Hands-On Machine Learning for Algorithmic Trading

Putting it all together

To prepare for the training of our network, we create a function that combines the preceding gradient definition and computes the relevant weight and bias updates from the training data and labels, and the current weight and bias values, as follows:

def compute_gradients(X, y_true, w_h, b_h, w_o, b_o):    """Evaluate gradients for parameter updates"""    # Compute hidden and output layer activations    hidden_activations = hidden_layer(X, w_h, b_h)    y_hat = output_layer(hidden_activations, w_o, b_o)    # Compute the output layer gradients    loss_grad = loss_gradient(y_hat, y_true)    out_weight_grad = output_weight_gradient(hidden_activations, loss_grad)    out_bias_grad = output_bias_gradient(loss_grad) # Compute the hidden layer gradients ...