train.py

import matplotlib.pyplot as plt
import numpy as np
import time
import tensorflow as tf
import pickle
import wfdb
from sklearn.utils import class_weight
from sklearn.model_selection import train_test_split

# Hyper-parameters
sequence_length = 240
epochs = 1000#int(input('Enter Number of Epochs (or enter default 1000): '))
FS = 100.0

def z_norm(result):
    result_mean = np.mean(result)
    result_std = np.std(result)
    result = (result - result_mean) / result_std
    return result

def split_data(X):
    X1 = []
    X2 = []
    for index in range(len(X)):
        X1.append([X[index][0], X[index][1]])
        X2.append([X[index][2], X[index][3]])

    return np.array(X1).astype('float64'), np.array(X2).astype('float64')

def get_data():
    with open('train_input.pickle','rb') as f: 
        X_train = np.asarray(pickle.load(f))
    with open('train_label.pickle','rb') as f: 
        y_train = np.asarray(pickle.load(f))
    with open('val_input.pickle','rb') as f: 
        X_val = np.asarray(pickle.load(f))
    with open('val_label.pickle','rb') as f: 
        y_val = np.asarray(pickle.load(f))
    with open('test_input.pickle','rb') as f: 
        X_test = np.asarray(pickle.load(f))
    with open('test_label.pickle','rb') as f: 
        y_test = np.asarray(pickle.load(f))


    #X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
    '''
    X_train = X_train[:, 0, :]
    X_test = X_test[:, 0, :]
    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))
    '''
    X_train1, X_train2 = split_data(X_train)
    X_val1, X_val2 = split_data(X_val)
    X_test1, X_test2 = split_data(X_test)

    X_train1 = np.transpose(X_train1, (0, 2, 1))
    #X_train2 = np.reshape(X_train2, (X_train2.shape[0], X_train2.shape[1], 1))
    X_test1 = np.transpose(X_test1, (0, 2, 1))
    #X_test2 = np.reshape(X_test2, (X_test2.shape[0], X_test2.shape[1], 1))
    X_val1 = np.transpose(X_val1, (0, 2, 1))
    return X_train1, X_train2, y_train, X_val1, X_val2, y_val, X_test1, X_test2, y_test



def build_model():
    
    layers = {'input': 2, 'hidden1': 256, 'hidden2': 256, 'hidden3': 256, 'output': 1}
    x1 = tf.keras.layers.Input(shape=(sequence_length, layers['input']))
    m1 = tf.keras.layers.LSTM(layers['hidden1'],                   
                    recurrent_dropout=0.5,
                   return_sequences=True)(x1)
    m1 = tf.keras.layers.LSTM(
            layers['hidden2'],
            recurrent_dropout=0.5,
            return_sequences=True)(m1)

    m1 = tf.keras.layers.LSTM(
            layers['hidden3'],
            recurrent_dropout=0.5,
            return_sequences=False)(m1)

    x2 = tf.keras.layers.Input(shape=(2,))
    m2 = tf.keras.layers.Dense(32)(x2)

    #merged = Merge([model1, model2], mode='concat')
    merged = tf.keras.layers.Concatenate(axis=1)([m1, m2])

    out = tf.keras.layers.Dense(8)(merged)
    out = tf.keras.layers.Dense(layers['output'], kernel_initializer='normal')(out)
    out = tf.keras.layers.Activation("sigmoid")(out)
    
    model = tf.keras.models.Model(inputs=[x1, x2], outputs=[out])

    start = time.time()
    model.compile(loss="binary_crossentropy", optimizer="adam",
                  metrics = ['accuracy'])
    print ("Compilation Time : ", time.time() - start)

    model.summary()
    return model


def run_network(model=None, data=None):
    global_start_time = time.time()

    print ('\nData Loaded. Compiling...\n')
    print('Loading data... ')
    X_train1, X_train2, y_train, X_val1, X_val2, y_val, X_test1, X_test2, y_test = get_data()

    class_w = class_weight.compute_class_weight(class_weight='balanced',
                                                     classes=np.unique(y_train),
                                                     y=y_train)

    print (class_w)

    if model is None:
        model = build_model()

    try:
        print("Training")

        class_w = {i : class_w[i] for i in range(2)}
        callback = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=3)
        history = model.fit([X_train1, X_train2], y_train, 
                            validation_data=([X_val1, X_val2], y_val),
                            callbacks=[callback],
                             epochs=epochs, batch_size=256, class_weight=class_w)

        import matplotlib.pyplot as plt
        '''
        plt.plot(history.losses)
        plt.ylabel('loss')
        plt.xlabel('epoch')
        plt.legend(['train'], loc='upper left')
        plt.show()
        '''
        # Evaluate Model
        y_pred = model.predict([X_test1, X_test2])
        scores = model.evaluate([X_test1, X_test2], y_test)
        print("%s: %.2f%%" % (model.metrics_names[1], scores[1] * 100))


    except KeyboardInterrupt:
        print("prediction exception")
        print ('Training duration (s) : ', time.time() - global_start_time)
        return model


    print ('Training duration (s) : ', time.time() - global_start_time)

    return model

run_network()