def load_dagger_generator(args):
port = args.port
ports = args.ports
assert ports
assert port in ports
assert args.dataset_name
assert args.eval_dataset_name
port_index = ports.index(port)
eval_param = Parameter()
eval_param.exp_index = args.exp_index
eval_param.exp_name = args.exp_name
eval_param.load()
eval_param.batch_size = 1
eval_param.dataset_data_names = [args.dataset_name]
eval_param.eval_data_name = args.eval_dataset_name
eval_param.max_data_length = -1
if eval_param.split_train:
train_dataset, valid_dataset = fetch_dataset_pair(eval_param)
else:
train_dataset = fetch_dataset(eval_param)
num_data = len(train_dataset)
index_func = partial(fetch_index_from_road_option,
low_level=eval_param.use_low_level_segment)
data_list = []
for i in range(num_data):
# road_option, data_frame = train_dataset.get_trajectory_data(i)
# images = data_frame.images
# drives = data_frame.drives
# fixme: this may not work if train_dataset is a HighLevelDataset? Cuz it does not even have this method!
road_option, images, drives = train_dataset.get_trajectory_data(i)
data_list.append({
'road_option': road_option,
'action_index': index_func(road_option),
'src_transform': drives[0].state.transform,
'dst_location': drives[-1].state.transform.location,
'length': len(images)
})
def chunker_list(seq, size):
return (seq[i::size] for i in range(size))
index_data_lists = list(
chunker_list(list(enumerate(data_list)), len(ports)))
index_data_list = index_data_lists[port_index]
return DaggerGeneratorEnvironment(args, eval_param, index_data_list)
def _prepare_evaluation_param(param: Parameter) -> Parameter:
assert param.eval_data_name
assert param.eval_info_name
assert param.eval_keyword
if param.model_level == 'low':
param.eval_keyword = fetch_road_option_from_str(param.eval_keyword.upper())
elif param.model_level == 'high':
param.eval_keyword = param.eval_keyword.lower()
else:
logger.info(param.model_level)
raise TypeError('invalid eval_keyword was given {}'.format(param.eval_keyword))
param.max_data_length = -1
param.shuffle = False
param.batch_size = 1
param.dataset_data_names = [param.eval_data_name]
param.dataset_info_names = [param.eval_info_name]
if param.model_level == 'low':
param.use_multi_cam = False
param.use_sequence = False
param.has_clusters = False
return param
