sine wave lstm

sinwave_lstm.py

#fork from https://github.com/jaungiers/LSTM-Neural-Network-for-Time-Series-Prediction
#minor fix, simpler model and normalize algorithm

import os
import time
import warnings
import numpy as np
from numpy import newaxis
from keras.layers.core import Dense, Activation, Dropout
from keras.layers.recurrent import LSTM
from keras.models import Sequential

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' #Hide messy TensorFlow warnings
warnings.filterwarnings("ignore") #Hide messy Numpy warnings

def load_data(filename, seq_len, normalise):
    f = open(filename, 'rb').read()
    data = f.decode().split('\n')
    
    data = [float(d) if len(d) > 0 else 0 for d in data]
    
   
    base = 0
    if normalise:
        base = max(data)
        data = [d/base for d in data]

    sequence_length = seq_len + 1
    result = []
    for index in range(len(data) - sequence_length):
        result.append(data[index: index + sequence_length])

    result = np.array(result, dtype=np.float64)
    
    row = round(0.9 * result.shape[0])        
 
   
    train = result[:int(row), :]
    np.random.shuffle(train)
    x_train = train[:, :-1]
    y_train = train[:, -1]
    x_test = result[int(row):, :-1]
    y_test = result[int(row):, -1]
    
    x_train = np.reshape(x_train, (x_train.shape[0], x_train.shape[1], 1))
    x_test = np.reshape(x_test, (x_test.shape[0], x_test.shape[1], 1))  
    
    return [base, x_train, y_train, x_test, y_test]

def build_model(input_shape):
    model = Sequential()
    model.add(LSTM(input_shape[0], input_shape=input_shape))
    model.add(Dense(1))
    model.compile(loss="mse", optimizer="rmsprop")
    
    return model

def predict_point_by_point(model, data):
    #Predict each timestep given the last sequence of true data, in effect only predicting 1 step ahead each time
    predicted = model.predict(data)
    predicted = np.reshape(predicted, (predicted.size,))
    return predicted

def predict_sequence_full(model, data):
    #Shift the window by 1 new prediction each time, re-run predictions on new window
    curr_frame = data[0]
    predicted = []
    for i in range(len(data)):
        predicted.append(model.predict(curr_frame[newaxis,:,:])[0,0])
        curr_frame = curr_frame[1:]
        curr_frame = np.insert(curr_frame, len(curr_frame), predicted[-1], axis=0)
    return predicted

def predict_sequences_multiple(model, data, prediction_len):
    #Predict sequence of 50 steps before shifting prediction run forward by 50 steps
    prediction_seqs = []
    for i in range(int(len(data)/prediction_len)):
        curr_frame = data[i*prediction_len]
        predicted = [] 
        for j in range(prediction_len):
            predicted.append(model.predict(curr_frame[newaxis,:,:])[0,0])
            curr_frame = curr_frame[1:]
            curr_frame = np.insert(curr_frame, len(curr_frame), predicted[-1], axis=0)
        prediction_seqs.append(predicted)
    return prediction_seqs


import time
import matplotlib.pyplot as plt

def plot_results(predicted_data, true_data):
    fig = plt.figure(facecolor='white')
    ax = fig.add_subplot(111)
    ax.plot(true_data, label='True Data', color="green")
    ax.plot(predicted_data, label='Prediction', color="blue")
    plt.legend()
    plt.show()

def plot_results_multiple(predicted_data, true_data, prediction_len):
    fig = plt.figure(facecolor='white')
    ax = fig.add_subplot(111)
    ax.plot(true_data, label='True Data', color="green")
    #Pad the list of predictions to shift it in the graph to it's correct start
    for i, data in enumerate(predicted_data):
        padding = [None for p in range(i * prediction_len)]
        ax.plot(padding + data.tolist(), label='Prediction', color="blue")
        #plt.legend()
    plt.show()

print('loading... ')

epochs  = 500
seq_len = 50
base, X_train, y_train, X_test, y_test = load_data('sinwave.csv', seq_len, True)

model = build_model((seq_len, 1))

print('trainning...')

history = model.fit(
    X_train,
    y_train,
    batch_size=64,
    nb_epoch=epochs,
    validation_split=0.05,
    verbose=False)

plt.plot(history.history['loss'])
plt.show()

print("last error", history.history['loss'][-1])

predict_len=7

predictions = predict_sequences_multiple(model, X_test, predict_len)
predictions = np.array(predictions)
plot_results_multiple(base * predictions, base * y_test, predict_len)

predicted = predict_sequence_full(model, X_test)
plot_results(base * np.array(predicted), base * y_test)

predicted = predict_point_by_point(model, X_test)        
plot_results(base * np.array(predicted), base * y_test)