params = json.load(f)
config['audio_rep'] = params
# set patch parameters
config['xInput'] = config['n_frames']
config['yInput'] = config['audio_rep']['n_mels']
# load audio representation paths
file_index = config_file.DATA_FOLDER + config[
'audio_representation_folder'] + 'index.tsv'
[audio_repr_paths,
id2audio_repr_path] = shared.load_id2audioReprPath(file_index)
# load training ground truth
file_ground_truth_train = config_file.DATA_FOLDER + config['gt_train']
[all_ids_train, id2gt_train] = shared.load_id2gt(file_ground_truth_train)
[_, id2label_train] = shared.load_id2label(file_ground_truth_train)
label2ids_train = shared.load_label2ids(id2label_train)
# load test ground truth
file_ground_truth_test = config_file.DATA_FOLDER + config['gt_test']
[all_ids_test, id2gt_test] = shared.load_id2gt(file_ground_truth_test)
[_, id2label_test] = shared.load_id2label(file_ground_truth_test)
label2ids_test = shared.load_label2ids(id2label_test)
# set output according to the experimental setup
classes_vector = list(range(config['num_classes_dataset']))
# tensorflow: define the model
with tf.name_scope('model'):
# which experiment we want to evaluate?
# Use the -l functionality to ensamble models: python arg.py -l 1234 2345 3456 4567
parser = argparse.ArgumentParser()
parser.add_argument('-l',
'--list',
nargs='+',
help='List of models to evaluate',
required=True)
args = parser.parse_args()
models = args.list
# load all audio representation paths
[audio_repr_paths, id2audio_repr_path] = shared.load_id2path(FILE_INDEX)
# load ground truth
[ids, id2gt] = shared.load_id2gt(FILE_GROUND_TRUTH_TEST)
print('# Test set', len(ids))
array_cost, pred_array, id_array = [], None, None
for model in models:
experiment_folder = config_file.DATA_FOLDER + 'experiments/' + str(
model) + '/'
config = json.load(open(experiment_folder + 'config.json'))
print('Experiment: ' + str(model))
print('\n' + str(config))
# pescador: define (finite, batched & parallel) streamer
pack = [config, 'overlap_sampling', config['n_frames'], False]
streams = [
params = json.load(f)
config['audio_rep'] = params
# set patch parameters
config['xInput'] = config['n_frames']
config['yInput'] = config['audio_rep']['n_mels']
# load audio representation paths
file_index = config_file.DATA_FOLDER + config[
'audio_representation_folder'] + 'index.tsv'
[audio_repr_paths,
id2audio_repr_path] = shared.load_id2audioReprPath(file_index)
# load training ground truth
file_ground_truth_train = config_file.DATA_FOLDER + config['gt_train']
[all_ids_train, id2gt_train] = shared.load_id2gt(file_ground_truth_train)
[_, id2label_train] = shared.load_id2label(file_ground_truth_train)
label2ids_train = shared.load_label2ids(id2label_train)
# load validation ground truth
file_ground_truth_val = config_file.DATA_FOLDER + config['gt_val']
[all_ids_val, id2gt_val] = shared.load_id2gt(file_ground_truth_val)
[_, id2label_val] = shared.load_id2label(file_ground_truth_val)
label2ids_val = shared.load_label2ids(id2label_val)
# set output according to the experimental setup
config['classes_vector'] = list(range(config['num_classes_dataset']))
# save experimental settings
experiment_id = 'fold_' + str(config_file.FOLD) + '_' + str(
shared.get_epoch_time())