def init_localoptima_selfplayer(seed, reward_type):
env = make_env(use_cnn=True,
reward_type=reward_type,
self_play=True,
save_score=True)
def make_model(env):
policy_kwargs = dict(
net_arch=[dict(pi=[128, 128, 128], vf=[128, 128, 128])])
return PPO2("CnnPolicy",
env,
verbose=1,
seed=seed,
self_play=True,
policy_kwargs=policy_kwargs)
train_until_localoptimum(make_model, seed, env, reward_type)
from stable_baselines import PPO2
from stable_baselines.common.policies import FeedForwardPolicy, register_policy
from train_func import do_train_multi
from env_getter import make_env, make_env_hex
import tensorflow as tf
from stable_baselines.a2c.utils import conv, linear, conv_to_fc, batch_to_seq, seq_to_batch, lstm
import numpy as np
policy_kwargs = dict(net_arch=[dict(pi=[128, 128, 128], vf=[128, 128, 128])])
env = make_env(use_cnn=True, reward_type=0, self_play=True)
model1 = PPO2("CnnPolicy",
env,
verbose=1,
self_play=True,
tensorboard_log="D:\\4-System\\tensor",
policy_kwargs=policy_kwargs)
model2 = PPO2("CnnPolicy",
env,
verbose=1,
self_play=True,
tensorboard_log="D:\\4-System\\tensor",
policy_kwargs=policy_kwargs)
model3 = PPO2("CnnPolicy",
env,
verbose=1,
self_play=True,
tensorboard_log="D:\\4-System\\tensor",
policy_kwargs=policy_kwargs)
model4 = PPO2("CnnPolicy",
env,
verbose=1,
from stable_baselines import PPO2
from stable_baselines.common.policies import FeedForwardPolicy, register_policy
from train_func import train_until_localoptimum, do_train_multi
from env_getter import make_env, make_env_hex
import tensorflow as tf
from stable_baselines.a2c.utils import conv, linear, conv_to_fc, batch_to_seq, seq_to_batch, lstm
import numpy as np
import sys
env = make_env(use_cnn=False, reward_type=0, self_play=True, save_score=False)
def make_model(env):
policy_kwargs = dict(
net_arch=[dict(pi=[128, 128, 128], vf=[128, 128, 128])])
return PPO2("MlpPolicy",
env,
verbose=1,
self_play=True,
policy_kwargs=policy_kwargs)
models = []
for i in range(0, 5):
models.append(make_model(env))
do_train_multi(models, "multiself", env)
from stable_baselines import A2C
from train_func import do_train
from env_getter import make_env
env = make_env()
model = A2C('MlpPolicy',
env,
verbose=1,
n_steps=71,
epsilon=2.36561309730191e-05,
alpha=0.930340721692521,
learning_rate=0.00223553342110284,
max_grad_norm=0.50523672966627,
ent_coef=0.00105748404479239,
vf_coef=0.0492764482019269,
gamma=0.953839228968649)
do_train(model, "a2cbox", env)