Keras Deep Learning Cookbook

init method

The init method takes the following parameters:

• model__: A Keras model.
• policy__: A keras-rl policy that is defined in (policy) (https://github.com/keras-rl/keras-rl/blob/master/rl/policy.py).
• test_policy__: A keras-rl policy.
• enable_double_dqn__: A Boolean that enables the target network as a second network, proposed by van Hasselt et al, to decrease overfitting.
• enable_dueling_dqn__: A Boolean that enables the dueling architecture, proposed by Mnih et al [2].
• dueling_type__: If enable_dueling_dqn is set to True, a type of dueling architecture must be chosen that calculates Q(s,a) from V(s) and A(s,a) differently. Note that avg is recommended in the (paper) (https://arxiv.org/abs/1511.06581). avg: Q(s,a;theta) = V(s;theta) + ...