Skip to content

Commit

Permalink
Created using Colaboratory
Browse files Browse the repository at this point in the history
  • Loading branch information
@sanikamal
sanikamal committed Oct 24, 2019
1 parent 8f2d98d commit 2a13497
Showing 1 changed file with 340 additions and 0 deletions.
340 changes: 340 additions & 0 deletions forecasting_with_lstm.ipynb
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,340 @@
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"accelerator": "GPU",
"colab": {
"name": "forecasting_with_lstm.ipynb",
"provenance": [],
"collapsed_sections": [],
"toc_visible": true,
"include_colab_link": true
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
}
},
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "view-in-github",
"colab_type": "text"
},
"source": [
"<a href=\"https://colab.research.google.com/github/sanikamal/time-series-analysis-and-forecasting-atoz/blob/master/forecasting_with_lstm.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
]
},
{
"cell_type": "markdown",
"metadata": {
"colab_type": "text",
"id": "E5VI4y76i14x"
},
"source": [
"# Forecasting with an LSTM"
]
},
{
"cell_type": "markdown",
"metadata": {
"colab_type": "text",
"id": "vidayERjaO5q"
},
"source": [
"## Setup"
]
},
{
"cell_type": "code",
"metadata": {
"id": "WQ5AjH_KY74O",
"colab_type": "code",
"colab": {}
},
"source": [
"# !pip install --upgrade tensorflow"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "gqWabzlJ63nL",
"colab": {}
},
"source": [
"import numpy as np\n",
"import matplotlib.pyplot as plt\n",
"import tensorflow as tf\n",
"\n",
"keras = tf.keras"
],
"execution_count": 0,
"outputs": []
},
{
"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",
"def trend(time, slope=0):\n",
" return slope * time\n",
" \n",
" \n",
"def seasonal_pattern(season_time):\n",
" \"\"\"Just an arbitrary pattern, you can change it if you wish\"\"\"\n",
" return np.where(season_time < 0.4,\n",
" np.cos(season_time * 2 * np.pi),\n",
" 1 / np.exp(3 * season_time))\n",
"\n",
" \n",
"def seasonality(time, period, amplitude=1, phase=0):\n",
" \"\"\"Repeats the same pattern at each period\"\"\"\n",
" season_time = ((time + phase) % period) / period\n",
" return amplitude * seasonal_pattern(season_time)\n",
" \n",
" \n",
"def white_noise(time, noise_level=1, seed=None):\n",
" rnd = np.random.RandomState(seed)\n",
" return rnd.randn(len(time)) * noise_level\n",
" \n",
"\n",
"def sequential_window_dataset(series, window_size):\n",
" series = tf.expand_dims(series, axis=-1)\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": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "iL2DDjV3lel6",
"colab": {}
},
"source": [
"time = np.arange(4 * 365 + 1)\n",
"\n",
"slope = 0.05\n",
"baseline = 10\n",
"amplitude = 40\n",
"series = baseline + trend(time, slope) + seasonality(time, period=365, amplitude=amplitude)\n",
"\n",
"noise_level = 5\n",
"noise = white_noise(time, noise_level, seed=42)\n",
"\n",
"series += noise\n",
"\n",
"plt.figure(figsize=(10, 6))\n",
"plot_series(time, series)\n",
"plt.show()"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "Zmp1JXKxk9Vb",
"colab": {}
},
"source": [
"split_time = 1000\n",
"time_train = time[:split_time]\n",
"x_train = series[:split_time]\n",
"time_valid = time[split_time:]\n",
"x_valid = series[split_time:]"
],
"execution_count": 0,
"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": 0,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"colab_type": "text",
"id": "EPjK0l9P8OJM"
},
"source": [
"## LSTM RNN Forecasting"
]
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "cSoUmW-x8OJN",
"colab": {}
},
"source": [
"keras.backend.clear_session()\n",
"tf.random.set_seed(42)\n",
"np.random.seed(42)\n",
"\n",
"window_size = 30\n",
"train_set = sequential_window_dataset(x_train, window_size)\n",
"\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",
" keras.layers.Lambda(lambda x: x * 200.0)\n",
"])\n",
"lr_schedule = keras.callbacks.LearningRateScheduler(\n",
" lambda epoch: 1e-8 * 10**(epoch / 20))\n",
"reset_states = ResetStatesCallback()\n",
"optimizer = keras.optimizers.SGD(lr=1e-8, momentum=0.9)\n",
"model.compile(loss=keras.losses.Huber(),\n",
" optimizer=optimizer,\n",
" metrics=[\"mae\"])\n",
"history = model.fit(train_set, epochs=100,\n",
" callbacks=[lr_schedule, reset_states])"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "KA0GM9sQ8OJO",
"colab": {}
},
"source": [
"plt.semilogx(history.history[\"lr\"], history.history[\"loss\"])\n",
"plt.axis([1e-8, 1e-4, 0, 30])"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "hiHR5pPL8OJP",
"colab": {}
},
"source": [
"keras.backend.clear_session()\n",
"tf.random.set_seed(42)\n",
"np.random.seed(42)\n",
"\n",
"window_size = 30\n",
"train_set = sequential_window_dataset(x_train, window_size)\n",
"valid_set = sequential_window_dataset(x_valid, window_size)\n",
"\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",
" keras.layers.Lambda(lambda x: x * 200.0)\n",
"])\n",
"optimizer = keras.optimizers.SGD(lr=5e-7, momentum=0.9)\n",
"model.compile(loss=keras.losses.Huber(),\n",
" optimizer=optimizer,\n",
" metrics=[\"mae\"])\n",
"reset_states = ResetStatesCallback()\n",
"model_checkpoint = keras.callbacks.ModelCheckpoint(\n",
" \"my_checkpoint.h5\", save_best_only=True)\n",
"early_stopping = keras.callbacks.EarlyStopping(patience=50)\n",
"model.fit(train_set, epochs=500,\n",
" validation_data=valid_set,\n",
" callbacks=[early_stopping, model_checkpoint, reset_states])"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "nPeZUfQy8OJQ",
"colab": {}
},
"source": [
"model = keras.models.load_model(\"my_checkpoint.h5\")"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "4tFrq5uW8OJR",
"colab": {}
},
"source": [
"rnn_forecast = model.predict(series[np.newaxis, :, np.newaxis])\n",
"rnn_forecast = rnn_forecast[0, split_time - 1:-1, 0]"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "ZfaR6nqj8OJT",
"colab": {}
},
"source": [
"plt.figure(figsize=(10, 6))\n",
"plot_series(time_valid, x_valid)\n",
"plot_series(time_valid, rnn_forecast)"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"colab_type": "code",
"id": "Wgf2u2Tp8OJV",
"colab": {}
},
"source": [
"keras.metrics.mean_absolute_error(x_valid, rnn_forecast).numpy()"
],
"execution_count": 0,
"outputs": []
}
]
}

0 comments on commit 2a13497

Please sign in to comment.