Baichenjia

import glob
import os
from gym import envs
import shutil

# collect all valid env names
envnames = []
for env_spec in envs.registry.all():
    name = env_spec.id
    if 'ram' not in name and "-v4" in name and 'Deterministic' not in name:

Baichenjia

import re

x = r""

x = re.sub("~", "", x)
x = re.sub(r"\$[\\0-9a-zA-Z\{\},()\[\]_+-=\^|\s*%']*\$", "EQ", x)
x = re.sub(r"emph{", "", x)
x = re.sub(r"ref\s*}", "1", x)
x = re.sub(r"\\citep\{[\w\s\d,-]*\}", "", x)
x = re.sub(r"\\cite\{[\w\s\d,]*\}", "", x)

Baichenjia

import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
config = tf.ConfigProto()
config.gpu_options.visible_device_list = '0'
config.gpu_options.allow_growth = True
tf.enable_eager_execution(config=config)

PRIOR_SCALE = 2.

Baichenjia

# Paper: Szita I, Lörincz A. Learning Tetris using the noisy cross-entropy method.[J]. Neural Computation, 2006, 18(12):2936.
# code: https://gym.openai.com/evaluations/eval_HIz0KjtWSvW06yKKPiaF5A

# annotation
# 1.Linear function approximation, the number of parameters to learn is the feature dimension of the state s + 1(bias)
# 2.Cross-Entropy Method is an evolutionary algorithm that searches for the optimal parameters by iterating. 
    First, “batch_size” vectors are sampled from a normal distribution of initial \mu and \sigma parameters,
    then these parameters are evaluated by a evaluate function, pip the top-n parameter vectors ordered by the evaluation result.
    use the new n_elite parameters to estimate the new \mu and \sigma.
# 3. Iterate through the method 2 to update the values of \mu and \sigma.