conf.model = utils.abspath(conf.model)
conf.unetxst_homographies = utils.abspath(
conf.unetxst_homographies
) if conf.unetxst_homographies is not None else conf.unetxst_homographies
conf.model_weights = utils.abspath(
conf.model_weights
) if conf.model_weights is not None else conf.model_weights
conf.output_dir = utils.abspath(conf.output_dir)
# load network architecture module
architecture = utils.load_module(conf.model)
# get max_samples_training random training samples
n_inputs = len(conf.input_training)
files_train_input = [
utils.get_files_in_folder(folder) for folder in conf.input_training
]
files_train_label = utils.get_files_in_folder(conf.label_training)
_, idcs = utils.sample_list(files_train_label,
n_samples=conf.max_samples_training)
files_train_input = [np.take(f, idcs) for f in files_train_input]
files_train_label = np.take(files_train_label, idcs)
image_shape_original_input = utils.load_image(
files_train_input[0][0]).shape[0:2]
image_shape_original_label = utils.load_image(files_train_label[0]).shape[0:2]
print(f"Found {len(files_train_label)} training samples")
# get max_samples_validation random validation samples
files_valid_input = [
utils.get_files_in_folder(folder) for folder in conf.input_validation
]
def evalate_children():
true_positive = 0
false_positive = 0
true_negative = 0
false_negative = 0
total = 0
try:
os.mkdir(f'experiments/{config.experiment_name}')
except FileExistsError:
pass
with open(f'experiments/{config.experiment_name}/child_evaluation.csv', 'w') as csvfile:
writer = csv.writer(csvfile, delimiter=',', quotechar='|', quoting=csv.QUOTE_MINIMAL)
writer.writerow(config.csv_header)
files = utils.get_child_files_in_folder('blogs/10s')
for idx, file in enumerate(files):
print(f'[{idx+1}/{len(files)}]Getting file {file}')
try:
text = parse_xml(file)
except xmltodict.expat.ExpatError:
file_name = file.split('/')[2]
os.replace(file, f'trash/{file_name}')
except AttributeError:
file_name = file.split('/')[2]
os.replace(file, f'trash/{file_name}')
print(f'Getting response for file {file}')
actual_age = file.split('/')[2].split('.')[2]
try:
categories = get_nlu_reponse(text)
predicted_category = categories[0]['label']
if predicted_category == '/Child':
true_positive += 1
result = 'true_positive'
else:
false_negative += 1
result = 'false_negative'
except:
file_name = file.split('/')[2]
os.replace(file, f'trash/{file_name}')
result = 'true_positive'
row = [
file,
actual_age,
'Child',
predicted_category,
result
]
writer.writerow(row)
csvfile.flush()
total += 1
for adult_folder in adult_folders:
files = utils.get_files_in_folder(f'blogs/{adult_folder}')
for idx, file in enumerate(files):
print(f'[{idx+1}/{len(files)}]Getting file {file}')
try:
text = parse_xml(file)
except xmltodict.expat.ExpatError:
file_name = file.split('/')[2]
os.replace(file, f'trash/{file_name}')
except AttributeError:
file_name = file.split('/')[2]
os.replace(file, f'trash/{file_name}')
print(f'Getting response for file {file}')
actual_age = file.split('/')[2].split('.')[2]
try:
categories = get_nlu_reponse(text)
predicted_category = categories[0]['label']
if predicted_category == '/Child':
false_positive += 1
result = 'false_positive'
else:
true_negative += 1
result = 'true_negative'
except:
file_name = file.split('/')[2]
os.replace(file, f'trash/{file_name}')
result = 'true_negative'
row = [
file,
actual_age,
'Adult',
predicted_category,
result
]
writer.writerow(row)
csvfile.flush()
total += 1
with open(f'experiments/{config.experiment_name}/child_results.txt', 'w') as result_file:
result_file.write(f'Total: {total} \n')
result_file.write(f'True positives: {true_positive} [{true_positive/total * 100}%]\n')
result_file.write(f'True negatives: {true_negative} [{true_negative / total * 100}%]\n')
result_file.write(f'False positives: {false_positive} [{false_positive / total * 100}%]\n')
result_file.write(f'False negatives: {false_negative} [{false_negative / total * 100}%]\n')
result_file.write(f'acc: {true_positive + true_negative} [{(true_positive + true_negative) / total * 100}%]\n')
conf.input_testing = [utils.abspath(path) for path in conf.input_testing]
conf.one_hot_palette_input = utils.abspath(conf.one_hot_palette_input)
conf.one_hot_palette_label = utils.abspath(conf.one_hot_palette_label)
conf.model = utils.abspath(conf.model)
conf.unetxst_homographies = utils.abspath(
conf.unetxst_homographies
) if conf.unetxst_homographies is not None else conf.unetxst_homographies
conf.model_weights = utils.abspath(conf.model_weights)
conf.prediction_dir = utils.abspath(conf.prediction_dir)
# load network architecture module
architecture = utils.load_module(conf.model)
# get max_samples_testing samples
files_input = [
utils.get_files_in_folder(folder) for folder in conf.input_testing
]
_, idcs = utils.sample_list(files_input[0], n_samples=conf.max_samples_testing)
files_input = [np.take(f, idcs) for f in files_input]
n_inputs = len(conf.input_testing)
n_samples = len(files_input[0])
image_shape_original = utils.load_image(files_input[0][0]).shape[0:2]
print(f"Found {n_samples} samples")
# parse one-hot-conversion.xml
conf.one_hot_palette_input = utils.parse_convert_xml(
conf.one_hot_palette_input)
conf.one_hot_palette_label = utils.parse_convert_xml(
conf.one_hot_palette_label)
n_classes_input = len(conf.one_hot_palette_input)
n_classes_label = len(conf.one_hot_palette_label)
conf.input_validation = [utils.abspath(path) for path in conf.input_validation]
conf.label_validation = utils.abspath(conf.label_validation)
conf.one_hot_palette_input = utils.abspath(conf.one_hot_palette_input)
conf.one_hot_palette_label = utils.abspath(conf.one_hot_palette_label)
conf.model = utils.abspath(conf.model)
conf.unetxst_homographies = utils.abspath(
conf.unetxst_homographies
) if conf.unetxst_homographies is not None else conf.unetxst_homographies
conf.model_weights = utils.abspath(conf.model_weights)
# load network architecture module
architecture = utils.load_module(conf.model)
# get max_samples_validation random validation samples
files_input = [
utils.get_files_in_folder(folder) for folder in conf.input_validation
]
files_label = utils.get_files_in_folder(conf.label_validation)
_, idcs = utils.sample_list(files_label, n_samples=conf.max_samples_validation)
files_input = [np.take(f, idcs) for f in files_input]
files_label = np.take(files_label, idcs)
n_inputs = len(conf.input_validation)
n_samples = len(files_label)
image_shape_original_input = utils.load_image(files_input[0][0]).shape[0:2]
image_shape_original_label = utils.load_image(files_label[0]).shape[0:2]
print(f"Found {n_samples} samples")
# parse one-hot-conversion.xml
conf.one_hot_palette_input = utils.parse_convert_xml(
conf.one_hot_palette_input)
conf.one_hot_palette_label = utils.parse_convert_xml(