digideep.agent package¶

Submodules¶

This module is dedicated to noise models used in other methods.

Each noise class should implement the __call__ method. See the examples EGreedyNoise and OrnsteinUhlenbeckNoise.

class digideep.agent.noises.EGreedyNoise(**params)[source]¶

This class implements simple e-greedy noise. The noise is sampled from uniform distribution.

Parameters:	std (python:float) – Standard deviation of the noise. e (python:float) – The probability of choosing a noisy action. lim (python:float) – Boundary of the noise (noise will be clipped beyond this value.)

Note

This class is not dependant on its history.

class digideep.agent.noises.OrnsteinUhlenbeckNoise(**params)[source]¶

An implementation of the Ornstein-Uhlenbeck noise.

The noise model is \({\displaystyle dx_{t}= heta (\mu -x_{t})\,dt+\sigma \,dW_{t}}\).

Parameters:	mu – Parameter \(\mu\) which indicates the final value that \(x\) will converge to. theta – Parameter :math:` heta`. sigma – Parameter \(\sigma\) which is the std of the additional normal noise. lim – The action limit, which can be a `np.array` for a vector of actions.

Note

This class is state serializable.