def main(args):
x_train, y_train, feature_names, selected_pixels_train, paths_train = ml_utils.create_samples(
args.dataset_names[0], args.dataset_paths[0], args.mat_files[0],
args.balanced[0], bool(strtobool(args.save_images)))
x_test, y_test, feature_names, selected_pixels_test, paths_test = ml_utils.create_samples(
args.dataset_names[1], args.dataset_paths[1], args.mat_files[1],
args.balanced[1], bool(strtobool(args.save_images)))
selected_indices, selected_features = ml_utils.get_feature_subset(
args, feature_names)
log_string = "Selected features:\n%s\n" % selected_features
print(log_string)
parameter_dict = ml_utils.get_parameter_dict(args)
clf = available_models[args.model](**parameter_dict)
scoring = {}
for metric in args.metrics:
score_func = available_scores[metric]
scoring[metric] = score_func if score_func is not None else None
test_images = [
"'%s'" % os.path.split(paths_test[0][idx])[-1]
for idx in range(len(paths_test[0]))
]
test_images = "[%s]" % ", ".join(test_images)
if not os.path.exists(args.save_results_to):
os.makedirs(args.save_results_to)
with open(os.path.join(args.save_results_to, "results.txt"), "w") as f:
f.write(log_string)
print("Classifier: %s" % str(clf))
f.write("\nClassifier: %s\n" % str(clf))
clf.fit(x_train, y_train)
predictions = clf.predict(x_test)
dsc = False
for metric in scoring.keys():
if metric == "dsc":
dsc = True
continue
if scoring[metric] is None:
value = 100 * clf.score(x_test, y_test)
else:
value = 100 * scoring[metric](y_test, predictions)
print("{}(%) = {:.2f}".format(metric, value))
f.write("\n{}(%) = {:.2f}\n".format(metric, value))
print("\nTest images per fold (dataset %s):\n%s" %
(args.dataset_names[1], test_images))
f.write("\nTest images per fold (dataset %s):\n%s" %
(args.dataset_names[1], test_images))
ml_utils.save_visual_results(selected_pixels_test, predictions, y_test,
paths_test, args.save_results_to)
if dsc:
ml_utils.calculate_dsc_from_result_folder(args.save_results_to)
def one_dataset_crossvalidation(selected_features="all"):
dataset_name = "cfd-pruned"
dataset_folder = "/media/winbuntu/databases/CrackForestDatasetPruned"
mat_path = "cfd-pruned.mat"
# mat_path = None
save_images = False
x, y, feature_names, selected_pixels, paths = ml_utils.create_samples(
dataset_name, dataset_folder, mat_path, False, save_images)
train_sgd_models(x,
y,
feature_names,
selected_pixels,
paths,
selected_features=selected_features)
def main(args):
x, y, feature_names, selected_pixels, paths = ml_utils.create_samples(args.dataset_name, args.dataset_path, None,
args.balanced,
bool(strtobool(args.save_images)))
return x, y, feature_names, selected_pixels, paths
def main(args):
x, y, feature_names, selected_pixels, paths = ml_utils.create_samples(args.dataset_name, args.dataset_path,
args.mat_file, args.balanced,
bool(strtobool(args.save_images)))
ml_utils.n_fold_cross_validation(args, available_models, available_scores, x, y, feature_names, selected_pixels,
paths)
dataset_name_2 = "cfd-pruned"
dataset_folder_2 = "/media/winbuntu/databases/CrackForestDatasetPruned"
mat_path_2 = "cfd-pruned.mat"
# mat_path_2 = None
balanced_2 = False
save_images_2 = False
test_models_cross_dataset([dataset_name_1, dataset_name_2],
[dataset_folder_1, dataset_folder_2],
[mat_path_1, mat_path_2],
[balanced_1, balanced_2],
[save_images_1, save_images_2],
selected_features=selected_features)
# selected_features = "Frangi's vesselness;Bottom-hat;Cross bottom-hat;Sliding mean 50x50;Sliding std 50x50;Sliding median 50x50;Sliding mad 50x50"
selected_features = "all"
# one_dataset_balanced_crossvalidation(selected_features)
dataset_name = "aigle-rn"
dataset_folder = "/media/winbuntu/databases/CrackDataset"
# mat_path = "cfd-pruned.mat"
mat_path = None
balanced = False
save_images = True
x, y, feature_names, selected_pixels, paths = ml_utils.create_samples(
dataset_name, dataset_folder, mat_path, balanced, save_images)
# feature_selection(x, y, feature_names)