\n",
    "# standardize data\n",
    "df = df.sub(mean, axis=1).div(std, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "\n",
    "def generator(data, lookback, delay, min_index, max_index,\n",
    "              shuffle=False, batch_size=128, step=6,\n",
    "              reverse=False):\n",
    "    if max_index is None:\n",
    "        max_index = len(data) - delay - 1\n",
    "    i = min_index + lookback\n",
    "\n",
    "    # create offset so we don't see the same 1/step th of data in non-shuffled\n",
    "    # scenario\n",
    "    offset = 1\n",
    "\n",
    "    while True:\n",
    "        if shuffle:\n",
    "            rows = np.random.randint(\n",
    "                min_index + lookback, max_index, size=batch_size)\n",
    "        else:\n",
    "            if i + batch_size >= max_index:\n",
    "                # won't this reset the generator to the same place, esentially not giving us 1/6 of the data?\n",
    "                # might be somewhat trivial since we're taking hourly samples,\n",
    "                # so data only has changed by an hour and we're reading across\n",
    "                # 8 years total dataset\n",
    "                i = min_index + lookback + offset\n",
    "                offset += 1\n",
    "                if offset == step:\n",
    "                    offset = 0\n",
    "            rows = np.arange(i, min(i + batch_size, max_index))\n",
    "            i += len(rows)\n",
    "        samples = np.zeros((len(rows),\n",
    "                            lookback // step,\n",
    "                            data.shape[-1]))\n",
    "        targets = np.zeros((len(rows),))\n",
    "        for j, row in enumerate(rows):\n",
    "            indices = range(rows[j] - lookback, rows[j], step)\n",
    "            samples[j] = data[indices]\n",
    "            targets[j] = data[rows[j] + delay][1]\n",
    "        if reverse:\n",
    "            yield samples[:, ::-1, :], targets\n",
    "        else:\n",
    "            yield samples, targets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "lookback = 1440\n",
    "step = 6\n",
    "delay = 144\n",
    "batch_size = 128\n",
    "num_features = df.shape[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_gen = generator(\n",
    "    df.values,\n",
    "    lookback=lookback,\n",
    "    delay=delay,\n",
    "    min_index=0,\n",
    "    max_index=200000,\n",
    "    shuffle=True,\n",
    "    step=step,\n",
    "    batch_size=batch_size)\n",
    "\n",
    "val_gen = generator(\n",
    "    df.values,\n",
    "    lookback=lookback,\n",
    "    delay=delay,\n",
    "    min_index=200001,\n",
    "    max_index=300000,\n",
    "    step=step,\n",
    "    batch_size=batch_size)\n",
    "\n",
    "test_gen = generator(\n",
    "    df.values,\n",
    "    lookback=lookback,\n",
    "    delay=delay,\n",
    "    min_index=300001,\n",
    "    max_index=None,\n",
    "    step=step,\n",
    "    batch_size=batch_size)\n",
    "\n",
    "# how many train steps to see entire dataset\n",
    "train_steps = (200000 - lookback) // batch_size\n",
    "\n",
    "# how many val steps to see entire dataset\n",
    "val_steps = (300000 - 200001 - lookback) // batch_size\n",
    "\n",
    "# how many test steps to see entire dataset\n",
    "test_steps = (len(df) - 300001 - lookback) // batch_size\n",
    "\n",
    "# need to check if we're processing the cursor at all, i.e. are we only\n",
    "# seeing 1/6 of the data since we're sampling every 6 hours\n",
    "# Actually, we're shuffling, so that should take care of it, but only for\n",
    "# training, in theory you want to run through 6 times, offestting,\n",
    "# something like that"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "500/500 [==============================] - 8s 16ms/step - loss: 0.4446 - val_loss: 0.4487\n",
      "Epoch 2/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.3869 - val_loss: 0.4538\n",
      "Epoch 3/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.3632 - val_loss: 0.4645\n",
      "Epoch 4/20\n",
      "500/500 [==============================] - 7s 14ms/step - loss: 0.3475 - val_loss: 0.4659\n",
      "Epoch 5/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.3345 - val_loss: 0.4698\n",
      "Epoch 6/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.3265 - val_loss: 0.4705\n",
      "Epoch 7/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.3170 - val_loss: 0.4704\n",
      "Epoch 8/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.3112 - val_loss: 0.4805\n",
      "Epoch 9/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.3019 - val_loss: 0.4980\n",
      "Epoch 10/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.2970 - val_loss: 0.4859\n",
      "Epoch 11/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.2944 - val_loss: 0.5000\n",
      "Epoch 12/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.2888 - val_loss: 0.5216\n",
      "Epoch 13/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.2851 - val_loss: 0.4968\n",
      "Epoch 14/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.2810 - val_loss: 0.5114\n",
      "Epoch 15/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.2786 - val_loss: 0.5073\n",
      "Epoch 16/20\n",
      "500/500 [==============================] - 7s 14ms/step - loss: 0.2746 - val_loss: 0.5186\n",
      "Epoch 17/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.2711 - val_loss: 0.5132\n",
      "Epoch 18/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.2686 - val_loss: 0.5073\n",
      "Epoch 19/20\n",
      "500/500 [==============================] - 7s 15ms/step - loss: 0.2662 - val_loss: 0.5183\n",
      "Epoch 20/20\n",
      "500/500 [==============================] - 7s 14ms/step - loss: 0.2643 - val_loss: 0.5242\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras import layers\n",
    "from keras.optimizers import RMSprop\n",
    "\n",
    "model = Sequential()\n",
    "model.add(layers.Conv1D(32, 5, activation='relu',\n",
    "                        input_shape=(None, df.shape[-1])))\n",
    "model.add(layers.MaxPooling1D(3))\n",
    "model.add(layers.Conv1D(32, 5, activation='relu'))\n",
    "model.add(layers.MaxPooling1D(3))\n",
    "model.add(layers.Conv1D(32, 5, activation='relu'))\n",
    "model.add(layers.GlobalMaxPooling1D())\n",
    "model.add(layers.Dense(1))\n",
    "\n",
    "model.compile(optimizer=RMSprop(), loss='mae')\n",
    "\n",
    "history = model.fit_generator(train_gen, steps_per_epoch=500,\n",
    "                              epochs=20, validation_data=val_gen,\n",
    "                              validation_steps=val_steps)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Min Val Loss: 0.448724445784\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fed4006fe90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print_curves(history)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Still not as good as naive baseline but fast."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### What if we use 1D Convnet to pre-process / downsample data, then use RNN?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "lookback = 1440\n",
    "\n",
    "# Changed step to 3 so we look at every 30 minutes\n",
    "step = 3\n",
    "delay = 144\n",
    "batch_size = 128\n",
    "num_features = df.shape[1]\n",
    "\n",
    "train_gen = generator(\n",
    "    df.values,\n",
    "    lookback=lookback,\n",
    "    delay=delay,\n",
    "    min_index=0,\n",
    "    max_index=200000,\n",
    "    shuffle=True,\n",
    "    step=step,\n",
    "    batch_size=batch_size)\n",
    "\n",
    "val_gen = generator(\n",
    "    df.values,\n",
    "    lookback=lookback,\n",
    "    delay=delay,\n",
    "    min_index=200001,\n",
    "    max_index=300000,\n",
    "    step=step,\n",
    "    batch_size=batch_size)\n",
    "\n",
    "test_gen = generator(\n",
    "    df.values,\n",
    "    lookback=lookback,\n",
    "    delay=delay,\n",
    "    min_index=300001,\n",
    "    max_index=None,\n",
    "    step=step,\n",
    "    batch_size=batch_size)\n",
    "\n",
    "# how many train steps to see entire dataset\n",
    "train_steps = (200000 - lookback) // batch_size\n",
    "\n",
    "# how many val steps to see entire dataset\n",
    "val_steps = (300000 - 200001 - lookback) // batch_size\n",
    "\n",
    "# how many test steps to see entire dataset\n",
    "test_steps = (len(df) - 300001 - lookback) // batch_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv1d_6 (Conv1D)            (None, None, 32)          2272      \n",
      "_________________________________________________________________\n",
      "max_pooling1d_4 (MaxPooling1 (None, None, 32)          0         \n",
      "_________________________________________________________________\n",
      "conv1d_7 (Conv1D)            (None, None, 32)          5152      \n",
      "_________________________________________________________________\n",
      "gru_1 (GRU)                  (None, 32)                6240      \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 1)                 33        \n",
      "=================================================================\n",
      "Total params: 13,697\n",
      "Trainable params: 13,697\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras import layers\n",
    "from keras.optimizers import RMSprop\n",
    "\n",
    "model = Sequential()\n",
    "model.add(layers.Conv1D(32, 5, activation='relu',\n",
    "                        input_shape=(None, df.shape[-1])))\n",
    "model.add(layers.MaxPooling1D(3))\n",
    "model.add(layers.Conv1D(32, 5, activation='relu'))\n",
    "model.add(layers.GRU(32, dropout=0.1, recurrent_dropout=0.5))\n",
    "model.add(layers.Dense(1))\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "500/500 [==============================] - 142s 285ms/step - loss: 0.3508 - val_loss: 0.2979\n",
      "Epoch 2/20\n",
      "500/500 [==============================] - 143s 287ms/step - loss: 0.3190 - val_loss: 0.2899\n",
      "Epoch 3/20\n",
      "500/500 [==============================] - 140s 281ms/step - loss: 0.3036 - val_loss: 0.2855\n",
      "Epoch 4/20\n",
      "500/500 [==============================] - 143s 286ms/step - loss: 0.2940 - val_loss: 0.2864\n",
      "Epoch 5/20\n",
      "500/500 [==============================] - 146s 292ms/step - loss: 0.2857 - val_loss: 0.2821\n",
      "Epoch 6/20\n",
      "500/500 [==============================] - 137s 274ms/step - loss: 0.2815 - val_loss: 0.2889\n",
      "Epoch 7/20\n",
      "500/500 [==============================] - 142s 284ms/step - loss: 0.2742 - val_loss: 0.2872\n",
      "Epoch 8/20\n",
      "500/500 [==============================] - 142s 284ms/step - loss: 0.2670 - val_loss: 0.2966\n",
      "Epoch 9/20\n",
      "500/500 [==============================] - 136s 272ms/step - loss: 0.2627 - val_loss: 0.2904\n",
      "Epoch 10/20\n",
      "500/500 [==============================] - 138s 277ms/step - loss: 0.2600 - val_loss: 0.2941\n",
      "Epoch 11/20\n",
      "500/500 [==============================] - 142s 283ms/step - loss: 0.2564 - val_loss: 0.3097\n",
      "Epoch 12/20\n",
      "500/500 [==============================] - 140s 280ms/step - loss: 0.2508 - val_loss: 0.3016\n",
      "Epoch 13/20\n",
      "499/500 [============================>.] - ETA: 0s - loss: 0.2495"
     ]
    }
   ],
   "source": [
    "model.compile(optimizer=RMSprop(), loss='mae')\n",
    "\n",
    "history = model.fit_generator(train_gen, steps_per_epoch=500,\n",
    "                              epochs=20, validation_data=val_gen,\n",
    "                              validation_steps=val_steps)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print_curves(history)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This wasn't as good as regularized GRU alone, but was a lot faster. We also looked at twice as much data - didn't seem to help as much in this case but good strategy to know."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}