mirror of
https://github.com/wassname/kair_algorithms_draft.git
synced 2026-08-25 11:19:30 +08:00
Kyunghwan Kim
d2769dfa9d
* Convert code format to python2.7 (SAC) * Convert code format python2.7 (TD3, all fD) * Remove no use import and black setting * Change SAC param * Change env name Reacher-v2 to v1 * Remove old version reacher training script * Convert code format python2.7 * Modify .travis.yml * Add install command python3.6 & black on Makefile * Fix seperator to tab on Makefile * Modify Makefile * Fix little error * Change td3 gamma parameter
59 lines
1.6 KiB
Python
59 lines
1.6 KiB
Python
# -*- coding: utf-8 -*-
|
|
"""Noise classes for algorithms."""
|
|
|
|
import copy
|
|
import random
|
|
|
|
import numpy as np
|
|
|
|
|
|
class GaussianNoise:
|
|
"""Gaussian Noise.
|
|
|
|
Taken from https://github.com/vitchyr/rlkit
|
|
"""
|
|
|
|
def __init__(self, action_dim, min_sigma=1.0, max_sigma=1.0, decay_period=1000000):
|
|
"""Initialization."""
|
|
self.action_dim = action_dim
|
|
self.min_sigma = min_sigma
|
|
self.max_sigma = max_sigma
|
|
self.decay_period = decay_period
|
|
|
|
def sample(self, t=0):
|
|
"""Get an action with gaussian noise."""
|
|
sigma = self.max_sigma - (self.max_sigma - self.min_sigma) * min(
|
|
1.0, t / self.decay_period
|
|
)
|
|
return np.random.normal(0, sigma, size=self.action_dim)
|
|
|
|
|
|
class OUNoise:
|
|
"""Ornstein-Uhlenbeck process.
|
|
|
|
Taken from Udacity deep-reinforcement-learning github repository:
|
|
https://github.com/udacity/deep-reinforcement-learning/blob/master/
|
|
ddpg-pendulum/ddpg_agent.py
|
|
"""
|
|
|
|
def __init__(self, size, mu=0.0, theta=0.15, sigma=0.2):
|
|
"""Initialize parameters and noise process."""
|
|
self.state = np.float64(0.0)
|
|
self.mu = mu * np.ones(size)
|
|
self.theta = theta
|
|
self.sigma = sigma
|
|
self.reset()
|
|
|
|
def reset(self):
|
|
"""Reset the internal state (= noise) to mean (mu)."""
|
|
self.state = copy.copy(self.mu)
|
|
|
|
def sample(self):
|
|
"""Update internal state and return it as a noise sample."""
|
|
x = self.state
|
|
dx = self.theta * (self.mu - x) + self.sigma * np.array(
|
|
[random.random() for _ in range(len(x))]
|
|
)
|
|
self.state = x + dx
|
|
return self.state
|