I’m new to ML, and was wondering if it is possible to acces the data from each hidden layer each step as it is possible to acces all the observation and action data.
Do you mean you’d like to access the weights/biases of the network layers of the policy/value function during training?
@andrewcoh_unity I would like to visualize the network while it is training. I am giving the network a 64x64 grid sensor as input, which I am able to visualize as an image trough the python API from the observations array. I would like to do that for each layer to see how it interprets the inputs and which neurons fire. So i guess I would need acess to each neurons weights/biases.