def prepare_and_run(subjects, classes, out_dir, options_path,
user_feature_paths, user_feature_type, fs_subject_dir,
train_perc, num_rep_cv, positive_class, sub_group_list,
feature_selection_size, num_procs, grid_search_level,
classifier, feat_select_method):
"Organizes the inputs and prepares them for CV"
feature_dir, method_list = make_method_list(fs_subject_dir,
user_feature_paths,
user_feature_type)
method_names, dataset_paths_file = import_datasets(method_list, out_dir,
subjects, classes,
feature_dir,
user_feature_type)
print('Requested processing for the following subgroups:'
'\n{}\n'.format('\n'.join([','.join(sg) for sg in sub_group_list])))
# iterating through the given set of subgroups
num_sg = len(sub_group_list)
for sgi, sub_group in enumerate(sub_group_list):
print('{}\nProcessing subgroup : {} ({}/{})'
'\n{}'.format('-' * 80, ','.join(sub_group), sgi + 1, num_sg,
'-' * 80))
out_dir_sg = pjoin(out_dir, sub_group_identifier(sub_group))
results_file_path = rhst.run(dataset_paths_file,
method_names,
out_dir_sg,
train_perc=train_perc,
num_repetitions=num_rep_cv,
positive_class=positive_class,
sub_group=sub_group,
feat_sel_size=feature_selection_size,
num_procs=num_procs,
grid_search_level=grid_search_level,
classifier_name=classifier,
feat_select_method=feat_select_method,
options_path=options_path)
print('\n\nSaving the visualizations to \n{}'.format(out_dir))
make_visualizations(results_file_path, out_dir_sg, options_path)
print('\n')
return
def load_results_from_folder(results_folder):
"""
Given a base output folder, possibly containing results for multiple sub-groups,
returns a dictionary of results, keyed in by sub group identifier.
"""
results = dict()
options = load_options(results_folder)
for ix, sg in enumerate(options['sub_groups']):
sg_id = sub_group_identifier(sg, ix)
results_file_path = pjoin(results_folder, sg_id, cfg.file_name_results)
if not pexists(
results_file_path) or os.path.getsize(results_file_path) <= 0:
raise IOError('Results file for sub group {} does not exist'
' or is empty!'.format(sg_id))
results[sg_id] = load_results_dict(results_file_path)
return results
def cli():
""" Main entry point. """
print('\nneuropredict version {} for Classification'.format(__version__))
from datetime import datetime
init_time = datetime.now()
print('\tTime stamp : {}\n'.format(
init_time.strftime('%Y-%m-%d %H:%M:%S')))
subjects, classes, out_dir, options_path, user_feature_paths, \
user_feature_type, fs_subject_dir, train_perc, num_rep_cv, positive_class, \
sub_group_list, feature_selection_size, impute_strategy, num_procs, \
grid_search_level, classifier, feat_select_method, covar_list, covar_method = \
parse_args()
feature_dir, method_list = make_method_list(fs_subject_dir,
user_feature_paths,
user_feature_type)
# noinspection PyTupleAssignmentBalance
method_names, outpath_list = import_datasets(method_list, out_dir,
subjects, classes,
feature_dir,
user_feature_type)
print('Requested processing for the following subgroups:'
'\n{}\n'.format('\n'.join([','.join(sg) for sg in sub_group_list])))
# iterating through the given set of subgroups
num_sg = len(sub_group_list)
result_paths = dict()
for sgi, sub_group in enumerate(sub_group_list):
print('{line}\nProcessing subgroup : {id_} ({idx}/{cnt})\n{line}'
''.format(line='-' * 80,
id_=','.join(sub_group),
idx=sgi + 1,
cnt=num_sg))
sub_group_id = sub_group_identifier(sub_group, sg_index=sgi + 1)
out_dir_sg = pjoin(out_dir, sub_group_id)
multi_ds = load_datasets(outpath_list,
task_type='classify',
subgroup=sub_group,
name=sub_group_id)
covariates, deconfounder = check_covariates(multi_ds, covar_list,
covar_method)
print(multi_ds)
impute_strategy = detect_missing_data(multi_ds, impute_strategy)
clf_expt = ClassificationWorkflow(
datasets=multi_ds,
pred_model=classifier,
impute_strategy=impute_strategy,
dim_red_method=feat_select_method,
covariates=covariates,
deconfounder=deconfounder,
reduced_dim=feature_selection_size,
train_perc=train_perc,
num_rep_cv=num_rep_cv,
scoring=cfg.default_metric_set_classification,
positive_class=positive_class,
grid_search_level=grid_search_level,
out_dir=out_dir_sg,
num_procs=num_procs,
user_options=options_path,
checkpointing=cfg.default_checkpointing)
result_paths[sub_group_id] = clf_expt.run()
timedelta = datetime.now() - init_time
print('All done. Elapsed time: {} HH:MM:SS\n'.format(timedelta))
return result_paths