Project - Stock Closing Price Prediction using Deep Learning, TensorFlow2 & Keras

28 / 32
Previous Index Next  

Building the Model

Finally, we have reached the modeling part.

As discussed previously, we shall use a GRU based model.

  • Firstly, let us import the necessary TensorFlow and Scikit-Learn libraries.

  • Next, we shall build the model by adding layers, compiling it, and then fitting the model on the train data.

A bit about the model we are going to build:

  • The model we are going to use is a stacked GRU.
  • So the output of one GRU layer should act as the input to the next GRU layer stacked above the former one and so on.
  • We have already discussed that the GRU/LSTM layers in Keras expect the input to be in three-dimensional.
  • Thus, we have to make sure that the output from a previous layer is formatted in a three-dimensional way so as to provide it as input to the next layer.
  • This job could be achieved by setting return_sequences=True in the GRU layers whose output would potentially act as the input to the next GRU layer.
INSTRUCTIONS

  • Import the below libraries.

    import tensorflow as tf
tf.random.set_seed(42)

from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.optimizers import Adam

  • Initialize model to a sequential model from keras.

    model = keras.<< your code comes here >>

  • Add the following layers to the model.

    # First GRU layer
model.add(layers.GRU(units=100, return_sequences=True, input_shape=(1,n_features), activation='tanh'))
model.add(layers.Dropout(0.2))

# Second GRU layer
model.add(layers.GRU(units=150, return_sequences=True, input_shape=(1,n_features), activation='tanh'))
model.add(layers.Dropout(0.2))

# Third GRU layer
model.add(layers.GRU(units=100, activation='tanh'))
model.add(layers.Dropout(0.2))

# The output layer
model.add(layers.Dense(units=1, kernel_initializer='he_uniform', activation='linear'))

    Observe the argument return_sequences, which is set to be True only for those layers which have a GRU layer after them(that is, the first and second GRU layers), unlike the third layer. This is so because the output of the third layer would be fed to a Dense layer but not a GRU/LSTM layer.

  • Compile the model as follows, by mentioning the learning rate lr=0.0005, loss='mean_squared_error' and metrics = ['mean_squared_error'].

    model.compile(loss='mean_squared_error', optimizer=Adam(lr = 0.0005) , metrics = ['mean_squared_error'])

  • Let us now see the summary of the model architecture.

    print(model.summary())

  • Use fit method of the model to start training on the train data. Use validation_data = (valX,valY). 

    history = model.fit(trainX,trainY,epochs=100,batch_size=128, verbose=1, validation_data = (valX,valY))
Get Hint See Answer


Note - Having trouble with the assessment engine? Follow the steps listed here

Previous Index Next  

Loading comments...