def _define_batch_env(environment_spec, num_agents):
"""Create environments and apply all desired wrappers."""
with tf.variable_scope("environments"):
envs = [environment_spec.env_lambda() for _ in range(num_agents)]
env = gym_env.T2TGymEnv(envs)
env = py_func_batch_env.PyFuncBatchEnv(env)
return env
def define_batch_env(constructor, num_agents, xvfb=False):
"""Create environments and apply all desired wrappers."""
with tf.variable_scope("environments"):
envs = [
ExternalProcessEnv(constructor, xvfb) for _ in range(num_agents)
]
env = batch_env.BatchEnv(envs, blocking=False)
env = py_func_batch_env.PyFuncBatchEnv(env)
return env
def define_batch_env(constructor, num_agents, xvfb=False):
"""Create environments and apply all desired wrappers.
Args:
constructor: Constructor of an OpenAI gym environment.
num_agents: Number of environments to combine in the batch.
xvfb: Frame buffer.
env_processes: Whether to step environment in external processes.
Returns:
In-graph environments object.
"""
with tf.variable_scope("environments"):
envs = [
ExternalProcessEnv(constructor, xvfb) for _ in range(num_agents)
]
env = batch_env.BatchEnv(envs, blocking=False)
env = py_func_batch_env.PyFuncBatchEnv(env)
return env
def batch_env_factory(environment_spec,
num_agents,
initial_frame_chooser=None):
"""Factory of batch envs."""
# TODO(konradczechowski): this is temporary function handling both old and
# new pipelines, refactor this when we move to the new pipeline.
if environment_spec.simulated_env:
cur_batch_env = _define_simulated_batch_env(environment_spec,
num_agents,
initial_frame_chooser)
else:
if "batch_env" in environment_spec:
msg = "Environment_spec should contain only 1 of (env_lambda, batch_env)."
assert "env_lambda" not in environment_spec, msg
batch_env = environment_spec.batch_env
assert batch_env.batch_size == num_agents
else:
batch_env = _define_batch_env(environment_spec, num_agents)
cur_batch_env = py_func_batch_env.PyFuncBatchEnv(batch_env)
return cur_batch_env
