Reinforcement Learning with TensorFlow

Use of experience replay

Another important feature added to deep Q-networks is experience replay. The idea behind this feature is that the agent can store its past experiences and use them in batches to train the deep neural network. Storing the experiences allows the agent to randomly draw the batches and help the network to learn from a variety of data instead of just formalizing decisions on immediate experiences. Each of these experiences are stored in a form of four dimensional vector comprising of state, action, reward, and next state.

In order to avoid storage issues, the buffer of experience replay is fixed, and as the new experiences get stored the old experiences get removed. For training neural networks, uniform batches of random ...