PyTorch 1.x Reinforcement Learning Cookbook

How it works...

In Step 2, the DQN class takes in four parameters: the number of input states and output actions, the number of hidden nodes (we herein just use one hidden layer as an example), and the learning rate. It initializes a neural network with one hidden layer, followed by a ReLU activation function. It takes in n_state units and generates one n_action output, which are the predicted state values for individual actions. An optimizer, Adam, is initialized along with each linear model. The loss function is the mean squared error.

Step 3 is for updating the network: given a training data point, the predictive result, along with the target value, is used to compute the loss and gradients. The neural network model is then updated via ...