"MIMIC": percentage_sigs_mimic,
"eICU": percentage_sigs_eicu
}
stats_dir = args.stats_dir
if not os.path.exists(f"{stats_dir}/DA/"):
os.makedirs(f"{stats_dir}/DA/")
with open(f"{stats_dir}/DA/percentage_sigs.pkl", "wb") as f:
pickle.dump(percentage_sigs, f)
del all_eicu, all_mimic
for mimic_model_info, eicu_model_info in zip(
init_models(input_dim=len(feature_names_eicu),
selection=args.models,
origin="MIMIC"),
init_models(input_dim=len(feature_names_eicu),
selection=args.models,
origin="eICU"),
):
print(mimic_model_info[2])
ood_detect_aucs, ood_recall = (
defaultdict(lambda: defaultdict(list)),
defaultdict(lambda: defaultdict(list)),
)
metrics = defaultdict(lambda: defaultdict(list))
ood_detect_aucs, ood_recall, metrics = ood_utils.run_ood_experiment_on_group(
id_data=mimic_data,
ood_data=eicu_data,
y_name = dh.load_target_name()
pipe = pipeline.Pipeline([("scaler", StandardScaler()),
("imputer", SimpleImputer())])
pipe.fit(train_data[feature_names])
X_train = pipe.transform(train_data[feature_names])
X_test = pipe.transform(test_data[feature_names])
X_val = pipe.transform(val_data[feature_names])
uncertainties = defaultdict(list)
for ne, scoring_funcs, method_name in init_models(
input_dim=len(feature_names),
selection=args.models,
origin=args.data_origin,
):
print(method_name)
predictions = []
for i in tqdm(range(N_SEEDS)):
ne.train(X_train, train_data[y_name].values, X_val,
val_data[y_name].values)
for scoring_func in scoring_funcs:
uncertainties[scoring_func] += [
ne.get_novelty_score(X_test, scoring_func=scoring_func)
]
print(len(uncertainties[scoring_func][0]))
# Loading the data
data_loader = load_data_from_origin(args.data_origin)
dh = DataHandler(**data_loader)
feature_names = dh.load_feature_names()
train_data, test_data, val_data = dh.load_data_splits()
y_name = dh.load_target_name()
if args.data_origin in MIMIC_ORIGINS:
train_newborns, test_newborns, val_newborns = dh.load_other_groups("newborns")
ood_mappings = dh.load_ood_mappings()
# loop over the different methods
for model_info in init_models(
input_dim=len(feature_names), selection=args.models, origin=args.data_origin
):
print(model_info[2])
ood_detect_aucs, ood_recall = (
defaultdict(lambda: defaultdict(list)),
defaultdict(lambda: defaultdict(list)),
)
metrics = defaultdict(lambda: defaultdict(list))
# Experiments on Newborns, only on MIMIC for now
if args.data_origin in MIMIC_ORIGINS:
id_data = DomainData(
train_data, test_data, val_data, feature_names, y_name, "in-domain"
)
ood_data = DomainData(
train_newborns,