linear_regress_TF2_4.py

num_features = X.shape[1]
weights = tf.random.normal( ( num_features , 1 ) ) 
bias = 0

epochs_plot = list()
loss_plot = list()

for i in range( num_epochs ) :
    
    epoch_loss = list()
    for b in range( int(num_samples/batch_size) ):
        x_batch , y_batch = iterator.get_next()
   
        output = h( x_batch , weights , bias ) 
        loss = epoch_loss.append( mean_squared_error( y_batch , output ).numpy() )
    
        dJ_dH = mean_squared_error_deriv( y_batch , output)
        dH_dW = x_batch
        dJ_dW = tf.reduce_mean( dJ_dH * dH_dW )
        dJ_dB = tf.reduce_mean( dJ_dH )
    
        weights -= ( learning_rate * dJ_dW )
        bias -= ( learning_rate * dJ_dB ) 
        
    loss = np.array( epoch_loss ).mean()
    epochs_plot.append( i + 1 )
    loss_plot.append( loss ) 
    
    print( 'Loss is {}'.format( loss ) )