Skip to content

Latest commit

 

History

History
21 lines (13 loc) · 779 Bytes

README.rst

File metadata and controls

21 lines (13 loc) · 779 Bytes

phynix_gym

Environments to train reinforcement learning agents.

Currently contains one environment, Minimize1DSimple, minimizing negative log-likelihood fit to a gaussian sample. The environment takes a couple of configuration arguments and should therefore be instantiated from the class. It adheres to the openai gym API and can also be instantiated with `gym.make("minimize-1d-simple-v0").

Example:

from phynix_gym import Minimize1DSimple

env = Minimize1DSimple()

nll, mu, mu_grad, sigma, sigma_grad = env.reset()  # return state
state, reward, done, info = env.step([0.1, 0.3])  # corresponds to [mu, sigma]
env.render()  # shows a nll contour plot, the position, gradient and the minimum