hiders = 1
seekers = 1
boxes = 1
ramps = 1
food = 0
rooms = 2
display = False
load_weights = True
save_weights = False
debug = False
env = hide_and_seek.make_env(n_hiders=hiders,
n_seekers=seekers,
n_boxes=boxes,
n_ramps=ramps,
n_food=food,
n_rooms=rooms,
n_lidar_per_agent=30,
visualize_lidar=True)
# # probably shouldn't use those two. but was testing.
# rewardWrapper = hide_and_seek.HideAndSeekRewardWrapper(env, n_hiders=hiders, n_seekers=seekers)
# trackStatW = hide_and_seek.TrackStatWrapper(env, boxes, ramps, food)
# run one episode
env.seed(42)
env.reset()
agents = []
# https://github.com/keras-rl/keras-rl/blob/master/rl/core.py
import numpy as np
import tensorflow as tf
tf.compat.v1.disable_eager_execution()
# environment parameters (there's more). Check make_env
hiders = 1
seekers = 1
boxes = 1
ramps = 1
food = 0
rooms = 2
display = True
load_weights = False
env = hide_and_seek.make_env(n_hiders=hiders, n_seekers=seekers, n_boxes=boxes, n_ramps=ramps, n_food=food, n_rooms=rooms)
# # probably shouldn't use those two. but was testing.
# rewardWrapper = hide_and_seek.HideAndSeekRewardWrapper(env, n_hiders=hiders, n_seekers=seekers)
# trackStatW = hide_and_seek.TrackStatWrapper(env, boxes, ramps, food)
# run one episode
env.seed(42)
env.reset()
agents = []
for i in range(hiders+seekers):
agents.append(get_agent(env,i))
if load_weights:
agents[-1].load_weights("agent_%i_weights.h5f"%(i))
def test_ppo(exp_dir, itr='last'):
_, get_action, lstm = load_policy_and_env(exp_dir, itr=itr)
env = hide_and_seek.make_env()
run_policy(env, get_action, lstm=lstm)
import tensorflow as tf
import gym
import time
from ppo_pkg.ppo import ppo
from mae_envs.envs import hide_and_seek
from ma_policy.ma_policy import MAPolicy
from testing.test_policy import test_ppo
env_fn = lambda: hide_and_seek.make_env() # choose desired environment
dir_str = '../Testing/exp/' # set output directory
now_str = time.asctime(
time.localtime()) # get current time stamp and set logger dict
logger_kwargs = dict(
output_dir=dir_str + now_str,
exp_name='hide_and_seek') # for saving information during training
ppo(
env_fn=env_fn, # run ppo training loop (check ppo.py for documentation)
pi_lr=3e-4,
vf_lr=3e-4,
steps_per_epoch=10000,
epochs=1000,
train_pi_iters=50,
train_v_iters=50,
logger_kwargs=logger_kwargs)
test_ppo(dir_str + now_str) # test learned policy