def __init__(self, env_name, out_file):
"""Initialize.
:param env_name: name of an Atari environment.
:param out_file: output file.
"""
# Super
Callback.__init__(self)
# Video Writer
frame_size = gym.make(env_name).observation_space.shape[:2]
fourcc = cv.VideoWriter_fourcc(*"HFYU")
fps = 20.0
self._writer = cv.VideoWriter(
filename=out_file,
fourcc=fourcc,
fps=fps,
frameSize=(frame_size[1], frame_size[0]),
isColor=True,
)
# Close properly
tools.QuitWithResources.add("VideoWriter",
lambda: self._writer.release())
def __init__(self, logdir, agent):
"""Initialize.
:param logdir: directory of tensorboard logs
:param agent: a QAgentDef instance.
"""
# Super
Callback.__init__(self)
# Store
self._agent = agent
# Both are dicts of {episode: data}
# where data is a dict of metrics accumulated during an episode
# {metric_name: episode_values}
self._episode_scalars = {}
self._episode_hists = {}
# Metrics returned after step and episode to be visualized as
# scalars or histograms
self._kerasrl_step_scalars = ["reward", "metrics"]
self._kerasrl_step_hists = ["action"]
self._kerasrl_episode_scalars = ["episode_reward", "nb_episode_steps"]
self._kerasrl_episode_hists = []
# Tf writer
self.summary_writer = tf.summary.create_file_writer(logdir)
def __init__(self, policy):
"""Initialize."""
# Super
Callback.__init__(self)
# Store
self._min_eps = policy._min_eps
self._max_eps = policy._max_eps
self._policy = policy
def __init__(self, env_name, skip_frames=None, port=None):
"""Initialize.
:param env_name: name of an Atari environment.
:param skip_frames: skip a random number of frames in [0, skip_frames].
:param port: if given, overrides the default port.
"""
# Super
Callback.__init__(self)
# Check
if skip_frames is not None and skip_frames <= 0:
raise ValueError("skip_frames must be positive")
# Init
self.sender = AtariFramesSender(env_name, port=port)
self.skip_frames = skip_frames
self._skips_left = 0
self._last_frame = None
def __init__(self, agent, path, interval):
"""Initialize.
:param agent: a keras-rl agent
:param path: directory of checkpoints
:param interval: save frequency in number of steps
"""
# Super
Callback.__init__(self)
# Store
self.agent = agent
self.interval = interval
self.init_step = 0
self.step = 0
self.episode = 0
self.counters_file = os.path.join(path, os.path.pardir,
"counters.json")
self.step_checkpoints = os.path.join(
path, "weights_{step}." + self.save_format)
def __init__(self):
Callback.__init__(self)
self.action_dict_list = dict()
self.ep_ctr = 0