def get_results(path, metric='dice', metrics_dir='test_metrics'):
from os.path import exists, join as jp
scores = []
ids = []
i = 0
while exists(jp(path, f'experiment_{i}')):
temp_json = load_json(
f'{path}/experiment_{i}/{metrics_dir}/{metric}.json')
scores.extend(list(temp_json.values()))
ids.extend(list(temp_json.keys()))
i += 1
return np.array(scores), np.array(ids), i
def exp2prc_df(exp_path, n_val=5, specific_ids=None):
"""Constructs pandas DataFrame with prc data from all predictions in ``exp_path``."""
dfs = []
for n in range(n_val):
prc_path = jp(exp_path, f'experiment_{n}', 'test_metrics',
'prc_records.json')
prc_dicts = load_json(prc_path)
for _id in prc_dicts.keys():
if specific_ids is None:
[d.update({'id': _id}) for d in prc_dicts[_id]]
dfs.append(pd.DataFrame.from_records(prc_dicts[_id]))
else:
if _id in specific_ids:
[d.update({'id': _id}) for d in prc_dicts[_id]]
dfs.append(pd.DataFrame.from_records(prc_dicts[_id]))
df = pd.concat(dfs)
return df
def load_results(experiment_path):
results = defaultdict(dict)
for fold_name in os.listdir(experiment_path):
if fold_name.startswith('experiment'):
for s in sets:
for metric in metrics:
try:
partial_result = load_json(
join(experiment_path, fold_name, s + '_metrics',
metric + '.json'))
if hasattr(
partial_result[list(partial_result.keys())[0]],
'__iter__'):
partial_result = {
i: np.array(r).flatten()
for i, r in partial_result.items()
}
results[f'{s}_{metric}'].update(partial_result)
except FileNotFoundError:
pass
return results
SPLIT_PATH = Path(sys.argv[2])
EXPERIMENT_PATH = Path(sys.argv[3])
FOLD = sys.argv[4]
CONFIG = {
'source_slice_spacing': 1.,
'target_slice_spacing': np.linspace(1, 5, 9),
'batch_size': 5,
'batches_per_epoch': 100,
'n_epochs': 300,
'lr': 1e-4,
'device': 'cuda',
}
try:
CONFIG.update(load_json(sys.argv[5]))
except (IndexError, FileNotFoundError):
pass
PATCH_SIZE = np.array([64, 64, 32])
# dataset
raw_dataset = BraTS2013(BRATS_PATH)
dataset = apply(CropToBrain(raw_dataset), load_image=partial(min_max_scale, axes=0))
train_dataset = cache_methods(ChangeSliceSpacing(dataset, new_slice_spacing=CONFIG['source_slice_spacing']))
# cross validation
split = load_json(SPLIT_PATH)
train_ids, val_ids, test_ids = split[int(FOLD)]
# batch iterator