def convert_hf_local_dataset_to_examples(path,
name=None,
version=None,
field_map=None,
label_map=None,
phase_map=None,
phase_list=None):
"""Helper function for reading from datasets.load_dataset and converting to examples
Args:
path: path argument (from datasets.load_dataset)
name: name argument (from datasets.load_dataset)
version: version argument (from datasets.load_dataset)
field_map: dictionary for renaming fields, non-exhaustive
label_map: dictionary for replacing labels, non-exhaustive
phase_map: dictionary for replacing phase names, non-exhaustive
phase_list: phases to keep (after phase_map)
Returns:
Dict[phase] -> list[examples]
"""
print('pppath', os.path.dirname(os.path.abspath(__file__)), path)
dataset = datasets.load_from_disk(
os.path.dirname(os.path.abspath(__file__)) + '/' + path)
if phase_map:
for old_phase_name, new_phase_name in phase_map.items():
replace_key(dataset,
old_key=old_phase_name,
new_key=new_phase_name)
if phase_list is None:
phase_list = dataset.keys()
examples_dict = {}
for phase in phase_list:
phase_examples = []
for raw_example in dataset[phase]:
if field_map:
for old_field_name, new_field_name in field_map.items():
replace_key(raw_example,
old_key=old_field_name,
new_key=new_field_name)
if label_map and "label" in raw_example:
# Optionally use an dict or function to map labels
label = raw_example["label"]
if isinstance(label_map, dict):
if raw_example["label"] in label_map:
label = label_map[raw_example["label"]]
elif callable(label_map):
label = label_map(raw_example["label"])
else:
raise TypeError(label_map)
raw_example["label"] = label
phase_examples.append(raw_example)
examples_dict[phase] = phase_examples
return examples_dict
def convert_nlp_dataset_to_examples(path,
name=None,
version=None,
field_map=None,
label_map=None,
phase_map=None,
phase_list=None):
"""Helper function for reading from nlp.load_dataset and converting to examples
Args:
path: path argument (from nlp.load_dataset)
name: name argument (from nlp.load_dataset)
version: version argument (from nlp.load_dataset)
field_map: dictionary for renaming fields, non-exhaustive
label_map: dictionary for replacing labels, non-exhaustive
phase_map: dictionary for replacing phase names, non-exhaustive
phase_list: phases to keep (after phase_map)
Returns:
Dict[phase] -> list[examples]
"""
dataset = nlp.load_dataset(path=path, name=name, version=version)
if phase_map:
for old_phase_name, new_phase_name in phase_map.items():
replace_key(dataset,
old_key=old_phase_name,
new_key=new_phase_name)
if phase_list is None:
phase_list = dataset.keys()
examples_dict = {}
for phase in phase_list:
phase_examples = []
for raw_example in dataset[phase]:
if field_map:
for old_field_name, new_field_name in field_map.items():
replace_key(raw_example,
old_key=old_field_name,
new_key=new_field_name)
if label_map and "label" in raw_example and raw_example[
"label"] in label_map:
raw_example["label"] = label_map[raw_example["label"]]
phase_examples.append(raw_example)
examples_dict[phase] = phase_examples
return examples_dict
def load_hf_dataset(path,
name=None,
version=None,
phase_map=None,
n_fold: int = None,
fold: int = None,
split_type: str = None):
phase_map = phase_map or {}
split_type = split_type or 'local_evaluation'
dataset_dict = load_dataset(path=path, name=name, version=version)
for old_phase_name, new_phase_name in phase_map.items():
replace_key(dataset_dict,
old_key=old_phase_name,
new_key=new_phase_name)
if n_fold is not None:
jiant2hf_phase_map = {val: key for key, val in phase_map.items()}
hf_train_phase = jiant2hf_phase_map.get('train', 'train')
hf_val_phase = jiant2hf_phase_map.get('val', 'val')
if split_type == 'local_evaluation':
hf_local_phase = '+'.join([hf_train_phase])
dataset_dict['test'] = dataset_dict['val']
elif split_type == 'submission':
hf_local_phase = '+'.join([hf_train_phase, hf_val_phase])
else:
raise ValueError(f'Unknown split type "{split_type}"')
hf_local_dataset = load_dataset(path=path,
name=name,
version=version,
split=hf_local_phase)
cv_dataset = build_cv_splits(
hf_local_dataset,
n_fold,
stratify=False,
)
dataset_dict['train'] = cv_dataset[f'fold-{fold}.train']
dataset_dict['val'] = cv_dataset[f'fold-{fold}.val']
return dataset_dict
def convert_hf_dataset_to_examples(
path,
name=None,
version=None,
field_map=None,
label_map=None,
phase_map=None,
phase_list=None,
n_fold: int = None,
fold: int = None,
return_all: bool = False,
experiment_id_for_metric=None,
cache_dir_for_metric=None,
):
"""Helper function for reading from datasets.load_dataset and converting to examples
Args:
path: path argument (from datasets.load_dataset)
name: name argument (from datasets.load_dataset)
version: version argument (from datasets.load_dataset)
field_map: dictionary for renaming fields, non-exhaustive
label_map: dictionary for replacing labels, non-exhaustive
phase_map: dictionary for replacing phase names, non-exhaustive
phase_list: phases to keep (after phase_map)
Returns:
Dict[phase] -> list[examples]
"""
# "mrpc.cv-5-0"
fold_dataset = load_hf_dataset(path=path,
name=name,
version=version,
phase_map=phase_map,
n_fold=n_fold,
fold=fold)
if phase_list is None:
phase_list = fold_dataset.keys()
examples_dict = {}
for phase in phase_list:
phase_examples = []
for raw_example in fold_dataset[phase]:
if field_map:
for old_field_name, new_field_name in field_map.items():
replace_key(raw_example,
old_key=old_field_name,
new_key=new_field_name)
if label_map and "label" in raw_example:
# Optionally use an dict or function to map labels
label = raw_example["label"]
if isinstance(label_map, dict):
if raw_example["label"] in label_map:
label = label_map[raw_example["label"]]
elif callable(label_map):
label = label_map(raw_example["label"])
else:
raise TypeError(label_map)
raw_example["label"] = label
phase_examples.append(raw_example)
examples_dict[phase] = phase_examples
metric = load_metric(path,
config_name=name,
experiment_id=experiment_id_for_metric,
cache_dir=cache_dir_for_metric)
if return_all:
return examples_dict, {
'hf_fold_dataset': fold_dataset,
# 'hf_full_dataset': full_dataset,
'hf_metric': metric
}
else:
return examples_dict