renom_rl.utility

Utilities (such as EpsilonGreedyFilter etc.) for the ReNomRL Modules are located in this section. There are modules that require these utilities. There are also modules which users can use to modify algorithms, such as how epsilon increases during training.

renom_rl.utility.animation

Animation allows users to create animation. Mainly developed for jupyter notebook.

renom_rl.utility.logger

Logger allows users to custom logging data. By modifying logger and implementing it to the algorithms, users can view their own original logging data, save logged data to csv or view graph data for immediate check.

renom_rl.utility.filter

Filters are used to add noise or choose random actions. In the renom_rl.utility.filter , there are various modules to design filter. Action Filters are used for random action decision in algorithms such as DQN, DDQN etc. Action Noise Filters are used for random action noises in algorithms such as DDPG etc. Epsilon are used to change epsilon value during epsilon greedy decision etc. Noise are used to generate noise. Action Range is used for limit sampling value. Action Sampler generates samples based on its input.

renom_rl.utility.graidents

Gradient Clipping is used to supress gradient from updating weights drastically.