If you open the forward_propagation.py you can see the code that is loading and printing the data. We have also implemented output_layer_fp function that is performing the forward propagation for the output layer. This was your task in the previous scenario.

The function is using numpy matmul function to perform matrix multiplication. As an activation, we used sigmoid provided in the helper.py script. Instead of the data it takes any input. In case of our architecture, this would be the output from the hidden layer.

You will use this function further when building the whole network architecture and performing forward propagation to predict the labels.

Hidden layers are the ones that are placed between the input and the output. They are usually not really hidden, which means you have full access to them. The name is more historical than anything.

Unlike the output layer, hidden ones are not usually activated using the sigmoid function. Instead, there are applied other like tanh or reLU.

Task 1

You are supposed to implement the hidden_layer_fp. The code should be similar to the one written for the output. The difference is the activation function. You can use the np.tanh for this.

Finally, we can perform the Forward Propagation for the whole network. First, we will prepare the input data, weights and biases so that they can be used in the process.

Task 2

Variable X was set to only contain the feature data. What you need to do is to initiate the weights and biases with the values of your choice. Remember to create the arrays with proper shapes, provided in the code.

Task 3

Once we have both types of layers, data and the weights, we can perform the Forward Propagation. Fill in the forward_propagation function using the following flow: X -> hidden(hidden_layer_fp) -> output(output_layer_fp).

As usual, once your code is ready you can run it with the following command:

python forward_propagation.py