ARS_tfv2

Requirements

• pybullet
• tensorflow v2.5
• ray
• gym

Usage

$ python main.py --help

$ python main.py --train
$ python main.py --play [weight_file_path]

usage: main.py [-h] [--train] [--play PLAY]

ARS

optional arguments:
  -h, --help   show this help message and exit
  --train      Train agent with given environment
  --play PLAY  Play with a given weight directory

Config

You can change config.py to fit your own flags.

hdims             # dimension of hidden layers
nu                # standard deviation of noise
actv = 'tanh'     # activation function
out_actv = 'tanh' # activation function

# ray
n_cpu = n_workers # number of cpu
b = (n_workers//5)# number of top-performing directions to use

# Update
total_steps
ep_len_rollout
alpha             # step size

# Evaluate
max_ep_len_eval
num_eval
evaluate_every
print_every