def __init__(self, env_name, config, upload_dir=None):
config.update({"alg": "PolicyGradient"})
Algorithm.__init__(self, env_name, config, upload_dir=upload_dir)
# TODO(ekl): preprocessor should be associated with the env elsewhere
if self.env_name == "Pong-v0":
preprocessor = AtariPixelPreprocessor()
elif self.env_name == "Pong-ram-v3":
preprocessor = AtariRamPreprocessor()
elif self.env_name == "CartPole-v0" or self.env_name == "CartPole-v1":
preprocessor = NoPreprocessor()
elif self.env_name == "Hopper-v1":
preprocessor = NoPreprocessor()
elif self.env_name == "Walker2d-v1":
preprocessor = NoPreprocessor()
elif self.env_name == "Humanoid-v1":
preprocessor = NoPreprocessor()
else:
preprocessor = AtariPixelPreprocessor()
self.preprocessor = preprocessor
self.global_step = 0
self.j = 0
self.kl_coeff = config["kl_coeff"]
self.model = Agent(self.env_name, 1, self.preprocessor, self.config,
self.logdir, False)
self.agents = [
RemoteAgent.remote(self.env_name, 1, self.preprocessor,
self.config, self.logdir, True)
for _ in range(config["num_agents"])
]
self.start_time = time.time()
def __init__(self, env_name, config):
Algorithm.__init__(self, env_name, config)
# TODO(ekl) the preprocessor should be associated with the env elsewhere
if self.env_name == "Pong-v0":
preprocessor = AtariPixelPreprocessor()
elif self.env_name == "Pong-ram-v3":
preprocessor = AtariRamPreprocessor()
elif self.env_name == "CartPole-v0":
preprocessor = NoPreprocessor()
elif self.env_name == "Walker2d-v1":
preprocessor = NoPreprocessor()
else:
preprocessor = AtariPixelPreprocessor()
self.preprocessor = preprocessor
self.global_step = 0
self.j = 0
self.kl_coeff = config["kl_coeff"]
self.model = Agent(
self.env_name, 1, self.preprocessor, self.config, False)
self.agents = [
RemoteAgent.remote(
self.env_name, 1, self.preprocessor, self.config, True)
for _ in range(config["num_agents"])]
def __init__(self, env_name, config, upload_dir=None):
config.update({"alg": "PolicyGradient"})
Algorithm.__init__(self, env_name, config, upload_dir=upload_dir)
self.global_step = 0
self.j = 0
self.kl_coeff = config["kl_coeff"]
self.model = Agent(self.env_name, 1, self.config, self.logdir, False)
self.agents = [
RemoteAgent.remote(
self.env_name, 1, self.config, self.logdir, True)
for _ in range(config["num_agents"])]
self.start_time = time.time()
def __init__(self, env_name, config, upload_dir=None):
config.update({"alg": "PolicyGradient"})
Algorithm.__init__(self, env_name, config, upload_dir=upload_dir)
self.global_step = 0
self.j = 0
self.kl_coeff = config["kl_coeff"]
self.model = Agent(self.env_name, 1, self.config, self.logdir, False)
self.agents = [
RemoteAgent.remote(
self.env_name, 1, self.config, self.logdir, True)
for _ in range(config["num_agents"])]
self.start_time = time.time()
# TF does not support to write logs to S3 at the moment
write_tf_logs = config["write_logs"] and self.logdir.startswith("file")
if write_tf_logs:
self.file_writer = tf.summary.FileWriter(
self.logdir, self.model.sess.graph)
else:
self.file_writer = None
self.saver = tf.train.Saver(max_to_keep=None)
# is detected
"use_tf_debugger": False,
# If True, we write checkpoints and tensorflow logging
"write_logs": True,
# Name of the model checkpoint file
"model_checkpoint_file": "iteration-%s.ckpt"
}
config = DEFAULT_CONFIG
config["model"]["fcnet_hiddens"] = [32, 32]
#dirname = "/tmp/ray/HIHIHI_PolicyGradient_2017-08-19_23-11-54x6m9b1il"
dirname = "/tmp/ray/HIHIHI_PolicyGradient_2017-08-20_02-53-02vwr5ax_v"
#dirname = "tmp_checkpoints/"
model = Agent("HIHIHI", 1, NoPreprocessor(), config, dirname, False)
saver = tf.train.Saver(max_to_keep=None)
saver.restore(model.sess, tf.train.latest_checkpoint(dirname))
##################
from pongjsenv import PongJSEnv
import matplotlib as mpl
import time
# mpl.use("MacOSX")
import matplotlib.pyplot as plt
game = PongJSEnv()
terminated = False