def upload_step_metrics(self, metrics):
metrics[StepMetrics.EPISODE.value] = self._number_of_trials_
self._progress_ = metrics[StepMetrics.PROG.value]
self._episode_status = metrics[StepMetrics.EPISODE_STATUS.value]
if self._episode_status in self.reset_count_dict:
self.reset_count_dict[self._episode_status] += 1
StepMetrics.validate_dict(metrics)
sim_trace_log(metrics)
if self.is_save_simtrace_enabled:
write_simtrace_to_local_file(self._simtrace_local_path, metrics)
def upload_step_metrics(self, metrics):
self._progress_ = metrics[StepMetrics.PROG.value]
self._episode_status = metrics[StepMetrics.EPISODE_STATUS.value]
self._episode_reward_ += metrics[StepMetrics.REWARD.value]
if not self._is_eval_:
metrics[StepMetrics.EPISODE.value] = self._episode_
StepMetrics.validate_dict(metrics)
sim_trace_log(metrics)
if self.is_save_simtrace_enabled:
write_simtrace_to_local_file(self._simtrace_local_path,
metrics)
def upload_step_metrics(self, metrics):
self._progress_ = metrics[StepMetrics.PROG.value]
self._episode_status = metrics[StepMetrics.EPISODE_STATUS.value]
# Community fix to have reward for evaluation runs during training
self._episode_reward_ += metrics[StepMetrics.REWARD.value]
if not self._is_eval_:
metrics[StepMetrics.EPISODE.value] = self._episode_
StepMetrics.validate_dict(metrics)
sim_trace_log(metrics)
if self.is_save_simtrace_enabled:
write_simtrace_to_local_file(self._simtrace_local_path,
metrics)
self._update_mp4_video_metrics(metrics)
def upload_step_metrics(self, metrics):
metrics[StepMetrics.EPISODE.value] = self._number_of_trials_
self._progress_ = metrics[StepMetrics.PROG.value]
self._episode_status = metrics[StepMetrics.EPISODE_STATUS.value]
if self._episode_status in self.reset_count_dict:
self.reset_count_dict[self._episode_status] += 1
StepMetrics.validate_dict(metrics)
sim_trace_log(metrics)
if self.is_save_simtrace_enabled:
write_simtrace_to_local_file(
os.path.join(os.path.join(ITERATION_DATA_LOCAL_FILE_PATH, self._agent_name_),
IterationDataLocalFileNames.SIM_TRACE_EVALUATION_LOCAL_FILE.value),
metrics)
self._update_mp4_video_metrics(metrics)
def upload_step_metrics(self, metrics):
self._progress_ = metrics[StepMetrics.PROG.value]
self._episode_status = metrics[StepMetrics.EPISODE_STATUS.value]
# Community fix to have reward for evaluation runs during training
self._episode_reward_ += metrics[StepMetrics.REWARD.value]
if not self._is_eval_:
metrics[StepMetrics.EPISODE.value] = self._episode_
StepMetrics.validate_dict(metrics)
sim_trace_log(metrics)
if self.is_save_simtrace_enabled:
write_simtrace_to_local_file(
os.path.join(os.path.join(ITERATION_DATA_LOCAL_FILE_PATH, self._agent_name_),
IterationDataLocalFileNames.SIM_TRACE_TRAINING_LOCAL_FILE.value),
metrics)
self._update_mp4_video_metrics(metrics)
def upload_step_metrics(self, metrics):
self._progress_ = metrics[StepMetrics.PROG.value]
self._episode_status = metrics[StepMetrics.EPISODE_STATUS.value]
self._episode_reward_ += metrics[StepMetrics.REWARD.value]
#! TODO have this work with new sim trace class
if not self._is_eval_:
metrics[StepMetrics.EPISODE.value] = self._episode_
self._episode_reward_ += metrics[StepMetrics.REWARD.value]
StepMetrics.validate_dict(metrics)
sim_trace_log(metrics)
if self.is_save_simtrace_enabled:
write_simtrace_to_local_file(
os.path.join(
os.path.join(ITERATION_DATA_LOCAL_FILE_PATH,
self._agent_name_),
IterationDataLocalFileNames.
SIM_TRACE_TRAINING_LOCAL_FILE.value), metrics)
def set_reward_and_metrics(reward_params, step_metrics, pos_dict, track_data,
next_index, prev_index, action, json_actions):
'''Populates the reward_params and step_metrics dictionaries with the common
metrics and parameters.
reward_params - Dictionary containing the input parameters to the reward function
step_metrics - Dictionary containing the metrics that are sent to s3
pos_dict - Dictionary containing the agent position data, keys defined in AgentPos
track_data - Object containing all the track information and geometry
next_index - The index of the next way point
prev_index - The index of the previous way point
action - Integer containing the action to take
json_actions - Dictionary that maps action into steering and angle
'''
try:
# Check that the required keys are present in the dicts that are being
# passed in, these methods will throw an exception if a key is missing
RewardParam.validate_dict(reward_params)
StepMetrics.validate_dict(step_metrics)
model_point = pos_dict[AgentPos.POINT.value]
# Geat the nearest points
nearest_pnts_dict = track_data.get_nearest_points(model_point)
# Compute distance from center and road width
nearest_dist_dict = track_data.get_nearest_dist(
nearest_pnts_dict, model_point)
# Compute the distance from the previous and next points
distance_from_prev, distance_from_next = \
track_data.get_distance_from_next_and_prev(model_point, prev_index,
next_index)
# Compute which points are on the track
wheel_on_track = track_data.points_on_track(
pos_dict[AgentPos.LINK_POINTS.value])
# Get the model orientation
model_orientation = pos_dict[AgentPos.ORIENTATION.value]
# Set the reward and metric parameters
reward_params[RewardParam.CENTER_DIST.value[0]] = \
nearest_dist_dict[TrackNearDist.NEAR_DIST_CENT.value]
reward_params[RewardParam.CLS_WAYPNY.value[0]] = [
prev_index, next_index
]
reward_params[RewardParam.LEFT_CENT.value[0]] = \
nearest_dist_dict[TrackNearDist.NEAR_DIST_IN.value] < \
nearest_dist_dict[TrackNearDist.NEAR_DIST_OUT.value]
reward_params[RewardParam.WAYPNTS.value[0]] = track_data.get_way_pnts()
reward_params[RewardParam.TRACK_WIDTH.value[0]] = \
nearest_pnts_dict[TrackNearPnts.NEAR_PNT_IN.value] \
.distance(nearest_pnts_dict[TrackNearPnts.NEAR_PNT_OUT.value])
reward_params[
RewardParam.TRACK_LEN.value[0]] = track_data.get_track_length()
step_metrics[StepMetrics.X.value] = \
reward_params[RewardParam.X.value[0]] = model_point.x
step_metrics[StepMetrics.Y.value] = \
reward_params[RewardParam.Y.value[0]] = model_point.y
step_metrics[StepMetrics.YAW.value] = \
reward_params[RewardParam.HEADING.value[0]] = \
Rotation.from_quat(model_orientation).as_euler('zyx')[0] * 180.0 / math.pi
step_metrics[StepMetrics.CLS_WAYPNT.value] = \
next_index if distance_from_next < distance_from_prev else prev_index
step_metrics[
StepMetrics.TRACK_LEN.value] = track_data.get_track_length()
step_metrics[StepMetrics.STEER.value] = \
reward_params[RewardParam.STEER.value[0]] = \
float(json_actions[action]['steering_angle'])
step_metrics[StepMetrics.THROTTLE.value] = \
reward_params[RewardParam.SPEED.value[0]] = \
float(json_actions[action]['speed'])
step_metrics[StepMetrics.WHEELS_TRACK.value] = \
reward_params[RewardParam.WHEELS_ON_TRACK.value[0]] = all(wheel_on_track)
step_metrics[StepMetrics.ACTION.value] = action
except KeyError as ex:
raise GenericRolloutException("Key {}, not found".format(ex))
except Exception as ex:
raise GenericRolloutException(
'Cannot compute reward and metrics: {}'.format(ex))