Deep learning models for time series forecasting

BiLSTM.py

### Create the Bidirectional LSTM model
from random import random
from numpy import array
from numpy import cumsum
from keras.models import Sequential
from tensorflow.keras.layers import Dense,Dropout
from keras.layers import TimeDistributed
from keras.layers import Bidirectional
from tensorflow.keras.layers import LSTM
import tensorflow as tf
# Set random seed for reproducibility: get the same result after each time running the model
tf.random.set_seed(1234)

# # Build the BiLSTM model
modell = Sequential()
modell.add(Bidirectional(LSTM(100, activation='relu', input_shape=(100,1)))) #elu
#modell.add(Bidirectional(LSTM(50, dropout=0.5)))
#modell.add(Bidirectional(LSTM(100, dropout=0.5)))
#modell.add(BatchNormalization(momentum=0.6))
modell.add(Dense(1))
modell.compile(loss='mean_squared_error', optimizer='adam') #rmsprop | adam


# use early stop to avoid over-fiting 
from tensorflow.keras.callbacks import EarlyStopping

monitor = EarlyStopping(monitor='val_loss', min_delta=1e-3, patience=30, 
        verbose=1, mode='auto', restore_best_weights=True)
# fit
history1=modell.fit(X_train,y_train,validation_data=(X_test,ytest),
        callbacks=[monitor],verbose=1,epochs=1000)

modell.summary()
# Output:
# Model: "sequential_2"
# _________________________________________________________________
#  Layer (type)                Output Shape              Param #   
# =================================================================
#  bidirectional_2 (Bidirectio  (None, 200)              81600     
#  nal)                                                            
                                                                 
#  dense_2 (Dense)             (None, 1)                 201       
                                                                 
# =================================================================
# Total params: 81,801
# Trainable params: 81,801
# Non-trainable params: 0
# _________________________________________________________________

# save the best model 
#modell.save("bilstm.h5")
# load the best model 
#new_model = tf.keras.models.load_model('bilstm.h5')