Hands-On Q-Learning with Python

Backpropagation

We use the sum of squares formula to compute the loss function, as follows:

Here, y - y_pred is the difference between the predicted and actual outputs.

We can implement it as follows:

loss = np.square(y_pred - y).sum()

We're now calculating the gradient descent with respect to the loss function. The gradient of a function, such as the derivative, is the slope of the tangent of the graph of that function. Gradient descent finds the steepest slope and adjusts the weights according to that slope:

The preceding diagram makes gradient ...