def main(settingsfname,
verbose=False,
store_models=True,
store_features=False,
save_training_detailed=False,
load_pickled=False,
parallel=0):
settings = utils.get_settings(settingsfname)
utils.print_verbose('=== Settings file ===', flag=verbose)
utils.print_verbose(settingsfname, flag=verbose)
utils.print_verbose('=== Settings loaded ===', flag=verbose)
utils.print_verbose(settings, flag=verbose)
utils.print_verbose('=======================', flag=verbose)
subjects = settings['SUBJECTS']
data = utils.get_data(settings, verbose=verbose)
metadata = utils.get_metadata()
features_that_parsed = [
feature for feature in settings['FEATURES']
if feature in list(data.keys())
]
settings['FEATURES'] = features_that_parsed
if not settings['FEATURES']:
raise EnvironmentError('No features could be loaded')
utils.print_verbose("=====Feature HDF5s parsed=====", flag=verbose)
model_pipe = utils.build_model_pipe(settings)
utils.print_verbose("=== Model Used ===\n"
"{0}\n==================".format(model_pipe),
flag=verbose)
# dictionary to store results
subject_predictions = {}
# dictionary to store features in
transformed_features = {}
# if we're loading pickled features then load them
if load_pickled:
if isinstance(load_pickled, str):
with open(load_pickled, "rb") as fh:
Xtra = pickle.load(fh)
else:
with open(
settingsfname.split(".")[0] + "_feature_dump.pickle",
"rb") as fh:
Xtra = pickle.load(fh)
else:
Xtra = None
# dictionary for final scores
auc_scores = {}
if not parallel:
for subject in subjects:
utils.print_verbose("=====Training {0} Model=====".format(
str(subject)),
flag=verbose)
if 'RFE' in settings:
transformed_features, auc = utils.train_RFE(
settings,
data,
metadata,
subject,
model_pipe,
transformed_features,
store_models,
store_features,
load_pickled,
settingsfname,
verbose,
extra_data=Xtra)
subject_predictions = None
elif 'CUSTOM' in settings:
results, auc = utils.train_custom_model(settings,
data,
metadata,
subject,
model_pipe,
store_models,
load_pickled,
verbose,
extra_data=Xtra)
subject_predictions[subject] = results
else:
results, auc = utils.train_model(settings,
data,
metadata,
subject,
model_pipe,
store_models,
load_pickled,
verbose,
extra_data=Xtra)
subject_predictions[subject] = results
auc_scores.update({subject: auc})
if parallel:
if 'RFE' in settings:
raise NotImplementedError('Parallel RFE is not implemented')
else:
output = joblib.Parallel(n_jobs=parallel)(
joblib.delayed(utils.train_model)(settings,
data,
metadata,
subject,
model_pipe,
store_models,
load_pickled,
verbose,
extra_data=Xtra,
parallel=parallel)
for subject in subjects)
results = [x[0] for x in output]
aucs = [x[1] for x in output]
for result in results:
subject_predictions.update(result)
for auc in aucs:
auc_scores.update(auc)
if save_training_detailed:
with open(save_training_detailed, "wb") as fh:
pickle.dump(subject_predictions[subject], fh)
combined_auc = utils.combined_auc_score(settings,
auc_scores,
subj_pred=subject_predictions)
print(
"predicted AUC score over all subjects: {0:.2f}".format(combined_auc))
auc_scores.update({'all': combined_auc})
utils.output_auc_scores(auc_scores, settings)
return auc_scores
def main(settings_file='SETTINGS.json', verbose=False):
# load the settings
settings = utils.get_settings(settings_file)
subjects = settings['SUBJECTS']
features = settings['FEATURES']
# load the data
data = utils.get_data(settings)
# load the metadata
metadata = utils.get_metadata()
features_that_parsed = list(data.keys())
settings['FEATURES'] = [
feature for feature in settings['FEATURES']
if feature in features_that_parsed
]
# check if features are 1-minute
if all('10feat' in feature for feature in settings['FEATURES']):
# set the flage
minutefeatures = True
elif not all('10feat' in feature for feature in settings['FEATURES']) and \
any('10feat' in feature for feature in settings['FEATURES']):
raise ValueError("Cannot mix 1-minute and 10-minute features.")
else:
minutefeatures = False
# iterate over subjects
prediction_dict = {}
for subject in subjects:
# load the trained model:
model = utils.read_trained_model(subject, settings, verbose=verbose)
# initialise the data assembler
assembler = utils.DataAssembler(settings, data, metadata)
# build test set
X = assembler.build_test(subject)
# make predictions
predictions = model.predict_proba(X)
# if using minute features combine the estimates
# on each segment by averaging
if minutefeatures:
segmentdict = {}
for segment, prediction in zip(assembler.test_segments,
predictions):
if segment not in segmentdict:
segmentdict[segment] = []
segmentdict[segment].append(prediction)
# gathered all predictions corresponding to a segment together
# now average them along their columns
for segment in assembler.test_segments:
segmentdict[segment] = np.vstack(segmentdict[segment])
segmentdict[segment] = np.mean(segmentdict[segment], axis=0)
for segment, prediction in zip(assembler.test_segments, predictions):
prediction_dict[segment] = prediction
utils.output_csv(prediction_dict, settings, verbose=verbose)
def main(settingsfname, verbose=False, store_models=True,
store_features=False, save_training_detailed=False,
load_pickled=False, parallel=0):
settings = utils.get_settings(settingsfname)
utils.print_verbose('=== Settings file ===', flag=verbose)
utils.print_verbose(settingsfname, flag=verbose)
utils.print_verbose('=== Settings loaded ===', flag=verbose)
utils.print_verbose(settings, flag=verbose)
utils.print_verbose('=======================', flag=verbose)
subjects = settings['SUBJECTS']
data = utils.get_data(settings, verbose=verbose)
metadata = utils.get_metadata()
features_that_parsed = [feature for feature in
settings['FEATURES'] if feature in list(data.keys())]
settings['FEATURES'] = features_that_parsed
if not settings['FEATURES']:
raise EnvironmentError('No features could be loaded')
utils.print_verbose("=====Feature HDF5s parsed=====", flag=verbose)
model_pipe = utils.build_model_pipe(settings)
utils.print_verbose("=== Model Used ===\n"
"{0}\n==================".format(model_pipe),
flag=verbose)
# dictionary to store results
subject_predictions = {}
# dictionary to store features in
transformed_features = {}
# if we're loading pickled features then load them
if load_pickled:
if isinstance(load_pickled, str):
with open(load_pickled, "rb") as fh:
Xtra = pickle.load(fh)
else:
with open(settingsfname.split(".")[0]
+ "_feature_dump.pickle", "rb") as fh:
Xtra = pickle.load(fh)
else:
Xtra = None
# dictionary for final scores
auc_scores = {}
if not parallel:
for subject in subjects:
utils.print_verbose(
"=====Training {0} Model=====".format(str(subject)),
flag=verbose)
if 'RFE' in settings:
transformed_features, auc = utils.train_RFE(settings,
data,
metadata,
subject,
model_pipe,
transformed_features,
store_models,
store_features,
load_pickled,
settingsfname,
verbose,
extra_data=Xtra)
subject_predictions = None
elif 'CUSTOM' in settings:
results, auc = utils.train_custom_model(settings,
data,
metadata,
subject,
model_pipe,
store_models,
load_pickled,
verbose,
extra_data=Xtra)
subject_predictions[subject] = results
else:
results, auc = utils.train_model(settings,
data,
metadata,
subject,
model_pipe,
store_models,
load_pickled,
verbose,
extra_data=Xtra)
subject_predictions[subject] = results
auc_scores.update({subject: auc})
if parallel:
if 'RFE' in settings:
raise NotImplementedError('Parallel RFE is not implemented')
else:
output = joblib.Parallel(n_jobs=parallel)(
joblib.delayed(utils.train_model)(settings,
data,
metadata,
subject,
model_pipe,
store_models,
load_pickled,
verbose,
extra_data=Xtra,
parallel=parallel)
for subject in subjects)
results = [x[0] for x in output]
aucs = [x[1] for x in output]
for result in results:
subject_predictions.update(result)
for auc in aucs:
auc_scores.update(auc)
if save_training_detailed:
with open(save_training_detailed, "wb") as fh:
pickle.dump(subject_predictions[subject], fh)
combined_auc = utils.combined_auc_score(settings,
auc_scores,
subj_pred=subject_predictions)
print(
"predicted AUC score over all subjects: {0:.2f}".format(combined_auc))
auc_scores.update({'all': combined_auc})
utils.output_auc_scores(auc_scores, settings)
return auc_scores
