Notebooks/5. LSTM_Model.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "LSTM Model.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rc7DTpLs9wiz",
        "colab_type": "text"
      },
      "source": [
        "# Building an LSTM model to predict the price of the S&P 500\n",
        "\n",
        "- LSTM stands for Long Short Term Memory\n",
        "\n",
        "The LSTM cell contains an RNN cell inside (dense layer with a tanh activation function), but it has other attributes that allow it to remember patterns over a longer period of time. The LSTM contains a short-term state vector that is used from one time step to the next identical to the RNN cell.\n",
        "\n",
        "Where LSTM cells differ is their Long-Term State vector. This long term state vector undergoes a multipicative and additive operation at each time step.The long term state vector starts off with a \"forget gate\" which is simply a sigmoid activation on top of a dense layer which produces an output from 0 - 1. This means if the forget gate ourputs a value of 1 then the long term state vector remains unchanged. On the opposite end, if the output valuse of the dense layer is 0 the long term state vector is nullified. This helps adjust quickly to highly volatile environments. If we are in a steady uptrend and then we suddenly have a big drop in the market, the long term state vector will be erased so the model can readjust quickly.\n",
        "\n",
        "The next gate is the input gate. The input gate is much like the forget gate, except it decides if the short term state vector gets added to the long term vector or is erased. \n",
        "\n",
        "The last gate called the \"output gate\" decides what the next hidden state should be. It is important to know that this hidden state output is the prediction for the next time step. Here, the previous hidden state (previous cell output) amd current input are jammed together through sigmoid function. Then we take our newly modified long term state vector which has already passed through the forget and input gates, and apply a tanh function to it. Finally, multiple both the tanh output (long term state vector output) and the sigmoid function output (previous hidden state and new input at this time step) to what information the hidden state needs to carry to the next time step.\n",
        "\n",
        "</br>\n",
        "<b>To review, the Forget gate decides what is relevant to keep from prior steps. The input gate decides what information is relevant to add from the current step. The output gate determines what the next hidden state should be.</b>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "vidayERjaO5q"
      },
      "source": [
        "## Setup"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "gqWabzlJ63nL",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 50
        },
        "outputId": "78765880-4bb8-41ca-923c-955826ee4301"
      },
      "source": [
        "import numpy as np\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "import tensorflow as tf\n",
        "import seaborn as sns\n",
        "\n",
        "from sklearn.preprocessing import MinMaxScaler\n",
        "\n",
        "keras = tf.keras\n",
        "\n",
        "# set style of charts\n",
        "sns.set(style=\"darkgrid\")\n",
        "plt.rcParams['figure.figsize'] = [10, 10]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/statsmodels/tools/_testing.py:19: FutureWarning: pandas.util.testing is deprecated. Use the functions in the public API at pandas.testing instead.\n",
            "  import pandas.util.testing as tm\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "cg1hfKCPldZG",
        "colab": {}
      },
      "source": [
        "def plot_series(time, series, format=\"-\", start=0, end=None, label=None):\n",
        "    plt.plot(time[start:end], series[start:end], format, label=label)\n",
        "    plt.xlabel(\"Time\")\n",
        "    plt.ylabel(\"Value\")\n",
        "    if label:\n",
        "        plt.legend(fontsize=14)\n",
        "    plt.grid(True)\n",
        "    \n",
        "\n",
        "def sequential_window_dataset(series, window_size):\n",
        "    ds = tf.data.Dataset.from_tensor_slices(series)\n",
        "    ds = ds.window(window_size + 1, shift=window_size, drop_remainder=True)\n",
        "    ds = ds.flat_map(lambda window: window.batch(window_size + 1))\n",
        "    ds = ds.map(lambda window: (window[:-1], window[1:]))\n",
        "    return ds.batch(1).prefetch(1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "iL2DDjV3lel6",
        "colab": {}
      },
      "source": [
        "# Read in data\n",
        "spy = pd.read_csv('SPY.csv')\n",
        "\n",
        "# Convert series into datetime type\n",
        "spy['Date'] = pd.to_datetime(spy['Date'])\n",
        "\n",
        "# Save target series\n",
        "series = spy['Close']\n",
        "\n",
        "# Create train data set\n",
        "train_split_date = '2014-12-31'\n",
        "train_split_index = np.where(spy.Date == train_split_date)[0][0]\n",
        "x_train = spy.loc[spy['Date'] <= train_split_date]['Close']\n",
        "\n",
        "# Create test data set\n",
        "test_split_date = '2019-01-02'\n",
        "test_split_index = np.where(spy.Date == test_split_date)[0][0]\n",
        "x_test = spy.loc[spy['Date'] >= test_split_date]['Close']\n",
        "\n",
        "# Create valid data set\n",
        "valid_split_index = (train_split_index.max(),test_split_index.min())\n",
        "x_valid = spy.loc[(spy['Date'] < test_split_date) & (spy['Date'] > train_split_date)]['Close']"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5BO5R1yY6moY",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 283
        },
        "outputId": "b5f0d87d-8697-4574-d53a-1243a2bfa158"
      },
      "source": [
        "# Plot all lines on one chart to see where one segment starts and another ends\n",
        "plt.plot(x_train, label = 'Train')\n",
        "plt.plot(x_valid, label = 'Validate')\n",
        "plt.plot(x_test, label = 'Test')\n",
        "plt.legend()\n",
        "print(x_train.index.max(),x_valid.index.min(),x_valid.index.max(),x_test.index.min(),x_test.index.max())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "5521 5522 6527 6528 6949\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1GiizAJ_oQjk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Reshape values\n",
        "x_train_values = x_train.values.reshape(-1, 1)\n",
        "x_valid_values = x_valid.values.reshape(-1, 1)\n",
        "x_test_values = x_test.values.reshape(-1, 1)\n",
        "\n",
        "#  Create Scaler Object\n",
        "x_train_scaler = MinMaxScaler(feature_range=(0, 1))\n",
        "\n",
        "# Fit x_train values\n",
        "normalized_x_train = x_train_scaler.fit_transform(x_train_values)\n",
        "\n",
        "# Fit x_valid values\n",
        "normalized_x_valid = x_train_scaler.transform(x_valid_values)\n",
        "\n",
        "# Fit x_test values\n",
        "normalized_x_test = x_train_scaler.transform(x_test_values)\n",
        "\n",
        "# All values normalized to training data\n",
        "spy_normalized_to_traindata = x_train_scaler.transform(series.values.reshape(-1, 1))\n",
        "\n",
        "# Example of how to iverse\n",
        "# inversed = scaler.inverse_transform(normalized_x_train).flatten()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "9fPenJpTtuDE",
        "colab": {}
      },
      "source": [
        "class ResetStatesCallback(keras.callbacks.Callback):\n",
        "    def on_epoch_begin(self, epoch, logs):\n",
        "        self.model.reset_states()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "EPjK0l9P8OJM"
      },
      "source": [
        "## Find the learning Rate"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "cSoUmW-x8OJN",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "49f3a9fa-1e42-4be4-ec13-77971aba21d6"
      },
      "source": [
        "# reset any stored data\n",
        "keras.backend.clear_session()\n",
        "tf.random.set_seed(42)\n",
        "np.random.seed(42)\n",
        "\n",
        "# Set window size and create input batch sequences\n",
        "window_size = 20\n",
        "train_set = sequential_window_dataset(normalized_x_train, window_size)\n",
        "\n",
        "# create model\n",
        "model = keras.models.Sequential([\n",
        "  keras.layers.LSTM(100, return_sequences=True, stateful=True,\n",
        "                    batch_input_shape=[1, None, 1]),\n",
        "  keras.layers.LSTM(100, return_sequences=True, stateful=True),\n",
        "  keras.layers.Dense(1),\n",
        "])\n",
        "\n",
        "# create lr\n",
        "lr_schedule = keras.callbacks.LearningRateScheduler(\n",
        "    lambda epoch: 1e-5 * 10**(epoch / 20))\n",
        "reset_states = ResetStatesCallback()\n",
        "\n",
        "# choose optimizer\n",
        "optimizer = keras.optimizers.Nadam(lr=1e-5)\n",
        "\n",
        "# compile model\n",
        "model.compile(loss=keras.losses.Huber(),\n",
        "              optimizer=optimizer,\n",
        "              metrics=[\"mae\"])\n",
        "\n",
        "# set history\n",
        "history = model.fit(train_set, epochs=100,\n",
        "                    callbacks=[lr_schedule, reset_states])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/100\n",
            "276/276 [==============================] - 3s 9ms/step - loss: 0.0109 - mae: 0.1198\n",
            "Epoch 2/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 7.1192e-04 - mae: 0.0270\n",
            "Epoch 3/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 4.6617e-04 - mae: 0.0222\n",
            "Epoch 4/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 3.1456e-04 - mae: 0.0189\n",
            "Epoch 5/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 2.2681e-04 - mae: 0.0165\n",
            "Epoch 6/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.8071e-04 - mae: 0.0150\n",
            "Epoch 7/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.5779e-04 - mae: 0.0139\n",
            "Epoch 8/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.4623e-04 - mae: 0.0133\n",
            "Epoch 9/100\n",
            "276/276 [==============================] - 2s 9ms/step - loss: 1.3998e-04 - mae: 0.0128\n",
            "Epoch 10/100\n",
            "276/276 [==============================] - 2s 9ms/step - loss: 1.3628e-04 - mae: 0.0125\n",
            "Epoch 11/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.3389e-04 - mae: 0.0123\n",
            "Epoch 12/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.3216e-04 - mae: 0.0122\n",
            "Epoch 13/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.3077e-04 - mae: 0.0121\n",
            "Epoch 14/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2950e-04 - mae: 0.0120\n",
            "Epoch 15/100\n",
            "276/276 [==============================] - 2s 9ms/step - loss: 1.2823e-04 - mae: 0.0119\n",
            "Epoch 16/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2691e-04 - mae: 0.0118\n",
            "Epoch 17/100\n",
            "276/276 [==============================] - 2s 9ms/step - loss: 1.2551e-04 - mae: 0.0117\n",
            "Epoch 18/100\n",
            "276/276 [==============================] - 2s 9ms/step - loss: 1.2404e-04 - mae: 0.0116\n",
            "Epoch 19/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2252e-04 - mae: 0.0115\n",
            "Epoch 20/100\n",
            "276/276 [==============================] - 2s 9ms/step - loss: 1.2101e-04 - mae: 0.0114\n",
            "Epoch 21/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1955e-04 - mae: 0.0113\n",
            "Epoch 22/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1821e-04 - mae: 0.0113\n",
            "Epoch 23/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1706e-04 - mae: 0.0112\n",
            "Epoch 24/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1618e-04 - mae: 0.0111\n",
            "Epoch 25/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1558e-04 - mae: 0.0111\n",
            "Epoch 26/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1525e-04 - mae: 0.0110\n",
            "Epoch 27/100\n",
            "276/276 [==============================] - 2s 9ms/step - loss: 1.1511e-04 - mae: 0.0110\n",
            "Epoch 28/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1508e-04 - mae: 0.0109\n",
            "Epoch 29/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1515e-04 - mae: 0.0109\n",
            "Epoch 30/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1541e-04 - mae: 0.0108\n",
            "Epoch 31/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1595e-04 - mae: 0.0107\n",
            "Epoch 32/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1677e-04 - mae: 0.0107\n",
            "Epoch 33/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1784e-04 - mae: 0.0106\n",
            "Epoch 34/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1909e-04 - mae: 0.0106\n",
            "Epoch 35/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1988e-04 - mae: 0.0105\n",
            "Epoch 36/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2090e-04 - mae: 0.0104\n",
            "Epoch 37/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2187e-04 - mae: 0.0104\n",
            "Epoch 38/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2421e-04 - mae: 0.0104\n",
            "Epoch 39/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2557e-04 - mae: 0.0103\n",
            "Epoch 40/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2726e-04 - mae: 0.0102\n",
            "Epoch 41/100\n",
            "276/276 [==============================] - 2s 9ms/step - loss: 1.2609e-04 - mae: 0.0101\n",
            "Epoch 42/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2533e-04 - mae: 0.0100\n",
            "Epoch 43/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.2102e-04 - mae: 0.0097\n",
            "Epoch 44/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1636e-04 - mae: 0.0095\n",
            "Epoch 45/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.1037e-04 - mae: 0.0092\n",
            "Epoch 46/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 9.7899e-05 - mae: 0.0087\n",
            "Epoch 47/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 9.3871e-05 - mae: 0.0086\n",
            "Epoch 48/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 8.9027e-05 - mae: 0.0085\n",
            "Epoch 49/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 9.2427e-05 - mae: 0.0086\n",
            "Epoch 50/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.3946e-04 - mae: 0.0103\n",
            "Epoch 51/100\n",
            "276/276 [==============================] - 3s 9ms/step - loss: 1.8082e-04 - mae: 0.0108\n",
            "Epoch 52/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 2.1126e-04 - mae: 0.0115\n",
            "Epoch 53/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 5.2716e-04 - mae: 0.0188\n",
            "Epoch 54/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0014 - mae: 0.0182\n",
            "Epoch 55/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 2.4640e-04 - mae: 0.0106\n",
            "Epoch 56/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 9.6223e-05 - mae: 0.0074\n",
            "Epoch 57/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 5.1194e-05 - mae: 0.0068\n",
            "Epoch 58/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 7.4970e-05 - mae: 0.0071\n",
            "Epoch 59/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 6.6583e-05 - mae: 0.0074\n",
            "Epoch 60/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 8.7840e-05 - mae: 0.0076\n",
            "Epoch 61/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 6.2480e-05 - mae: 0.0074\n",
            "Epoch 62/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 7.7943e-05 - mae: 0.0074\n",
            "Epoch 63/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 1.7969e-04 - mae: 0.0108\n",
            "Epoch 64/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 4.2404e-04 - mae: 0.0170\n",
            "Epoch 65/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0142 - mae: 0.0549\n",
            "Epoch 66/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0025 - mae: 0.0239\n",
            "Epoch 67/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0023 - mae: 0.0252\n",
            "Epoch 68/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0016 - mae: 0.0223\n",
            "Epoch 69/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 8.4764e-04 - mae: 0.0224\n",
            "Epoch 70/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0019 - mae: 0.0257\n",
            "Epoch 71/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0014 - mae: 0.0234\n",
            "Epoch 72/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0010 - mae: 0.0219\n",
            "Epoch 73/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0012 - mae: 0.0218\n",
            "Epoch 74/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0027 - mae: 0.0275\n",
            "Epoch 75/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0019 - mae: 0.0261\n",
            "Epoch 76/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0033 - mae: 0.0284\n",
            "Epoch 77/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0084 - mae: 0.0330\n",
            "Epoch 78/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0037 - mae: 0.0268\n",
            "Epoch 79/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0206 - mae: 0.0745\n",
            "Epoch 80/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0055 - mae: 0.0393\n",
            "Epoch 81/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0207 - mae: 0.0777\n",
            "Epoch 82/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0175 - mae: 0.0959\n",
            "Epoch 83/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0063 - mae: 0.0565\n",
            "Epoch 84/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0163 - mae: 0.1093\n",
            "Epoch 85/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0189 - mae: 0.1207\n",
            "Epoch 86/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0224 - mae: 0.1100\n",
            "Epoch 87/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0589 - mae: 0.1791\n",
            "Epoch 88/100\n",
            "276/276 [==============================] - 2s 9ms/step - loss: 0.0356 - mae: 0.1607\n",
            "Epoch 89/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0507 - mae: 0.1814\n",
            "Epoch 90/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0649 - mae: 0.2157\n",
            "Epoch 91/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.1557 - mae: 0.3670\n",
            "Epoch 92/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.0385 - mae: 0.1544\n",
            "Epoch 93/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.1620 - mae: 0.3821\n",
            "Epoch 94/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.0608 - mae: 0.2522\n",
            "Epoch 95/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.3353 - mae: 0.7034\n",
            "Epoch 96/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.2039 - mae: 0.5022\n",
            "Epoch 97/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.2743 - mae: 0.6712\n",
            "Epoch 98/100\n",
            "276/276 [==============================] - 2s 8ms/step - loss: 0.3448 - mae: 0.7910\n",
            "Epoch 99/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.4008 - mae: 0.8490\n",
            "Epoch 100/100\n",
            "276/276 [==============================] - 2s 7ms/step - loss: 0.5534 - mae: 1.0528\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "KA0GM9sQ8OJO",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 293
        },
        "outputId": "f8484d5a-5417-45cc-e0b9-0273d6c44c4e"
      },
      "source": [
        "# Plot the learning rate chart\n",
        "plt.semilogx(history.history[\"lr\"], history.history[\"loss\"])\n",
        "plt.axis([1e-8, 1, 0, 0.01])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1e-08, 1.0, 0.0, 0.01)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7EyiYesc-EPU",
        "colab_type": "text"
      },
      "source": [
        "## Build and Train LSTM Model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "hiHR5pPL8OJP",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "f45ec9c3-b1c0-4e87-b3ab-32c137222988"
      },
      "source": [
        "# reset any stored data\n",
        "keras.backend.clear_session()\n",
        "tf.random.set_seed(42)\n",
        "np.random.seed(42)\n",
        "\n",
        "# set window size and create input batch sequence\n",
        "window_size = 30\n",
        "train_set = sequential_window_dataset(normalized_x_train, window_size)\n",
        "valid_set = sequential_window_dataset(normalized_x_valid, window_size)\n",
        "\n",
        "# create model\n",
        "model = keras.models.Sequential([\n",
        "  keras.layers.LSTM(100, return_sequences=True, stateful=True,\n",
        "                         batch_input_shape=[1, None, 1]),\n",
        "  keras.layers.LSTM(100, return_sequences=True, stateful=True),\n",
        "  keras.layers.Dense(1),\n",
        "])\n",
        "\n",
        "# set optimizer\n",
        "optimizer = keras.optimizers.Nadam(lr=1e-4)\n",
        "\n",
        "# compile model\n",
        "model.compile(loss=keras.losses.Huber(),\n",
        "              optimizer=optimizer,\n",
        "              metrics=[\"mae\"])\n",
        "\n",
        "# reset states\n",
        "reset_states = ResetStatesCallback()\n",
        "\n",
        "#set up save best only checkpoint\n",
        "model_checkpoint = keras.callbacks.ModelCheckpoint(\n",
        "    \"my_checkpoint\", save_best_only=True)\n",
        "\n",
        "early_stopping = keras.callbacks.EarlyStopping(patience=50)\n",
        "\n",
        "# fit model\n",
        "model.fit(train_set, epochs=500,\n",
        "          validation_data=valid_set,\n",
        "          callbacks=[early_stopping, model_checkpoint, reset_states])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/500\n",
            "    180/Unknown - 1s 8ms/step - loss: 1.7256e-04 - mae: 0.0138INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 50ms/step - loss: 1.9735e-04 - mae: 0.0144 - val_loss: 0.0045 - val_mae: 0.0720\n",
            "Epoch 2/500\n",
            "178/184 [============================>.] - ETA: 0s - loss: 3.0633e-04 - mae: 0.0162INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 49ms/step - loss: 3.0381e-04 - mae: 0.0162 - val_loss: 0.0020 - val_mae: 0.0469\n",
            "Epoch 3/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 1.2606e-04 - mae: 0.0117INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 50ms/step - loss: 1.3149e-04 - mae: 0.0120 - val_loss: 0.0011 - val_mae: 0.0361\n",
            "Epoch 4/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 1.0997e-04 - mae: 0.0108INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 49ms/step - loss: 1.1313e-04 - mae: 0.0109 - val_loss: 9.3129e-04 - val_mae: 0.0336\n",
            "Epoch 5/500\n",
            "178/184 [============================>.] - ETA: 0s - loss: 1.0368e-04 - mae: 0.0104INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 1.0802e-04 - mae: 0.0106 - val_loss: 8.6388e-04 - val_mae: 0.0324\n",
            "Epoch 6/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 9.9785e-05 - mae: 0.0101INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 49ms/step - loss: 1.0382e-04 - mae: 0.0103 - val_loss: 8.3215e-04 - val_mae: 0.0315\n",
            "Epoch 7/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 9.6271e-05 - mae: 0.0099INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 49ms/step - loss: 9.9913e-05 - mae: 0.0101 - val_loss: 8.1579e-04 - val_mae: 0.0310\n",
            "Epoch 8/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 9.3842e-05 - mae: 0.0098INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 49ms/step - loss: 9.6270e-05 - mae: 0.0099 - val_loss: 8.0799e-04 - val_mae: 0.0306\n",
            "Epoch 9/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 8.9459e-05 - mae: 0.0096INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 47ms/step - loss: 9.2871e-05 - mae: 0.0098 - val_loss: 8.0563e-04 - val_mae: 0.0304\n",
            "Epoch 10/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 8.9697e-05 - mae: 0.0096 - val_loss: 8.0667e-04 - val_mae: 0.0302\n",
            "Epoch 11/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 8.6734e-05 - mae: 0.0094 - val_loss: 8.0824e-04 - val_mae: 0.0301\n",
            "Epoch 12/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 8.1572e-05 - mae: 0.0091INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 8.3988e-05 - mae: 0.0093 - val_loss: 8.0512e-04 - val_mae: 0.0300\n",
            "Epoch 13/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 7.9157e-05 - mae: 0.0090INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 49ms/step - loss: 8.1474e-05 - mae: 0.0091 - val_loss: 7.9039e-04 - val_mae: 0.0297\n",
            "Epoch 14/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 7.6667e-05 - mae: 0.0089INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 7.9210e-05 - mae: 0.0090 - val_loss: 7.5965e-04 - val_mae: 0.0292\n",
            "Epoch 15/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 7.4955e-05 - mae: 0.0088INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 50ms/step - loss: 7.7191e-05 - mae: 0.0089 - val_loss: 7.1782e-04 - val_mae: 0.0285\n",
            "Epoch 16/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 7.3416e-05 - mae: 0.0087INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 49ms/step - loss: 7.5360e-05 - mae: 0.0088 - val_loss: 6.7717e-04 - val_mae: 0.0279\n",
            "Epoch 17/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 7.1132e-05 - mae: 0.0085INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 46ms/step - loss: 7.3629e-05 - mae: 0.0086 - val_loss: 6.4497e-04 - val_mae: 0.0273\n",
            "Epoch 18/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 6.9809e-05 - mae: 0.0084INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 7.1963e-05 - mae: 0.0085 - val_loss: 6.1962e-04 - val_mae: 0.0269\n",
            "Epoch 19/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 6.7965e-05 - mae: 0.0083INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 7.0385e-05 - mae: 0.0084 - val_loss: 5.9805e-04 - val_mae: 0.0264\n",
            "Epoch 20/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 6.6526e-05 - mae: 0.0082INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 6.8904e-05 - mae: 0.0083 - val_loss: 5.7880e-04 - val_mae: 0.0261\n",
            "Epoch 21/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 6.5182e-05 - mae: 0.0081INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 6.7517e-05 - mae: 0.0082 - val_loss: 5.6098e-04 - val_mae: 0.0257\n",
            "Epoch 22/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 6.4184e-05 - mae: 0.0080INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 6.6223e-05 - mae: 0.0082 - val_loss: 5.4456e-04 - val_mae: 0.0254\n",
            "Epoch 23/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 6.3075e-05 - mae: 0.0080INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 46ms/step - loss: 6.5013e-05 - mae: 0.0081 - val_loss: 5.2906e-04 - val_mae: 0.0250\n",
            "Epoch 24/500\n",
            "184/184 [==============================] - ETA: 0s - loss: 6.3883e-05 - mae: 0.0080INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 6.3883e-05 - mae: 0.0080 - val_loss: 5.1510e-04 - val_mae: 0.0247\n",
            "Epoch 25/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 6.0651e-05 - mae: 0.0078INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 6.2819e-05 - mae: 0.0079 - val_loss: 5.0175e-04 - val_mae: 0.0245\n",
            "Epoch 26/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 5.9703e-05 - mae: 0.0077INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 50ms/step - loss: 6.1826e-05 - mae: 0.0079 - val_loss: 4.9019e-04 - val_mae: 0.0242\n",
            "Epoch 27/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 5.9095e-05 - mae: 0.0077INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 6.0884e-05 - mae: 0.0078 - val_loss: 4.7851e-04 - val_mae: 0.0240\n",
            "Epoch 28/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 5.8368e-05 - mae: 0.0077INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 10s 52ms/step - loss: 6.0011e-05 - mae: 0.0078 - val_loss: 4.6922e-04 - val_mae: 0.0237\n",
            "Epoch 29/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 5.7370e-05 - mae: 0.0076INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 5.9171e-05 - mae: 0.0077 - val_loss: 4.5805e-04 - val_mae: 0.0235\n",
            "Epoch 30/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 5.6681e-05 - mae: 0.0075INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 5.8413e-05 - mae: 0.0077 - val_loss: 4.5179e-04 - val_mae: 0.0233\n",
            "Epoch 31/500\n",
            "178/184 [============================>.] - ETA: 0s - loss: 5.6037e-05 - mae: 0.0075INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 46ms/step - loss: 5.7640e-05 - mae: 0.0076 - val_loss: 4.3911e-04 - val_mae: 0.0232\n",
            "Epoch 32/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 5.7020e-05 - mae: 0.0076 - val_loss: 4.3969e-04 - val_mae: 0.0229\n",
            "Epoch 33/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 5.4413e-05 - mae: 0.0074INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 5.6231e-05 - mae: 0.0075 - val_loss: 4.2241e-04 - val_mae: 0.0231\n",
            "Epoch 34/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 5.5913e-05 - mae: 0.0075 - val_loss: 4.5365e-04 - val_mae: 0.0227\n",
            "Epoch 35/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 5.4879e-05 - mae: 0.0074 - val_loss: 4.4921e-04 - val_mae: 0.0248\n",
            "Epoch 36/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 5.6064e-05 - mae: 0.0075 - val_loss: 7.4789e-04 - val_mae: 0.0295\n",
            "Epoch 37/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 5.2889e-05 - mae: 0.0073 - val_loss: 5.5608e-04 - val_mae: 0.0283\n",
            "Epoch 38/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 5.8446e-05 - mae: 0.0078 - val_loss: 0.0011 - val_mae: 0.0383\n",
            "Epoch 39/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 5.0359e-05 - mae: 0.0072 - val_loss: 5.4204e-04 - val_mae: 0.0257\n",
            "Epoch 40/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 5.4952e-05 - mae: 0.0076 - val_loss: 5.2967e-04 - val_mae: 0.0242\n",
            "Epoch 41/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 5.1010e-05 - mae: 0.0071 - val_loss: 4.6375e-04 - val_mae: 0.0254\n",
            "Epoch 42/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 5.3681e-05 - mae: 0.0074 - val_loss: 7.2583e-04 - val_mae: 0.0289\n",
            "Epoch 43/500\n",
            "184/184 [==============================] - 2s 10ms/step - loss: 4.9810e-05 - mae: 0.0072 - val_loss: 4.8792e-04 - val_mae: 0.0261\n",
            "Epoch 44/500\n",
            "184/184 [==============================] - 2s 10ms/step - loss: 5.4073e-05 - mae: 0.0075 - val_loss: 7.4445e-04 - val_mae: 0.0294\n",
            "Epoch 45/500\n",
            "184/184 [==============================] - 2s 10ms/step - loss: 4.8501e-05 - mae: 0.0071 - val_loss: 4.6631e-04 - val_mae: 0.0250\n",
            "Epoch 46/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 5.2256e-05 - mae: 0.0074 - val_loss: 5.9232e-04 - val_mae: 0.0255\n",
            "Epoch 47/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 4.8093e-05 - mae: 0.0070 - val_loss: 4.5548e-04 - val_mae: 0.0251\n",
            "Epoch 48/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 5.1635e-05 - mae: 0.0073 - val_loss: 6.0837e-04 - val_mae: 0.0261\n",
            "Epoch 49/500\n",
            "184/184 [==============================] - 2s 10ms/step - loss: 4.7344e-05 - mae: 0.0070 - val_loss: 4.4147e-04 - val_mae: 0.0244\n",
            "Epoch 50/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 5.0546e-05 - mae: 0.0072 - val_loss: 5.3542e-04 - val_mae: 0.0242\n",
            "Epoch 51/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 4.6612e-05 - mae: 0.0069 - val_loss: 4.2456e-04 - val_mae: 0.0240\n",
            "Epoch 52/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 4.9641e-05 - mae: 0.0071 - val_loss: 4.9913e-04 - val_mae: 0.0234\n",
            "Epoch 53/500\n",
            "183/184 [============================>.] - ETA: 0s - loss: 4.5026e-05 - mae: 0.0068INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 46ms/step - loss: 4.5965e-05 - mae: 0.0069 - val_loss: 4.0971e-04 - val_mae: 0.0235\n",
            "Epoch 54/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 4.8793e-05 - mae: 0.0071 - val_loss: 4.6118e-04 - val_mae: 0.0224\n",
            "Epoch 55/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 4.3730e-05 - mae: 0.0068INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 4.5238e-05 - mae: 0.0068 - val_loss: 3.9326e-04 - val_mae: 0.0230\n",
            "Epoch 56/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 4.7922e-05 - mae: 0.0070 - val_loss: 4.2562e-04 - val_mae: 0.0216\n",
            "Epoch 57/500\n",
            "184/184 [==============================] - ETA: 0s - loss: 4.4530e-05 - mae: 0.0068INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 4.4530e-05 - mae: 0.0068 - val_loss: 3.7701e-04 - val_mae: 0.0224\n",
            "Epoch 58/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 4.7113e-05 - mae: 0.0069 - val_loss: 3.9500e-04 - val_mae: 0.0208\n",
            "Epoch 59/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 4.2310e-05 - mae: 0.0066INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 52ms/step - loss: 4.3809e-05 - mae: 0.0067 - val_loss: 3.6147e-04 - val_mae: 0.0220\n",
            "Epoch 60/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 4.6352e-05 - mae: 0.0069 - val_loss: 3.6787e-04 - val_mae: 0.0201\n",
            "Epoch 61/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 4.1627e-05 - mae: 0.0066INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 44ms/step - loss: 4.3073e-05 - mae: 0.0067 - val_loss: 3.4678e-04 - val_mae: 0.0215\n",
            "Epoch 62/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 4.4920e-05 - mae: 0.0068INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 4.5637e-05 - mae: 0.0068 - val_loss: 3.4350e-04 - val_mae: 0.0195\n",
            "Epoch 63/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 4.1069e-05 - mae: 0.0065INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 48ms/step - loss: 4.2328e-05 - mae: 0.0066 - val_loss: 3.3327e-04 - val_mae: 0.0211\n",
            "Epoch 64/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 4.4441e-05 - mae: 0.0067INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 46ms/step - loss: 4.4975e-05 - mae: 0.0068 - val_loss: 3.2141e-04 - val_mae: 0.0189\n",
            "Epoch 65/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 4.0274e-05 - mae: 0.0064INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 50ms/step - loss: 4.1576e-05 - mae: 0.0065 - val_loss: 3.2125e-04 - val_mae: 0.0208\n",
            "Epoch 66/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 4.3601e-05 - mae: 0.0067INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 46ms/step - loss: 4.4372e-05 - mae: 0.0067 - val_loss: 3.0122e-04 - val_mae: 0.0183\n",
            "Epoch 67/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 4.0817e-05 - mae: 0.0065 - val_loss: 3.1093e-04 - val_mae: 0.0205\n",
            "Epoch 68/500\n",
            "177/184 [===========================>..] - ETA: 0s - loss: 4.3351e-05 - mae: 0.0066INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 50ms/step - loss: 4.3830e-05 - mae: 0.0067 - val_loss: 2.8251e-04 - val_mae: 0.0177\n",
            "Epoch 69/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 4.0054e-05 - mae: 0.0064 - val_loss: 3.0230e-04 - val_mae: 0.0202\n",
            "Epoch 70/500\n",
            "183/184 [============================>.] - ETA: 0s - loss: 4.2977e-05 - mae: 0.0066INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 44ms/step - loss: 4.3346e-05 - mae: 0.0067 - val_loss: 2.6492e-04 - val_mae: 0.0172\n",
            "Epoch 71/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.9292e-05 - mae: 0.0063 - val_loss: 2.9493e-04 - val_mae: 0.0200\n",
            "Epoch 72/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 4.2309e-05 - mae: 0.0066INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 4.2902e-05 - mae: 0.0067 - val_loss: 2.4782e-04 - val_mae: 0.0166\n",
            "Epoch 73/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.8542e-05 - mae: 0.0063 - val_loss: 2.8845e-04 - val_mae: 0.0198\n",
            "Epoch 74/500\n",
            "184/184 [==============================] - ETA: 0s - loss: 4.2480e-05 - mae: 0.0066INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 46ms/step - loss: 4.2480e-05 - mae: 0.0066 - val_loss: 2.3046e-04 - val_mae: 0.0160\n",
            "Epoch 75/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.7819e-05 - mae: 0.0062 - val_loss: 2.8274e-04 - val_mae: 0.0196\n",
            "Epoch 76/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 4.1426e-05 - mae: 0.0066INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 4.2064e-05 - mae: 0.0066 - val_loss: 2.1230e-04 - val_mae: 0.0154\n",
            "Epoch 77/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.7131e-05 - mae: 0.0061 - val_loss: 2.7812e-04 - val_mae: 0.0195\n",
            "Epoch 78/500\n",
            "181/184 [============================>.] - ETA: 0s - loss: 4.1133e-05 - mae: 0.0065INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 4.1650e-05 - mae: 0.0066 - val_loss: 1.9326e-04 - val_mae: 0.0147\n",
            "Epoch 79/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.6477e-05 - mae: 0.0061 - val_loss: 2.7515e-04 - val_mae: 0.0194\n",
            "Epoch 80/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 4.0795e-05 - mae: 0.0065INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 50ms/step - loss: 4.1239e-05 - mae: 0.0066 - val_loss: 1.7372e-04 - val_mae: 0.0139\n",
            "Epoch 81/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.5853e-05 - mae: 0.0060 - val_loss: 2.7449e-04 - val_mae: 0.0194\n",
            "Epoch 82/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 4.0546e-05 - mae: 0.0065INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 4.0827e-05 - mae: 0.0065 - val_loss: 1.5442e-04 - val_mae: 0.0131\n",
            "Epoch 83/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.5251e-05 - mae: 0.0060 - val_loss: 2.7726e-04 - val_mae: 0.0196\n",
            "Epoch 84/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 3.9863e-05 - mae: 0.0065INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 50ms/step - loss: 4.0410e-05 - mae: 0.0065 - val_loss: 1.3649e-04 - val_mae: 0.0122\n",
            "Epoch 85/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.4672e-05 - mae: 0.0059 - val_loss: 2.8538e-04 - val_mae: 0.0200\n",
            "Epoch 86/500\n",
            "179/184 [============================>.] - ETA: 0s - loss: 3.9752e-05 - mae: 0.0065INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 46ms/step - loss: 3.9994e-05 - mae: 0.0065 - val_loss: 1.2155e-04 - val_mae: 0.0114\n",
            "Epoch 87/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.4111e-05 - mae: 0.0059 - val_loss: 3.0085e-04 - val_mae: 0.0206\n",
            "Epoch 88/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 3.9070e-05 - mae: 0.0064INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 3.9578e-05 - mae: 0.0064 - val_loss: 1.1186e-04 - val_mae: 0.0108\n",
            "Epoch 89/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.3570e-05 - mae: 0.0058 - val_loss: 3.2664e-04 - val_mae: 0.0217\n",
            "Epoch 90/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 3.8674e-05 - mae: 0.0064INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 8s 45ms/step - loss: 3.9165e-05 - mae: 0.0064 - val_loss: 1.1014e-04 - val_mae: 0.0106\n",
            "Epoch 91/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.3047e-05 - mae: 0.0058 - val_loss: 3.6512e-04 - val_mae: 0.0234\n",
            "Epoch 92/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.8753e-05 - mae: 0.0064 - val_loss: 1.1945e-04 - val_mae: 0.0111\n",
            "Epoch 93/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.2542e-05 - mae: 0.0057 - val_loss: 4.1914e-04 - val_mae: 0.0257\n",
            "Epoch 94/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.8344e-05 - mae: 0.0063 - val_loss: 1.4271e-04 - val_mae: 0.0124\n",
            "Epoch 95/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.2054e-05 - mae: 0.0056 - val_loss: 4.8965e-04 - val_mae: 0.0284\n",
            "Epoch 96/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.7930e-05 - mae: 0.0063 - val_loss: 1.8263e-04 - val_mae: 0.0146\n",
            "Epoch 97/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.1584e-05 - mae: 0.0056 - val_loss: 5.7988e-04 - val_mae: 0.0314\n",
            "Epoch 98/500\n",
            "184/184 [==============================] - 1s 8ms/step - loss: 3.7522e-05 - mae: 0.0063 - val_loss: 2.4050e-04 - val_mae: 0.0174\n",
            "Epoch 99/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.1133e-05 - mae: 0.0055 - val_loss: 6.8964e-04 - val_mae: 0.0345\n",
            "Epoch 100/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.7144e-05 - mae: 0.0062 - val_loss: 3.1538e-04 - val_mae: 0.0202\n",
            "Epoch 101/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.0696e-05 - mae: 0.0055 - val_loss: 8.1580e-04 - val_mae: 0.0376\n",
            "Epoch 102/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.6810e-05 - mae: 0.0062 - val_loss: 4.0357e-04 - val_mae: 0.0230\n",
            "Epoch 103/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.0270e-05 - mae: 0.0054 - val_loss: 9.5227e-04 - val_mae: 0.0405\n",
            "Epoch 104/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.6535e-05 - mae: 0.0062 - val_loss: 4.9825e-04 - val_mae: 0.0255\n",
            "Epoch 105/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.9860e-05 - mae: 0.0054 - val_loss: 0.0011 - val_mae: 0.0433\n",
            "Epoch 106/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.6360e-05 - mae: 0.0062 - val_loss: 5.8356e-04 - val_mae: 0.0273\n",
            "Epoch 107/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.9462e-05 - mae: 0.0054 - val_loss: 0.0012 - val_mae: 0.0455\n",
            "Epoch 108/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.6344e-05 - mae: 0.0062 - val_loss: 6.3270e-04 - val_mae: 0.0279\n",
            "Epoch 109/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.9081e-05 - mae: 0.0053 - val_loss: 0.0013 - val_mae: 0.0467\n",
            "Epoch 110/500\n",
            "184/184 [==============================] - 2s 10ms/step - loss: 3.6576e-05 - mae: 0.0062 - val_loss: 6.0335e-04 - val_mae: 0.0266\n",
            "Epoch 111/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8718e-05 - mae: 0.0053 - val_loss: 0.0012 - val_mae: 0.0455\n",
            "Epoch 112/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.7138e-05 - mae: 0.0062 - val_loss: 4.6580e-04 - val_mae: 0.0226\n",
            "Epoch 113/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8434e-05 - mae: 0.0053 - val_loss: 9.8567e-04 - val_mae: 0.0411\n",
            "Epoch 114/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.8149e-05 - mae: 0.0063 - val_loss: 2.6739e-04 - val_mae: 0.0165\n",
            "Epoch 115/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8229e-05 - mae: 0.0053 - val_loss: 5.9813e-04 - val_mae: 0.0319\n",
            "Epoch 116/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.9109e-05 - mae: 0.0063 - val_loss: 1.4597e-04 - val_mae: 0.0119\n",
            "Epoch 117/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8043e-05 - mae: 0.0052 - val_loss: 5.2040e-04 - val_mae: 0.0298\n",
            "Epoch 118/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.9693e-05 - mae: 0.0064 - val_loss: 1.2966e-04 - val_mae: 0.0126\n",
            "Epoch 119/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8008e-05 - mae: 0.0053 - val_loss: 1.6490e-04 - val_mae: 0.0145\n",
            "Epoch 120/500\n",
            "178/184 [============================>.] - ETA: 0s - loss: 3.9209e-05 - mae: 0.0062INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 3.8981e-05 - mae: 0.0062 - val_loss: 8.7059e-05 - val_mae: 0.0089\n",
            "Epoch 121/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.9347e-05 - mae: 0.0052 - val_loss: 0.0010 - val_mae: 0.0429\n",
            "Epoch 122/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 4.0796e-05 - mae: 0.0066 - val_loss: 2.0102e-04 - val_mae: 0.0143\n",
            "Epoch 123/500\n",
            "183/184 [============================>.] - ETA: 0s - loss: 3.0821e-05 - mae: 0.0058INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 46ms/step - loss: 3.0981e-05 - mae: 0.0058 - val_loss: 8.1521e-05 - val_mae: 0.0089\n",
            "Epoch 124/500\n",
            "182/184 [============================>.] - ETA: 0s - loss: 3.2110e-05 - mae: 0.0057INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 51ms/step - loss: 3.2382e-05 - mae: 0.0057 - val_loss: 7.7045e-05 - val_mae: 0.0086\n",
            "Epoch 125/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8298e-05 - mae: 0.0054 - val_loss: 8.1593e-05 - val_mae: 0.0090\n",
            "Epoch 126/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8765e-05 - mae: 0.0054 - val_loss: 4.5796e-04 - val_mae: 0.0277\n",
            "Epoch 127/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.1548e-05 - mae: 0.0057 - val_loss: 2.2297e-04 - val_mae: 0.0191\n",
            "Epoch 128/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8263e-05 - mae: 0.0054 - val_loss: 1.5941e-04 - val_mae: 0.0143\n",
            "Epoch 129/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 4.0265e-05 - mae: 0.0062 - val_loss: 1.5114e-04 - val_mae: 0.0142\n",
            "Epoch 130/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.6027e-05 - mae: 0.0050 - val_loss: 3.2915e-04 - val_mae: 0.0225\n",
            "Epoch 131/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.5690e-05 - mae: 0.0062 - val_loss: 2.9269e-04 - val_mae: 0.0219\n",
            "Epoch 132/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8605e-05 - mae: 0.0055 - val_loss: 2.0469e-04 - val_mae: 0.0155\n",
            "Epoch 133/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.7053e-05 - mae: 0.0060 - val_loss: 1.7733e-04 - val_mae: 0.0145\n",
            "Epoch 134/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.5811e-05 - mae: 0.0050 - val_loss: 2.0614e-04 - val_mae: 0.0170\n",
            "Epoch 135/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.0614e-05 - mae: 0.0056 - val_loss: 3.8316e-04 - val_mae: 0.0247\n",
            "Epoch 136/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.7274e-05 - mae: 0.0053 - val_loss: 1.9005e-04 - val_mae: 0.0155\n",
            "Epoch 137/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.6182e-05 - mae: 0.0060 - val_loss: 2.2242e-04 - val_mae: 0.0168\n",
            "Epoch 138/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.5473e-05 - mae: 0.0049 - val_loss: 3.4264e-04 - val_mae: 0.0229\n",
            "Epoch 139/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.5744e-05 - mae: 0.0062 - val_loss: 1.5961e-04 - val_mae: 0.0151\n",
            "Epoch 140/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.8491e-05 - mae: 0.0055 - val_loss: 1.8356e-04 - val_mae: 0.0145\n",
            "Epoch 141/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.0798e-05 - mae: 0.0056 - val_loss: 2.0131e-04 - val_mae: 0.0154\n",
            "Epoch 142/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.5246e-05 - mae: 0.0050 - val_loss: 1.3638e-04 - val_mae: 0.0132\n",
            "Epoch 143/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.7634e-05 - mae: 0.0053 - val_loss: 8.9536e-05 - val_mae: 0.0100\n",
            "Epoch 144/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.8137e-05 - mae: 0.0051 - val_loss: 4.2874e-04 - val_mae: 0.0270\n",
            "Epoch 145/500\n",
            "180/184 [============================>.] - ETA: 0s - loss: 3.3950e-05 - mae: 0.0059INFO:tensorflow:Assets written to: my_checkpoint/assets\n",
            "184/184 [==============================] - 9s 48ms/step - loss: 3.4106e-05 - mae: 0.0059 - val_loss: 7.4275e-05 - val_mae: 0.0083\n",
            "Epoch 146/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8369e-05 - mae: 0.0055 - val_loss: 9.6272e-05 - val_mae: 0.0103\n",
            "Epoch 147/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.7855e-05 - mae: 0.0053 - val_loss: 1.1140e-04 - val_mae: 0.0113\n",
            "Epoch 148/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.6126e-05 - mae: 0.0051 - val_loss: 1.5093e-04 - val_mae: 0.0134\n",
            "Epoch 149/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.6071e-05 - mae: 0.0051 - val_loss: 1.4494e-04 - val_mae: 0.0129\n",
            "Epoch 150/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.6890e-05 - mae: 0.0052 - val_loss: 2.4838e-04 - val_mae: 0.0193\n",
            "Epoch 151/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.0265e-05 - mae: 0.0056 - val_loss: 4.4795e-04 - val_mae: 0.0275\n",
            "Epoch 152/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.8674e-05 - mae: 0.0055 - val_loss: 2.2602e-04 - val_mae: 0.0161\n",
            "Epoch 153/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.9033e-05 - mae: 0.0060 - val_loss: 1.9250e-04 - val_mae: 0.0151\n",
            "Epoch 154/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.4295e-05 - mae: 0.0049 - val_loss: 1.8922e-04 - val_mae: 0.0147\n",
            "Epoch 155/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.5094e-05 - mae: 0.0050 - val_loss: 2.2200e-04 - val_mae: 0.0179\n",
            "Epoch 156/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.9438e-05 - mae: 0.0055 - val_loss: 3.9457e-04 - val_mae: 0.0253\n",
            "Epoch 157/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.7761e-05 - mae: 0.0055 - val_loss: 2.1949e-04 - val_mae: 0.0158\n",
            "Epoch 158/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 3.3925e-05 - mae: 0.0057 - val_loss: 2.1847e-04 - val_mae: 0.0162\n",
            "Epoch 159/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.3167e-05 - mae: 0.0047 - val_loss: 1.3078e-04 - val_mae: 0.0125\n",
            "Epoch 160/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.5411e-05 - mae: 0.0050 - val_loss: 2.1563e-04 - val_mae: 0.0178\n",
            "Epoch 161/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.7480e-05 - mae: 0.0052 - val_loss: 4.6293e-04 - val_mae: 0.0277\n",
            "Epoch 162/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.8065e-05 - mae: 0.0055 - val_loss: 2.2093e-04 - val_mae: 0.0159\n",
            "Epoch 163/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.5712e-05 - mae: 0.0058 - val_loss: 2.0812e-04 - val_mae: 0.0159\n",
            "Epoch 164/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.2774e-05 - mae: 0.0047 - val_loss: 1.4666e-04 - val_mae: 0.0129\n",
            "Epoch 165/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.4753e-05 - mae: 0.0050 - val_loss: 2.6860e-04 - val_mae: 0.0203\n",
            "Epoch 166/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.9058e-05 - mae: 0.0054 - val_loss: 2.4770e-04 - val_mae: 0.0196\n",
            "Epoch 167/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.7732e-05 - mae: 0.0054 - val_loss: 1.7727e-04 - val_mae: 0.0143\n",
            "Epoch 168/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.0698e-05 - mae: 0.0055 - val_loss: 1.8806e-04 - val_mae: 0.0152\n",
            "Epoch 169/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.2617e-05 - mae: 0.0047 - val_loss: 1.2716e-04 - val_mae: 0.0121\n",
            "Epoch 170/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.4598e-05 - mae: 0.0049 - val_loss: 2.5980e-04 - val_mae: 0.0201\n",
            "Epoch 171/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.8175e-05 - mae: 0.0053 - val_loss: 2.6704e-04 - val_mae: 0.0206\n",
            "Epoch 172/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.7499e-05 - mae: 0.0054 - val_loss: 1.7189e-04 - val_mae: 0.0141\n",
            "Epoch 173/500\n",
            "184/184 [==============================] - 2s 10ms/step - loss: 3.1509e-05 - mae: 0.0055 - val_loss: 1.7279e-04 - val_mae: 0.0146\n",
            "Epoch 174/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.2385e-05 - mae: 0.0046 - val_loss: 1.4485e-04 - val_mae: 0.0129\n",
            "Epoch 175/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.3825e-05 - mae: 0.0048 - val_loss: 2.0018e-04 - val_mae: 0.0170\n",
            "Epoch 176/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.7152e-05 - mae: 0.0052 - val_loss: 3.2044e-04 - val_mae: 0.0227\n",
            "Epoch 177/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.6715e-05 - mae: 0.0053 - val_loss: 1.7910e-04 - val_mae: 0.0144\n",
            "Epoch 178/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.1299e-05 - mae: 0.0055 - val_loss: 1.7829e-04 - val_mae: 0.0148\n",
            "Epoch 179/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.1882e-05 - mae: 0.0045 - val_loss: 1.2520e-04 - val_mae: 0.0120\n",
            "Epoch 180/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.3712e-05 - mae: 0.0048 - val_loss: 2.2237e-04 - val_mae: 0.0183\n",
            "Epoch 181/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.7030e-05 - mae: 0.0051 - val_loss: 2.6881e-04 - val_mae: 0.0206\n",
            "Epoch 182/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.6721e-05 - mae: 0.0053 - val_loss: 1.6771e-04 - val_mae: 0.0139\n",
            "Epoch 183/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 3.0127e-05 - mae: 0.0054 - val_loss: 1.6916e-04 - val_mae: 0.0145\n",
            "Epoch 184/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.1674e-05 - mae: 0.0045 - val_loss: 1.2452e-04 - val_mae: 0.0119\n",
            "Epoch 185/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.3358e-05 - mae: 0.0048 - val_loss: 1.9900e-04 - val_mae: 0.0171\n",
            "Epoch 186/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.6378e-05 - mae: 0.0051 - val_loss: 2.8705e-04 - val_mae: 0.0214\n",
            "Epoch 187/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.6344e-05 - mae: 0.0053 - val_loss: 1.6809e-04 - val_mae: 0.0139\n",
            "Epoch 188/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.9999e-05 - mae: 0.0054 - val_loss: 1.6703e-04 - val_mae: 0.0144\n",
            "Epoch 189/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.1408e-05 - mae: 0.0045 - val_loss: 1.1610e-04 - val_mae: 0.0115\n",
            "Epoch 190/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.3151e-05 - mae: 0.0048 - val_loss: 1.8424e-04 - val_mae: 0.0163\n",
            "Epoch 191/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.5714e-05 - mae: 0.0050 - val_loss: 3.0156e-04 - val_mae: 0.0219\n",
            "Epoch 192/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.5988e-05 - mae: 0.0052 - val_loss: 1.6665e-04 - val_mae: 0.0139\n",
            "Epoch 193/500\n",
            "184/184 [==============================] - 2s 9ms/step - loss: 2.9839e-05 - mae: 0.0054 - val_loss: 1.6582e-04 - val_mae: 0.0143\n",
            "Epoch 194/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.1207e-05 - mae: 0.0044 - val_loss: 1.0762e-04 - val_mae: 0.0111\n",
            "Epoch 195/500\n",
            "184/184 [==============================] - 2s 8ms/step - loss: 2.3044e-05 - mae: 0.0048 - val_loss: 1.4368e-04 - val_mae: 0.0139\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7fc122834208>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kVkhv3sk-LlA",
        "colab_type": "text"
      },
      "source": [
        "## Make Predictions"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "nPeZUfQy8OJQ",
        "colab": {}
      },
      "source": [
        "# recall best model\n",
        "model = keras.models.load_model(\"my_checkpoint\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "4tFrq5uW8OJR",
        "colab": {}
      },
      "source": [
        "# make predictions\n",
        "rnn_forecast = model.predict(normalized_x_test[np.newaxis,:])\n",
        "rnn_forecast = rnn_forecast.flatten()\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gsSbjyD8r66o",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "e75ed9ed-3968-4028-9441-61e3b6ecc067"
      },
      "source": [
        "# Example of how to iverse\n",
        "rnn_unscaled_forecast = x_train_scaler.inverse_transform(rnn_forecast.reshape(-1,1)).flatten()\n",
        "rnn_unscaled_forecast.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(422,)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 22
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "ZfaR6nqj8OJT",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 410
        },
        "outputId": "d90cbcf3-3a0e-4e50-d995-89ef602779a9"
      },
      "source": [
        "# plot results\n",
        "plt.figure(figsize=(10,6))\n",
        "\n",
        "plt.ylabel('Dollars $')\n",
        "plt.xlabel('Timestep in Days')\n",
        "plt.title(f'LSTM {window_size} Forecast')\n",
        "plot_series(x_test.index, x_test, label=\"Actual\")\n",
        "plot_series(x_test.index, rnn_unscaled_forecast, label=\"Forecast\")\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 720x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "Wgf2u2Tp8OJV",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "21687791-1ebf-4eba-8686-cf159d82ea24"
      },
      "source": [
        "# calculate MAE\n",
        "keras.metrics.mean_absolute_error(x_test, rnn_unscaled_forecast).numpy()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1.1875452"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 25
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yRBHsgYm3qhT",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}