def _init(self):
self.global_step = 0
self.local_evaluator = PPOEvaluator(self.env_creator, self.config,
self.logdir, False)
RemotePPOEvaluator = ray.remote(
num_cpus=self.config["num_cpus_per_worker"],
num_gpus=self.config["num_gpus_per_worker"])(PPOEvaluator)
self.remote_evaluators = [
RemotePPOEvaluator.remote(self.env_creator, self.config,
self.logdir, True)
for _ in range(self.config["num_workers"])
]
self.optimizer = LocalMultiGPUOptimizer(
{
"sgd_batch_size": self.config["sgd_batchsize"],
"sgd_stepsize": self.config["sgd_stepsize"],
"num_sgd_iter": self.config["num_sgd_iter"],
"timesteps_per_batch": self.config["timesteps_per_batch"]
},
self.local_evaluator,
self.remote_evaluators,
)
self.saver = tf.train.Saver(max_to_keep=None)
def _init(self):
self.global_step = 0
self.kl_coeff = self.config["kl_coeff"]
self.local_evaluator = PPOEvaluator(self.registry, self.env_creator,
self.config, self.logdir, False)
self.remote_evaluators = [
RemotePPOEvaluator.remote(self.registry, self.env_creator,
self.config, self.logdir, True)
for _ in range(self.config["num_workers"])
]
self.start_time = time.time()
if self.config["write_logs"]:
self.file_writer = tf.summary.FileWriter(
self.logdir, self.local_evaluator.sess.graph)
else:
self.file_writer = None
self.saver = tf.train.Saver(max_to_keep=None)