def test_parse_model_metadata_only_action(create_model_metadata_action_space):
"""This function tests the functionality of parse_model_metadata function
in markov/s3/model_metadata parse_model_metadata when we pass a model metadata file with only
action space and no sensor or neural network information.
Args:
create_model_metadata_action_space (String): Gives the path for model metadata file for testing
"""
sensor, network, simapp_version = ModelMetadata.parse_model_metadata(create_model_metadata_action_space)
os.remove(create_model_metadata_action_space)
assert sensor == [Input.OBSERVATION.value]
assert network == NeuralNetwork.DEEP_CONVOLUTIONAL_NETWORK_SHALLOW.value
assert simapp_version == SIMAPP_VERSION_1
def test_parse_model_metadata(create_model_metadata):
"""This function tests the functionality of parse_model_metadata function
in markov/s3/model_metadata parse_model_metadata when we pass a model metadata file with
sensor and neural network information
Args:
create_model_metadata (String): Gives the path for model metadata file for testing
"""
sensor, network, simapp_version = ModelMetadata.parse_model_metadata(create_model_metadata)
os.remove(create_model_metadata)
assert sensor == ["STEREO_CAMERAS"]
assert network == "DEEP_CONVOLUTIONAL_NETWORK_SHALLOW"
assert simapp_version == SIMAPP_VERSION_2
def test_load_model_metadata(s3_bucket, aws_region, model_metadata_s3_key):
"""This function checks the functionality of get_model_metadata_info in
in markov/s3/model_metadata.py
The function checks if model_metadata.json file is downloaded into the required directory.
If the function fails, it will generate an exception which will call log_and_exit internally.
Hence the test will fail.
Args:
s3_bucket (String): S3_BUCKET
aws_region (String): AWS_REGION
model_metadata_s3_key (String): MODEL_METADATA_S3_KEY
"""
model_metadata_local_path = 'test_model_metadata.json'
model_metadata = ModelMetadata(bucket=s3_bucket,
s3_key=model_metadata_s3_key,
region_name=aws_region,
local_path=model_metadata_local_path)
model_metadata.get_model_metadata_info()
assert os.path.isfile(model_metadata_local_path)
# Remove file downloaded
if os.path.isfile(model_metadata_local_path):
os.remove(model_metadata_local_path)
def main():
""" Main function for tournament"""
try:
# parse argument
s3_region = sys.argv[1]
s3_bucket = sys.argv[2]
s3_prefix = sys.argv[3]
s3_yaml_name = sys.argv[4]
# create boto3 session/client and download yaml/json file
session = boto3.session.Session()
s3_endpoint_url = os.environ.get("S3_ENDPOINT_URL", None)
s3_client = S3Client(region_name=s3_region,
s3_endpoint_url=s3_endpoint_url)
# Intermediate tournament files
queue_pickle_name = 'tournament_candidate_queue.pkl'
queue_pickle_s3_key = os.path.normpath(
os.path.join(s3_prefix, queue_pickle_name))
local_queue_pickle_path = os.path.abspath(
os.path.join(os.getcwd(), queue_pickle_name))
report_pickle_name = 'tournament_report.pkl'
report_pickle_s3_key = os.path.normpath(
os.path.join(s3_prefix, report_pickle_name))
local_report_pickle_path = os.path.abspath(
os.path.join(os.getcwd(), report_pickle_name))
final_report_name = 'tournament_report.json'
final_report_s3_key = os.path.normpath(
os.path.join(s3_prefix, final_report_name))
try:
s3_client.download_file(bucket=s3_bucket,
s3_key=queue_pickle_s3_key,
local_path=local_queue_pickle_path)
s3_client.download_file(bucket=s3_bucket,
s3_key=report_pickle_s3_key,
local_path=local_report_pickle_path)
except:
pass
# download yaml file
yaml_file = YamlFile(
agent_type=AgentType.TOURNAMENT.value,
bucket=s3_bucket,
s3_key=get_s3_key(s3_prefix, s3_yaml_name),
region_name=s3_region,
s3_endpoint_url=s3_endpoint_url,
local_path=YAML_LOCAL_PATH_FORMAT.format(s3_yaml_name))
yaml_dict = yaml_file.get_yaml_values()
if os.path.exists(local_queue_pickle_path):
with open(local_queue_pickle_path, 'rb') as f:
tournament_candidate_queue = pickle.load(f)
with open(local_report_pickle_path, 'rb') as f:
tournament_report = pickle.load(f)
logger.info('tournament_candidate_queue loaded from existing file')
else:
logger.info('tournament_candidate_queue initialized')
tournament_candidate_queue = deque()
for agent_idx, _ in enumerate(
yaml_dict[YamlKey.MODEL_S3_BUCKET_YAML_KEY.value]):
tournament_candidate_queue.append((
yaml_dict[YamlKey.MODEL_S3_BUCKET_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.MODEL_S3_PREFIX_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.MODEL_METADATA_FILE_S3_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.METRICS_S3_BUCKET_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.METRICS_S3_PREFIX_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.SIMTRACE_S3_BUCKET_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.SIMTRACE_S3_PREFIX_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.MP4_S3_BUCKET_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.MP4_S3_PREFIX_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.DISPLAY_NAME_YAML_KEY.value][agent_idx],
# TODO: Deprecate the DISPLAY_NAME and use only the RACER_NAME without if else check
"" if None in yaml_dict.get(YamlKey.RACER_NAME_YAML_KEY.value, [None]) \
else yaml_dict[YamlKey.RACER_NAME_YAML_KEY.value][agent_idx],
yaml_dict[YamlKey.BODY_SHELL_TYPE_YAML_KEY.value][agent_idx]
))
tournament_report = {"race_results": []}
race_idx = len(tournament_report["race_results"])
while len(tournament_candidate_queue) > 1:
car1 = tournament_candidate_queue.popleft()
car2 = tournament_candidate_queue.popleft()
(car1_model_s3_bucket, car1_s3_prefix, car1_model_metadata,
car1_metrics_bucket, car1_metrics_s3_key, car1_simtrace_bucket,
car1_simtrace_prefix, car1_mp4_bucket, car1_mp4_prefix,
car1_display_name, car1_racer_name, car1_body_shell_type) = car1
(car2_model_s3_bucket, car2_s3_prefix, car2_model_metadata,
car2_metrics_bucket, car2_metrics_s3_key, car2_simtrace_bucket,
car2_simtrace_prefix, car2_mp4_bucket, car2_mp4_prefix,
car2_display_name, car2_racer_name, car2_body_shell_type) = car2
race_yaml_dict = generate_race_yaml(yaml_dict=yaml_dict,
car1=car1,
car2=car2,
race_idx=race_idx)
if s3_endpoint_url is not None:
race_yaml_dict["S3_ENDPOINT_URL"] = s3_endpoint_url
race_model_s3_buckets = [
car1_model_s3_bucket, car2_model_s3_bucket
]
race_model_metadatas = [car1_model_metadata, car2_model_metadata]
body_shell_types = [car1_body_shell_type, car2_body_shell_type]
# List of directories created
dirs_to_delete = list()
yaml_dir = os.path.abspath(os.path.join(os.getcwd(),
str(race_idx)))
os.makedirs(yaml_dir)
dirs_to_delete.append(yaml_dir)
race_yaml_path = os.path.abspath(
os.path.join(yaml_dir, 'evaluation_params.yaml'))
with open(race_yaml_path, 'w') as race_yaml_file:
yaml.dump(race_yaml_dict, race_yaml_file)
# List of racecar names that should include second camera while launching
racecars_with_stereo_cameras = list()
# List of racecar names that should include lidar while launching
racecars_with_lidars = list()
# List of SimApp versions
simapp_versions = list()
for agent_index, model_s3_bucket in enumerate(
race_model_s3_buckets):
racecar_name = 'racecar_' + str(agent_index)
json_key = race_model_metadatas[agent_index]
# download model metadata
try:
model_metadata = ModelMetadata(
bucket=model_s3_bucket,
s3_key=json_key,
region_name=s3_region,
s3_endpoint_url=s3_endpoint_url,
local_path=MODEL_METADATA_LOCAL_PATH_FORMAT.format(
racecar_name))
dirs_to_delete.append(model_metadata.local_dir)
except Exception as e:
log_and_exit(
"Failed to download model_metadata file: s3_bucket: {}, s3_key: {}, {}"
.format(model_s3_bucket, json_key,
e), SIMAPP_SIMULATION_WORKER_EXCEPTION,
SIMAPP_EVENT_ERROR_CODE_500)
sensors, _, simapp_version = model_metadata.get_model_metadata_info(
)
simapp_versions.append(str(simapp_version))
if Input.STEREO.value in sensors:
racecars_with_stereo_cameras.append(racecar_name)
if Input.LIDAR.value in sensors or Input.SECTOR_LIDAR.value in sensors:
racecars_with_lidars.append(racecar_name)
cmd = [
os.path.join(os.path.dirname(os.path.abspath(__file__)),
"tournament_race_node.py"),
str(race_idx), race_yaml_path,
','.join(racecars_with_stereo_cameras),
','.join(racecars_with_lidars), ','.join(simapp_versions),
','.join(body_shell_types)
]
try:
return_code, _, stderr = run_cmd(cmd_args=cmd,
shell=False,
stdout=None,
stderr=None)
except KeyboardInterrupt:
logger.info(
"KeyboardInterrupt raised, SimApp must be faulted! exiting..."
)
return
# Retrieve winner and append tournament report
with open('race_report.pkl', 'rb') as f:
race_report = pickle.load(f)
race_report['race_idx'] = race_idx
winner = car1 if race_report[
'winner'] == car1_display_name else car2
logger.info("race {}'s winner: {}".format(race_idx,
race_report['winner']))
tournament_candidate_queue.append(winner)
tournament_report["race_results"].append(race_report)
# Clean up directories created
for dir_to_delete in dirs_to_delete:
shutil.rmtree(dir_to_delete, ignore_errors=True)
race_idx += 1
s3_extra_args = get_s3_kms_extra_args()
# Persist latest queue and report to use after job restarts.
with open(local_queue_pickle_path, 'wb') as f:
pickle.dump(tournament_candidate_queue, f, protocol=2)
s3_client.upload_file(bucket=s3_bucket,
s3_key=queue_pickle_s3_key,
local_path=local_queue_pickle_path,
s3_kms_extra_args=s3_extra_args)
with open(local_report_pickle_path, 'wb') as f:
pickle.dump(tournament_report, f, protocol=2)
s3_client.upload_file(bucket=s3_bucket,
s3_key=report_pickle_s3_key,
local_path=local_report_pickle_path,
s3_kms_extra_args=s3_extra_args)
# If there is more than 1 candidates then restart the simulation job otherwise
# tournament is finished, persists final report and ends the job.
if len(tournament_candidate_queue) > 1:
restart_simulation_job(
os.environ.get('AWS_ROBOMAKER_SIMULATION_JOB_ARN'),
s3_region)
break
else:
# Persist final tournament report in json format
# and terminate the job by canceling it
s3_client.put_object(bucket=s3_bucket,
s3_key=final_report_s3_key,
body=json.dumps(tournament_report),
s3_kms_extra_args=s3_extra_args)
cancel_simulation_job(
os.environ.get('AWS_ROBOMAKER_SIMULATION_JOB_ARN'),
s3_region)
except ValueError as ex:
log_and_exit("User modified model_metadata.json: {}".format(ex),
SIMAPP_SIMULATION_WORKER_EXCEPTION,
SIMAPP_EVENT_ERROR_CODE_400)
except Exception as e:
log_and_exit("Tournament node failed: {}".format(e),
SIMAPP_SIMULATION_WORKER_EXCEPTION,
SIMAPP_EVENT_ERROR_CODE_500)
def main():
screen.set_use_colors(False)
parser = argparse.ArgumentParser()
parser.add_argument('-pk',
'--preset_s3_key',
help="(string) Name of a preset to download from S3",
type=str,
required=False)
parser.add_argument(
'-ek',
'--environment_s3_key',
help="(string) Name of an environment file to download from S3",
type=str,
required=False)
parser.add_argument('--model_metadata_s3_key',
help="(string) Model Metadata File S3 Key",
type=str,
required=False)
parser.add_argument(
'-c',
'--checkpoint_dir',
help=
'(string) Path to a folder containing a checkpoint to write the model to.',
type=str,
default='./checkpoint')
parser.add_argument(
'--pretrained_checkpoint_dir',
help='(string) Path to a folder for downloading a pre-trained model',
type=str,
default=PRETRAINED_MODEL_DIR)
parser.add_argument('--s3_bucket',
help='(string) S3 bucket',
type=str,
default=os.environ.get(
"SAGEMAKER_SHARED_S3_BUCKET_PATH", "gsaur-test"))
parser.add_argument('--s3_prefix',
help='(string) S3 prefix',
type=str,
default='sagemaker')
parser.add_argument('--s3_endpoint_url',
help='(string) S3 endpoint URL',
type=str,
default=os.environ.get("S3_ENDPOINT_URL", None))
parser.add_argument('--framework',
help='(string) tensorflow or mxnet',
type=str,
default='tensorflow')
parser.add_argument('--pretrained_s3_bucket',
help='(string) S3 bucket for pre-trained model',
type=str)
parser.add_argument('--pretrained_s3_prefix',
help='(string) S3 prefix for pre-trained model',
type=str,
default='sagemaker')
parser.add_argument(
'--pretrained_checkpoint',
help='(string) Choose which checkpoint to use (best | last)',
type=str,
default="best")
parser.add_argument('--aws_region',
help='(string) AWS region',
type=str,
default=os.environ.get("AWS_REGION", "us-east-1"))
args, _ = parser.parse_known_args()
logger.info("S3 bucket: %s \n S3 prefix: %s \n S3 endpoint URL: %s",
args.s3_bucket, args.s3_prefix, args.s3_endpoint_url)
s3_client = S3Client(region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
max_retry_attempts=0)
# download model metadata
# TODO: replace 'agent' with name of each agent
model_metadata_download = ModelMetadata(
bucket=args.s3_bucket,
s3_key=args.model_metadata_s3_key,
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=MODEL_METADATA_LOCAL_PATH_FORMAT.format('agent'))
_, network_type, version = model_metadata_download.get_model_metadata_info(
)
# upload model metadata
model_metadata_upload = ModelMetadata(
bucket=args.s3_bucket,
s3_key=get_s3_key(args.s3_prefix, MODEL_METADATA_S3_POSTFIX),
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=MODEL_METADATA_LOCAL_PATH_FORMAT.format('agent'))
model_metadata_upload.persist(
s3_kms_extra_args=utils.get_s3_kms_extra_args())
shutil.copy2(model_metadata_download.local_path, SM_MODEL_OUTPUT_DIR)
success_custom_preset = False
if args.preset_s3_key:
preset_local_path = "./markov/presets/preset.py"
try:
s3_client.download_file(bucket=args.s3_bucket,
s3_key=args.preset_s3_key,
local_path=preset_local_path)
success_custom_preset = True
except botocore.exceptions.ClientError:
pass
if not success_custom_preset:
logger.info(
"Could not download the preset file. Using the default DeepRacer preset."
)
else:
preset_location = "markov.presets.preset:graph_manager"
graph_manager = short_dynamic_import(preset_location,
ignore_module_case=True)
s3_client.upload_file(
bucket=args.s3_bucket,
s3_key=os.path.normpath("%s/presets/preset.py" %
args.s3_prefix),
local_path=preset_local_path,
s3_kms_extra_args=utils.get_s3_kms_extra_args())
if success_custom_preset:
logger.info("Using preset: %s" % args.preset_s3_key)
if not success_custom_preset:
params_blob = os.environ.get('SM_TRAINING_ENV', '')
if params_blob:
params = json.loads(params_blob)
sm_hyperparams_dict = params["hyperparameters"]
else:
sm_hyperparams_dict = {}
#! TODO each agent should have own config
agent_config = {
'model_metadata': model_metadata_download,
ConfigParams.CAR_CTRL_CONFIG.value: {
ConfigParams.LINK_NAME_LIST.value: [],
ConfigParams.VELOCITY_LIST.value: {},
ConfigParams.STEERING_LIST.value: {},
ConfigParams.CHANGE_START.value: None,
ConfigParams.ALT_DIR.value: None,
ConfigParams.ACTION_SPACE_PATH.value:
model_metadata_download.local_path,
ConfigParams.REWARD.value: None,
ConfigParams.AGENT_NAME.value: 'racecar'
}
}
agent_list = list()
agent_list.append(create_training_agent(agent_config))
graph_manager, robomaker_hyperparams_json = get_graph_manager(
hp_dict=sm_hyperparams_dict,
agent_list=agent_list,
run_phase_subject=None)
# Upload hyperparameters to SageMaker shared s3 bucket
hyperparameters = Hyperparameters(bucket=args.s3_bucket,
s3_key=get_s3_key(
args.s3_prefix,
HYPERPARAMETER_S3_POSTFIX),
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url)
hyperparameters.persist(
hyperparams_json=robomaker_hyperparams_json,
s3_kms_extra_args=utils.get_s3_kms_extra_args())
# Attach sample collector to graph_manager only if sample count > 0
max_sample_count = int(sm_hyperparams_dict.get("max_sample_count", 0))
if max_sample_count > 0:
sample_collector = SampleCollector(
bucket=args.s3_bucket,
s3_prefix=args.s3_prefix,
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
max_sample_count=max_sample_count,
sampling_frequency=int(
sm_hyperparams_dict.get("sampling_frequency", 1)))
graph_manager.sample_collector = sample_collector
# persist IP config from sagemaker to s3
ip_config = IpConfig(bucket=args.s3_bucket,
s3_prefix=args.s3_prefix,
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url)
ip_config.persist(s3_kms_extra_args=utils.get_s3_kms_extra_args())
use_pretrained_model = args.pretrained_s3_bucket and args.pretrained_s3_prefix
# Handle backward compatibility
if use_pretrained_model:
# checkpoint s3 instance for pretrained model
# TODO: replace 'agent' for multiagent training
checkpoint = Checkpoint(bucket=args.pretrained_s3_bucket,
s3_prefix=args.pretrained_s3_prefix,
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
agent_name='agent',
checkpoint_dir=args.pretrained_checkpoint_dir)
# make coach checkpoint compatible
if version < SIMAPP_VERSION_2 and not checkpoint.rl_coach_checkpoint.is_compatible(
):
checkpoint.rl_coach_checkpoint.make_compatible(
checkpoint.syncfile_ready)
# Get the correct pre-trained checkpoint
if args.pretrained_checkpoint.lower() == "best":
# get best model checkpoint string
model_checkpoint_name = checkpoint.deepracer_checkpoint_json.get_deepracer_best_checkpoint(
)
else:
# get the last model checkpoint string
model_checkpoint_name = checkpoint.deepracer_checkpoint_json.get_deepracer_last_checkpoint(
)
# Select the best checkpoint model by uploading rl coach .coach_checkpoint file
checkpoint.rl_coach_checkpoint.update(
model_checkpoint_name=model_checkpoint_name,
s3_kms_extra_args=utils.get_s3_kms_extra_args())
# add checkpoint into checkpoint_dict
checkpoint_dict = {'agent': checkpoint}
# load pretrained model
ds_params_instance_pretrained = S3BotoDataStoreParameters(
checkpoint_dict=checkpoint_dict)
data_store_pretrained = S3BotoDataStore(ds_params_instance_pretrained,
graph_manager, True)
data_store_pretrained.load_from_store()
memory_backend_params = DeepRacerRedisPubSubMemoryBackendParameters(
redis_address="localhost",
redis_port=6379,
run_type=str(RunType.TRAINER),
channel=args.s3_prefix,
network_type=network_type)
graph_manager.memory_backend_params = memory_backend_params
# checkpoint s3 instance for training model
checkpoint = Checkpoint(bucket=args.s3_bucket,
s3_prefix=args.s3_prefix,
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
agent_name='agent',
checkpoint_dir=args.checkpoint_dir)
checkpoint_dict = {'agent': checkpoint}
ds_params_instance = S3BotoDataStoreParameters(
checkpoint_dict=checkpoint_dict)
graph_manager.data_store_params = ds_params_instance
graph_manager.data_store = S3BotoDataStore(ds_params_instance,
graph_manager)
task_parameters = TaskParameters()
task_parameters.experiment_path = SM_MODEL_OUTPUT_DIR
task_parameters.checkpoint_save_secs = 20
if use_pretrained_model:
task_parameters.checkpoint_restore_path = args.pretrained_checkpoint_dir
task_parameters.checkpoint_save_dir = args.checkpoint_dir
training_worker(
graph_manager=graph_manager,
task_parameters=task_parameters,
user_batch_size=json.loads(robomaker_hyperparams_json)["batch_size"],
user_episode_per_rollout=json.loads(
robomaker_hyperparams_json)["num_episodes_between_training"])
def main():
""" Main function for downloading yaml params """
# parse argument
s3_region = sys.argv[1]
s3_bucket = sys.argv[2]
s3_prefix = sys.argv[3]
s3_yaml_name = sys.argv[4]
launch_name = sys.argv[5]
yaml_key = os.path.normpath(os.path.join(s3_prefix, s3_yaml_name))
try:
s3_endpoint_url = os.environ.get("S3_ENDPOINT_URL", None)
if s3_endpoint_url is not None:
logging.info('Endpoint URL {}'.format(s3_endpoint_url))
rospy.set_param('S3_ENDPOINT_URL', s3_endpoint_url)
if AgentType.ROLLOUT.value in launch_name:
# For rollout, launch_name is "rollout_rl_agent.launch"
agent_type = AgentType.ROLLOUT.value
elif AgentType.EVALUATION.value in launch_name:
# For eval, launch_name is "evaluation_rl_agent.launch"
agent_type = AgentType.EVALUATION.value
else:
log_and_exit(
"Unknown agent type in launch file: {}".format(launch_name),
SIMAPP_SIMULATION_WORKER_EXCEPTION,
SIMAPP_EVENT_ERROR_CODE_500)
# download yaml file
yaml_key = get_s3_key(s3_prefix, s3_yaml_name)
yaml_file = YamlFile(
agent_type=agent_type,
bucket=s3_bucket,
s3_key=yaml_key,
region_name=s3_region,
s3_endpoint_url=s3_endpoint_url,
local_path=YAML_LOCAL_PATH_FORMAT.format(s3_yaml_name))
yaml_file.get_yaml_values()
# List of racecar names that should include second camera while launching
racecars_with_stereo_cameras = list()
# List of racecar names that should include lidar while launching
racecars_with_lidars = list()
# List of SimApp versions
simapp_versions = list()
for agent_index, model_s3_bucket in enumerate(
yaml_file.model_s3_buckets):
racecar_name = 'racecar_' + str(agent_index) \
if yaml_file.is_multicar else 'racecar'
json_key = yaml_file.model_metadata_s3_keys[agent_index]
# download model metadata
model_metadata = ModelMetadata(
bucket=model_s3_bucket,
s3_key=json_key,
region_name=s3_region,
s3_endpoint_url=s3_endpoint_url,
local_path=MODEL_METADATA_LOCAL_PATH_FORMAT.format(
racecar_name))
sensors, _, simapp_version = model_metadata.get_model_metadata_info(
)
simapp_versions.append(str(simapp_version))
if Input.STEREO.value in sensors:
racecars_with_stereo_cameras.append(racecar_name)
if Input.LIDAR.value in sensors or Input.SECTOR_LIDAR.value in sensors:
racecars_with_lidars.append(racecar_name)
cmd = [
''.join(("roslaunch deepracer_simulation_environment {} ".format(
launch_name),
"local_yaml_path:={} ".format(yaml_file.local_path),
"racecars_with_stereo_cameras:={} ".format(
','.join(racecars_with_stereo_cameras)),
"racecars_with_lidars:={} ".format(
','.join(racecars_with_lidars)),
"multicar:={} ".format(yaml_file.is_multicar),
"body_shell_types:={} ".format(','.join(
yaml_file.body_shell_types)),
"simapp_versions:={} ".format(','.join(simapp_versions)),
"f1:={}".format(yaml_file.is_f1)))
]
Popen(cmd, shell=True, executable="/bin/bash")
except botocore.exceptions.ClientError as ex:
log_and_exit(
"Download params and launch of agent node S3 ClientError: s3_bucket: {}, yaml_key: {}, {}"
.format(s3_bucket, yaml_key, ex),
SIMAPP_SIMULATION_WORKER_EXCEPTION, SIMAPP_EVENT_ERROR_CODE_500)
except botocore.exceptions.EndpointConnectionError:
log_and_exit("No Internet connection or s3 service unavailable",
SIMAPP_SIMULATION_WORKER_EXCEPTION,
SIMAPP_EVENT_ERROR_CODE_500)
except ValueError as ex:
log_and_exit("User modified model_metadata.json: {}".format(ex),
SIMAPP_SIMULATION_WORKER_EXCEPTION,
SIMAPP_EVENT_ERROR_CODE_500)
except Exception as ex:
log_and_exit(
"Download params and launch of agent node failed: s3_bucket: {}, yaml_key: {}, {}"
.format(s3_bucket, yaml_key, ex),
SIMAPP_SIMULATION_WORKER_EXCEPTION, SIMAPP_EVENT_ERROR_CODE_500)
def validate(s3_bucket, s3_prefix, aws_region):
screen.set_use_colors(False)
screen.log_title(" S3 bucket: {} \n S3 prefix: {}".format(
s3_bucket, s3_prefix))
# download model metadata
model_metadata = ModelMetadata(bucket=s3_bucket,
s3_key=get_s3_key(
s3_prefix, MODEL_METADATA_S3_POSTFIX),
region_name=aws_region,
local_path=MODEL_METADATA_LOCAL_PATH)
# Create model local path
os.makedirs(LOCAL_MODEL_DIR)
try:
# Handle backward compatibility
observation_list, _, version = model_metadata.get_model_metadata_info()
except Exception as ex:
log_and_exit("Failed to parse model_metadata file: {}".format(ex),
SIMAPP_VALIDATION_WORKER_EXCEPTION,
SIMAPP_EVENT_ERROR_CODE_400)
# Below get_transition_data function must called before create_training_agent function
# to avoid 500 in case unsupported Sensor is received.
# create_training_agent will exit with 500 if unsupported sensor is received,
# and get_transition_data function below will exit with 400 if unsupported sensor is received.
# We want to return 400 in model validation case if unsupported sensor is received.
# Thus, call this get_transition_data function before create_traning_agent function!
transitions = get_transition_data(observation_list)
if version < SIMAPP_VERSION_2 and \
not utils.has_current_ckpnt_name(s3_bucket, s3_prefix, aws_region):
utils.make_compatible(s3_bucket, s3_prefix, aws_region,
SyncFiles.TRAINER_READY.value)
agent_config = {
'model_metadata': model_metadata,
ConfigParams.CAR_CTRL_CONFIG.value: {
ConfigParams.LINK_NAME_LIST.value: [],
ConfigParams.VELOCITY_LIST.value: {},
ConfigParams.STEERING_LIST.value: {},
ConfigParams.CHANGE_START.value: None,
ConfigParams.ALT_DIR.value: None,
ConfigParams.ACTION_SPACE_PATH.value: model_metadata.local_path,
ConfigParams.REWARD.value: None,
ConfigParams.AGENT_NAME.value: 'racecar'
}
}
agent_list = list()
agent_list.append(create_training_agent(agent_config))
sm_hyperparams_dict = {}
graph_manager, _ = get_graph_manager(hp_dict=sm_hyperparams_dict,
agent_list=agent_list,
run_phase_subject=None)
ds_params_instance = S3BotoDataStoreParameters(
aws_region=aws_region,
bucket_names={'agent': s3_bucket},
s3_folders={'agent': s3_prefix},
base_checkpoint_dir=LOCAL_MODEL_DIR)
graph_manager.data_store = S3BotoDataStore(ds_params_instance,
graph_manager,
ignore_lock=True)
task_parameters = TaskParameters()
task_parameters.checkpoint_restore_path = LOCAL_MODEL_DIR
_validate(graph_manager=graph_manager,
task_parameters=task_parameters,
transitions=transitions,
s3_bucket=s3_bucket,
s3_prefix=s3_prefix,
aws_region=aws_region)
def test_parse_model_metadata_exception():
"""This function tests the functionality of parse_model_metadata function
in markov/s3/model_metadata parse_model_metadata when an exception occurs
"""
with pytest.raises(Exception, match=r".*Model metadata does not exist:.*"):
sensor, network, simapp_version = ModelMetadata.parse_model_metadata("dummy_file.json")
def main():
""" Main function for evaluation worker """
parser = argparse.ArgumentParser()
parser.add_argument('-p',
'--preset',
help="(string) Name of a preset to run \
(class name from the 'presets' directory.)",
type=str,
required=False)
parser.add_argument('--s3_bucket',
help='list(string) S3 bucket',
type=str,
nargs='+',
default=rospy.get_param("MODEL_S3_BUCKET",
["gsaur-test"]))
parser.add_argument('--s3_prefix',
help='list(string) S3 prefix',
type=str,
nargs='+',
default=rospy.get_param("MODEL_S3_PREFIX",
["sagemaker"]))
parser.add_argument('--aws_region',
help='(string) AWS region',
type=str,
default=rospy.get_param("AWS_REGION", "us-east-1"))
parser.add_argument('--number_of_trials',
help='(integer) Number of trials',
type=int,
default=int(rospy.get_param("NUMBER_OF_TRIALS", 10)))
parser.add_argument(
'-c',
'--local_model_directory',
help='(string) Path to a folder containing a checkpoint \
to restore the model from.',
type=str,
default='./checkpoint')
parser.add_argument('--number_of_resets',
help='(integer) Number of resets',
type=int,
default=int(rospy.get_param("NUMBER_OF_RESETS", 0)))
parser.add_argument('--penalty_seconds',
help='(float) penalty second',
type=float,
default=float(rospy.get_param("PENALTY_SECONDS", 2.0)))
parser.add_argument('--job_type',
help='(string) job type',
type=str,
default=rospy.get_param("JOB_TYPE", "EVALUATION"))
parser.add_argument('--is_continuous',
help='(boolean) is continous after lap completion',
type=bool,
default=utils.str2bool(
rospy.get_param("IS_CONTINUOUS", False)))
parser.add_argument('--race_type',
help='(string) Race type',
type=str,
default=rospy.get_param("RACE_TYPE", "TIME_TRIAL"))
parser.add_argument('--off_track_penalty',
help='(float) off track penalty second',
type=float,
default=float(rospy.get_param("OFF_TRACK_PENALTY",
2.0)))
parser.add_argument('--collision_penalty',
help='(float) collision penalty second',
type=float,
default=float(rospy.get_param("COLLISION_PENALTY",
5.0)))
args = parser.parse_args()
arg_s3_bucket = args.s3_bucket
arg_s3_prefix = args.s3_prefix
logger.info("S3 bucket: %s \n S3 prefix: %s", arg_s3_bucket, arg_s3_prefix)
metrics_s3_buckets = rospy.get_param('METRICS_S3_BUCKET')
metrics_s3_object_keys = rospy.get_param('METRICS_S3_OBJECT_KEY')
arg_s3_bucket, arg_s3_prefix = utils.force_list(
arg_s3_bucket), utils.force_list(arg_s3_prefix)
metrics_s3_buckets = utils.force_list(metrics_s3_buckets)
metrics_s3_object_keys = utils.force_list(metrics_s3_object_keys)
validate_list = [
arg_s3_bucket, arg_s3_prefix, metrics_s3_buckets,
metrics_s3_object_keys
]
simtrace_s3_bucket = rospy.get_param('SIMTRACE_S3_BUCKET', None)
mp4_s3_bucket = rospy.get_param('MP4_S3_BUCKET', None)
if simtrace_s3_bucket:
simtrace_s3_object_prefix = rospy.get_param('SIMTRACE_S3_PREFIX')
simtrace_s3_bucket = utils.force_list(simtrace_s3_bucket)
simtrace_s3_object_prefix = utils.force_list(simtrace_s3_object_prefix)
validate_list.extend([simtrace_s3_bucket, simtrace_s3_object_prefix])
if mp4_s3_bucket:
mp4_s3_object_prefix = rospy.get_param('MP4_S3_OBJECT_PREFIX')
mp4_s3_bucket = utils.force_list(mp4_s3_bucket)
mp4_s3_object_prefix = utils.force_list(mp4_s3_object_prefix)
validate_list.extend([mp4_s3_bucket, mp4_s3_object_prefix])
if not all([lambda x: len(x) == len(validate_list[0]), validate_list]):
log_and_exit(
"Eval worker error: Incorrect arguments passed: {}".format(
validate_list), SIMAPP_SIMULATION_WORKER_EXCEPTION,
SIMAPP_EVENT_ERROR_CODE_500)
if args.number_of_resets != 0 and args.number_of_resets < MIN_RESET_COUNT:
raise GenericRolloutException(
"number of resets is less than {}".format(MIN_RESET_COUNT))
# Instantiate Cameras
if len(arg_s3_bucket) == 1:
configure_camera(namespaces=['racecar'])
else:
configure_camera(namespaces=[
'racecar_{}'.format(str(agent_index))
for agent_index in range(len(arg_s3_bucket))
])
agent_list = list()
s3_bucket_dict = dict()
s3_prefix_dict = dict()
checkpoint_dict = dict()
simtrace_video_s3_writers = []
start_positions = get_start_positions(len(arg_s3_bucket))
done_condition = utils.str_to_done_condition(
rospy.get_param("DONE_CONDITION", any))
park_positions = utils.pos_2d_str_to_list(
rospy.get_param("PARK_POSITIONS", []))
# if not pass in park positions for all done condition case, use default
if not park_positions:
park_positions = [DEFAULT_PARK_POSITION for _ in arg_s3_bucket]
for agent_index, _ in enumerate(arg_s3_bucket):
agent_name = 'agent' if len(arg_s3_bucket) == 1 else 'agent_{}'.format(
str(agent_index))
racecar_name = 'racecar' if len(
arg_s3_bucket) == 1 else 'racecar_{}'.format(str(agent_index))
s3_bucket_dict[agent_name] = arg_s3_bucket[agent_index]
s3_prefix_dict[agent_name] = arg_s3_prefix[agent_index]
# download model metadata
model_metadata = ModelMetadata(
bucket=arg_s3_bucket[agent_index],
s3_key=get_s3_key(arg_s3_prefix[agent_index],
MODEL_METADATA_S3_POSTFIX),
region_name=args.aws_region,
local_path=MODEL_METADATA_LOCAL_PATH_FORMAT.format(agent_name))
_, _, version = model_metadata.get_model_metadata_info()
# checkpoint s3 instance
checkpoint = Checkpoint(bucket=arg_s3_bucket[agent_index],
s3_prefix=arg_s3_prefix[agent_index],
region_name=args.aws_region,
agent_name=agent_name,
checkpoint_dir=args.local_model_directory)
# make coach checkpoint compatible
if version < SIMAPP_VERSION_2 and not checkpoint.rl_coach_checkpoint.is_compatible(
):
checkpoint.rl_coach_checkpoint.make_compatible(
checkpoint.syncfile_ready)
# get best model checkpoint string
model_checkpoint_name = checkpoint.deepracer_checkpoint_json.get_deepracer_best_checkpoint(
)
# Select the best checkpoint model by uploading rl coach .coach_checkpoint file
checkpoint.rl_coach_checkpoint.update(
model_checkpoint_name=model_checkpoint_name,
s3_kms_extra_args=utils.get_s3_kms_extra_args())
checkpoint_dict[agent_name] = checkpoint
agent_config = {
'model_metadata': model_metadata,
ConfigParams.CAR_CTRL_CONFIG.value: {
ConfigParams.LINK_NAME_LIST.value: [
link_name.replace('racecar', racecar_name)
for link_name in LINK_NAMES
],
ConfigParams.VELOCITY_LIST.value: [
velocity_topic.replace('racecar', racecar_name)
for velocity_topic in VELOCITY_TOPICS
],
ConfigParams.STEERING_LIST.value: [
steering_topic.replace('racecar', racecar_name)
for steering_topic in STEERING_TOPICS
],
ConfigParams.CHANGE_START.value:
utils.str2bool(rospy.get_param('CHANGE_START_POSITION',
False)),
ConfigParams.ALT_DIR.value:
utils.str2bool(
rospy.get_param('ALTERNATE_DRIVING_DIRECTION', False)),
ConfigParams.ACTION_SPACE_PATH.value:
model_metadata.local_path,
ConfigParams.REWARD.value:
reward_function,
ConfigParams.AGENT_NAME.value:
racecar_name,
ConfigParams.VERSION.value:
version,
ConfigParams.NUMBER_OF_RESETS.value:
args.number_of_resets,
ConfigParams.PENALTY_SECONDS.value:
args.penalty_seconds,
ConfigParams.NUMBER_OF_TRIALS.value:
args.number_of_trials,
ConfigParams.IS_CONTINUOUS.value:
args.is_continuous,
ConfigParams.RACE_TYPE.value:
args.race_type,
ConfigParams.COLLISION_PENALTY.value:
args.collision_penalty,
ConfigParams.OFF_TRACK_PENALTY.value:
args.off_track_penalty,
ConfigParams.START_POSITION.value:
start_positions[agent_index],
ConfigParams.DONE_CONDITION.value:
done_condition
}
}
metrics_s3_config = {
MetricsS3Keys.METRICS_BUCKET.value:
metrics_s3_buckets[agent_index],
MetricsS3Keys.METRICS_KEY.value:
metrics_s3_object_keys[agent_index],
# Replaced rospy.get_param('AWS_REGION') to be equal to the argument being passed
# or default argument set
MetricsS3Keys.REGION.value:
args.aws_region
}
aws_region = rospy.get_param('AWS_REGION', args.aws_region)
if simtrace_s3_bucket:
simtrace_video_s3_writers.append(
SimtraceVideo(
upload_type=SimtraceVideoNames.SIMTRACE_EVAL.value,
bucket=simtrace_s3_bucket[agent_index],
s3_prefix=simtrace_s3_object_prefix[agent_index],
region_name=aws_region,
local_path=SIMTRACE_EVAL_LOCAL_PATH_FORMAT.format(
agent_name)))
if mp4_s3_bucket:
simtrace_video_s3_writers.extend([
SimtraceVideo(
upload_type=SimtraceVideoNames.PIP.value,
bucket=mp4_s3_bucket[agent_index],
s3_prefix=mp4_s3_object_prefix[agent_index],
region_name=aws_region,
local_path=CAMERA_PIP_MP4_LOCAL_PATH_FORMAT.format(
agent_name)),
SimtraceVideo(
upload_type=SimtraceVideoNames.DEGREE45.value,
bucket=mp4_s3_bucket[agent_index],
s3_prefix=mp4_s3_object_prefix[agent_index],
region_name=aws_region,
local_path=CAMERA_45DEGREE_LOCAL_PATH_FORMAT.format(
agent_name)),
SimtraceVideo(
upload_type=SimtraceVideoNames.TOPVIEW.value,
bucket=mp4_s3_bucket[agent_index],
s3_prefix=mp4_s3_object_prefix[agent_index],
region_name=aws_region,
local_path=CAMERA_TOPVIEW_LOCAL_PATH_FORMAT.format(
agent_name))
])
run_phase_subject = RunPhaseSubject()
agent_list.append(
create_rollout_agent(
agent_config,
EvalMetrics(agent_name, metrics_s3_config, args.is_continuous),
run_phase_subject))
agent_list.append(create_obstacles_agent())
agent_list.append(create_bot_cars_agent())
# ROS service to indicate all the robomaker markov packages are ready for consumption
signal_robomaker_markov_package_ready()
PhaseObserver('/agent/training_phase', run_phase_subject)
enable_domain_randomization = utils.str2bool(
rospy.get_param('ENABLE_DOMAIN_RANDOMIZATION', False))
sm_hyperparams_dict = {}
graph_manager, _ = get_graph_manager(
hp_dict=sm_hyperparams_dict,
agent_list=agent_list,
run_phase_subject=run_phase_subject,
enable_domain_randomization=enable_domain_randomization,
done_condition=done_condition)
ds_params_instance = S3BotoDataStoreParameters(
checkpoint_dict=checkpoint_dict)
graph_manager.data_store = S3BotoDataStore(params=ds_params_instance,
graph_manager=graph_manager,
ignore_lock=True)
graph_manager.env_params.seed = 0
task_parameters = TaskParameters()
task_parameters.checkpoint_restore_path = args.local_model_directory
evaluation_worker(graph_manager=graph_manager,
number_of_trials=args.number_of_trials,
task_parameters=task_parameters,
simtrace_video_s3_writers=simtrace_video_s3_writers,
is_continuous=args.is_continuous,
park_positions=park_positions)
def main():
screen.set_use_colors(False)
parser = argparse.ArgumentParser()
parser.add_argument(
'-c',
'--checkpoint_dir',
help=
'(string) Path to a folder containing a checkpoint to restore the model from.',
type=str,
default='./checkpoint')
parser.add_argument('--s3_bucket',
help='(string) S3 bucket',
type=str,
default=rospy.get_param("SAGEMAKER_SHARED_S3_BUCKET",
"gsaur-test"))
parser.add_argument('--s3_prefix',
help='(string) S3 prefix',
type=str,
default=rospy.get_param("SAGEMAKER_SHARED_S3_PREFIX",
"sagemaker"))
parser.add_argument('--s3_endpoint_url',
help='(string) S3 endpoint URL',
type=str,
default=rospy.get_param("S3_ENDPOINT_URL", None))
parser.add_argument(
'--num_workers',
help="(int) The number of workers started in this pool",
type=int,
default=int(rospy.get_param("NUM_WORKERS", 1)))
parser.add_argument('--rollout_idx',
help="(int) The index of current rollout worker",
type=int,
default=0)
parser.add_argument('-r',
'--redis_ip',
help="(string) IP or host for the redis server",
default='localhost',
type=str)
parser.add_argument('-rp',
'--redis_port',
help="(int) Port of the redis server",
default=6379,
type=int)
parser.add_argument('--aws_region',
help='(string) AWS region',
type=str,
default=rospy.get_param("AWS_REGION", "us-east-1"))
parser.add_argument('--reward_file_s3_key',
help='(string) Reward File S3 Key',
type=str,
default=rospy.get_param("REWARD_FILE_S3_KEY", None))
parser.add_argument('--model_metadata_s3_key',
help='(string) Model Metadata File S3 Key',
type=str,
default=rospy.get_param("MODEL_METADATA_FILE_S3_KEY",
None))
# For training job, reset is not allowed. penalty_seconds, off_track_penalty, and
# collision_penalty will all be 0 be default
parser.add_argument('--number_of_resets',
help='(integer) Number of resets',
type=int,
default=int(rospy.get_param("NUMBER_OF_RESETS", 0)))
parser.add_argument('--penalty_seconds',
help='(float) penalty second',
type=float,
default=float(rospy.get_param("PENALTY_SECONDS", 0.0)))
parser.add_argument('--job_type',
help='(string) job type',
type=str,
default=rospy.get_param("JOB_TYPE", "TRAINING"))
parser.add_argument('--is_continuous',
help='(boolean) is continous after lap completion',
type=bool,
default=utils.str2bool(
rospy.get_param("IS_CONTINUOUS", False)))
parser.add_argument('--race_type',
help='(string) Race type',
type=str,
default=rospy.get_param("RACE_TYPE", "TIME_TRIAL"))
parser.add_argument('--off_track_penalty',
help='(float) off track penalty second',
type=float,
default=float(rospy.get_param("OFF_TRACK_PENALTY",
0.0)))
parser.add_argument('--collision_penalty',
help='(float) collision penalty second',
type=float,
default=float(rospy.get_param("COLLISION_PENALTY",
0.0)))
parser.add_argument('--round_robin_advance_dist',
help='(float) round robin distance 0-1',
type=float,
default=float(
rospy.get_param("ROUND_ROBIN_ADVANCE_DIST", 0.05)))
parser.add_argument('--start_position_offset',
help='(float) offset start 0-1',
type=float,
default=float(
rospy.get_param("START_POSITION_OFFSET", 0.0)))
args = parser.parse_args()
logger.info("S3 bucket: %s", args.s3_bucket)
logger.info("S3 prefix: %s", args.s3_prefix)
logger.info("S3 endpoint URL: %s" % args.s3_endpoint_url)
# Download and import reward function
# TODO: replace 'agent' with name of each agent for multi-agent training
reward_function_file = RewardFunction(
bucket=args.s3_bucket,
s3_key=args.reward_file_s3_key,
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=REWARD_FUCTION_LOCAL_PATH_FORMAT.format('agent'))
reward_function = reward_function_file.get_reward_function()
# Instantiate Cameras
configure_camera(namespaces=['racecar'])
preset_file_success, _ = download_custom_files_if_present(
s3_bucket=args.s3_bucket,
s3_prefix=args.s3_prefix,
aws_region=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url)
# download model metadata
# TODO: replace 'agent' with name of each agent
model_metadata = ModelMetadata(
bucket=args.s3_bucket,
s3_key=args.model_metadata_s3_key,
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=MODEL_METADATA_LOCAL_PATH_FORMAT.format('agent'))
_, _, version = model_metadata.get_model_metadata_info()
agent_config = {
'model_metadata': model_metadata,
ConfigParams.CAR_CTRL_CONFIG.value: {
ConfigParams.LINK_NAME_LIST.value:
LINK_NAMES,
ConfigParams.VELOCITY_LIST.value:
VELOCITY_TOPICS,
ConfigParams.STEERING_LIST.value:
STEERING_TOPICS,
ConfigParams.CHANGE_START.value:
utils.str2bool(rospy.get_param('CHANGE_START_POSITION', True)),
ConfigParams.ALT_DIR.value:
utils.str2bool(
rospy.get_param('ALTERNATE_DRIVING_DIRECTION', False)),
ConfigParams.ACTION_SPACE_PATH.value:
model_metadata.local_path,
ConfigParams.REWARD.value:
reward_function,
ConfigParams.AGENT_NAME.value:
'racecar',
ConfigParams.VERSION.value:
version,
ConfigParams.NUMBER_OF_RESETS.value:
args.number_of_resets,
ConfigParams.PENALTY_SECONDS.value:
args.penalty_seconds,
ConfigParams.NUMBER_OF_TRIALS.value:
None,
ConfigParams.IS_CONTINUOUS.value:
args.is_continuous,
ConfigParams.RACE_TYPE.value:
args.race_type,
ConfigParams.COLLISION_PENALTY.value:
args.collision_penalty,
ConfigParams.OFF_TRACK_PENALTY.value:
args.off_track_penalty,
ConfigParams.ROUND_ROBIN_ADVANCE_DIST.value:
args.round_robin_advance_dist,
ConfigParams.START_POSITION_OFFSET.value:
args.start_position_offset
}
}
#! TODO each agent should have own s3 bucket
metrics_key = rospy.get_param('METRICS_S3_OBJECT_KEY')
if args.num_workers > 1 and args.rollout_idx > 0:
key_tuple = os.path.splitext(metrics_key)
metrics_key = "{}_{}{}".format(key_tuple[0], str(args.rollout_idx),
key_tuple[1])
metrics_s3_config = {
MetricsS3Keys.METRICS_BUCKET.value:
rospy.get_param('METRICS_S3_BUCKET'),
MetricsS3Keys.METRICS_KEY.value: metrics_key,
MetricsS3Keys.ENDPOINT_URL.value:
rospy.get_param('S3_ENDPOINT_URL', None),
MetricsS3Keys.REGION.value: rospy.get_param('AWS_REGION')
}
run_phase_subject = RunPhaseSubject()
agent_list = list()
#TODO: replace agent for multi agent training
# checkpoint s3 instance
# TODO replace agent with agent_0 and so on for multiagent case
checkpoint = Checkpoint(bucket=args.s3_bucket,
s3_prefix=args.s3_prefix,
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
agent_name='agent',
checkpoint_dir=args.checkpoint_dir)
agent_list.append(
create_rollout_agent(
agent_config,
TrainingMetrics(
agent_name='agent',
s3_dict_metrics=metrics_s3_config,
deepracer_checkpoint_json=checkpoint.deepracer_checkpoint_json,
ckpnt_dir=os.path.join(args.checkpoint_dir, 'agent'),
run_phase_sink=run_phase_subject,
use_model_picker=(args.rollout_idx == 0)), run_phase_subject))
agent_list.append(create_obstacles_agent())
agent_list.append(create_bot_cars_agent())
# ROS service to indicate all the robomaker markov packages are ready for consumption
signal_robomaker_markov_package_ready()
PhaseObserver('/agent/training_phase', run_phase_subject)
aws_region = rospy.get_param('AWS_REGION', args.aws_region)
simtrace_s3_bucket = rospy.get_param('SIMTRACE_S3_BUCKET', None)
mp4_s3_bucket = rospy.get_param('MP4_S3_BUCKET',
None) if args.rollout_idx == 0 else None
if simtrace_s3_bucket:
simtrace_s3_object_prefix = rospy.get_param('SIMTRACE_S3_PREFIX')
if args.num_workers > 1:
simtrace_s3_object_prefix = os.path.join(simtrace_s3_object_prefix,
str(args.rollout_idx))
if mp4_s3_bucket:
mp4_s3_object_prefix = rospy.get_param('MP4_S3_OBJECT_PREFIX')
simtrace_video_s3_writers = []
#TODO: replace 'agent' with 'agent_0' for multi agent training and
# mp4_s3_object_prefix, mp4_s3_bucket will be a list, so need to access with index
if simtrace_s3_bucket:
simtrace_video_s3_writers.append(
SimtraceVideo(
upload_type=SimtraceVideoNames.SIMTRACE_TRAINING.value,
bucket=simtrace_s3_bucket,
s3_prefix=simtrace_s3_object_prefix,
region_name=aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=SIMTRACE_TRAINING_LOCAL_PATH_FORMAT.format(
'agent')))
if mp4_s3_bucket:
simtrace_video_s3_writers.extend([
SimtraceVideo(
upload_type=SimtraceVideoNames.PIP.value,
bucket=mp4_s3_bucket,
s3_prefix=mp4_s3_object_prefix,
region_name=aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=CAMERA_PIP_MP4_LOCAL_PATH_FORMAT.format('agent')),
SimtraceVideo(
upload_type=SimtraceVideoNames.DEGREE45.value,
bucket=mp4_s3_bucket,
s3_prefix=mp4_s3_object_prefix,
region_name=aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=CAMERA_45DEGREE_LOCAL_PATH_FORMAT.format('agent')),
SimtraceVideo(
upload_type=SimtraceVideoNames.TOPVIEW.value,
bucket=mp4_s3_bucket,
s3_prefix=mp4_s3_object_prefix,
region_name=aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=CAMERA_TOPVIEW_LOCAL_PATH_FORMAT.format('agent'))
])
# TODO: replace 'agent' with specific agent name for multi agent training
ip_config = IpConfig(bucket=args.s3_bucket,
s3_prefix=args.s3_prefix,
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=IP_ADDRESS_LOCAL_PATH.format('agent'))
redis_ip = ip_config.get_ip_config()
# Download hyperparameters from SageMaker shared s3 bucket
# TODO: replace 'agent' with name of each agent
hyperparameters = Hyperparameters(
bucket=args.s3_bucket,
s3_key=get_s3_key(args.s3_prefix, HYPERPARAMETER_S3_POSTFIX),
region_name=args.aws_region,
s3_endpoint_url=args.s3_endpoint_url,
local_path=HYPERPARAMETER_LOCAL_PATH_FORMAT.format('agent'))
sm_hyperparams_dict = hyperparameters.get_hyperparameters_dict()
enable_domain_randomization = utils.str2bool(
rospy.get_param('ENABLE_DOMAIN_RANDOMIZATION', False))
if preset_file_success:
preset_location = os.path.join(CUSTOM_FILES_PATH, "preset.py")
preset_location += ":graph_manager"
graph_manager = short_dynamic_import(preset_location,
ignore_module_case=True)
logger.info("Using custom preset file!")
else:
graph_manager, _ = get_graph_manager(
hp_dict=sm_hyperparams_dict,
agent_list=agent_list,
run_phase_subject=run_phase_subject,
enable_domain_randomization=enable_domain_randomization)
# If num_episodes_between_training is smaller than num_workers then cancel worker early.
episode_steps_per_rollout = graph_manager.agent_params.algorithm.num_consecutive_playing_steps.num_steps
# Reduce number of workers if allocated more than num_episodes_between_training
if args.num_workers > episode_steps_per_rollout:
logger.info(
"Excess worker allocated. Reducing from {} to {}...".format(
args.num_workers, episode_steps_per_rollout))
args.num_workers = episode_steps_per_rollout
if args.rollout_idx >= episode_steps_per_rollout or args.rollout_idx >= args.num_workers:
err_msg_format = "Exiting excess worker..."
err_msg_format += "(rollout_idx[{}] >= num_workers[{}] or num_episodes_between_training[{}])"
logger.info(
err_msg_format.format(args.rollout_idx, args.num_workers,
episode_steps_per_rollout))
# Close the down the job
utils.cancel_simulation_job(
os.environ.get('AWS_ROBOMAKER_SIMULATION_JOB_ARN'),
rospy.get_param('AWS_REGION'))
memory_backend_params = DeepRacerRedisPubSubMemoryBackendParameters(
redis_address=redis_ip,
redis_port=6379,
run_type=str(RunType.ROLLOUT_WORKER),
channel=args.s3_prefix,
num_workers=args.num_workers,
rollout_idx=args.rollout_idx)
graph_manager.memory_backend_params = memory_backend_params
checkpoint_dict = {'agent': checkpoint}
ds_params_instance = S3BotoDataStoreParameters(
checkpoint_dict=checkpoint_dict)
graph_manager.data_store = S3BotoDataStore(ds_params_instance,
graph_manager)
task_parameters = TaskParameters()
task_parameters.checkpoint_restore_path = args.checkpoint_dir
rollout_worker(graph_manager=graph_manager,
num_workers=args.num_workers,
rollout_idx=args.rollout_idx,
task_parameters=task_parameters,
simtrace_video_s3_writers=simtrace_video_s3_writers)