[-1] + _input_shape)
_test_labels = to_categorical(test_pairs['label'].tolist())
# Build model
_keras_model = VggishModel(class_weights=_class_weights,
input_shape=_input_shape,
num_classes=_num_classes,
batch_size=64,
learning_rate=0.0001,
metric_baseline=0.5,
num_epochs=30,
load_pretrained=True,
feature_type='raw',
output_dir='outputs')
# Build container.
_container = DataContainer(data=_train_pairs,
labeled_percent=0.1,
num_classes=_num_classes,
feature_shape=_keras_model.input_shape,
label_shape=_keras_model.output_shape)
print(_keras_model.get_model_summary())
run_entropy(data_container=_container,
model=_keras_model,
test_features=_test_features,
test_lablels=_test_labels,
stats_path='outputs/stats/al_stats.pd',
max_round=60)
if flags.framework_type == 'entropy' or \
flags.framework_type == 'edpc' or \
flags.framework_type == 'hist_select':
_embeddings = _keras_model.embedding(
np.reshape(_pairs['feature'].tolist(), [-1] + _input_shape))
else:
_embeddings = None
# Reset weights params.
_keras_model.reload_weights()
# Build container.
_container = DataContainer(data=_train_pairs,
labeled_percent=flags.labeled_percent,
num_classes=_num_classes,
feature_shape=_keras_model.input_shape,
label_shape=_keras_model.output_shape,
embeddings=_embeddings)
print(_keras_model.get_model_summary())
# Create framework object.
framework = framework_creator(
data_container=_container,
model=_keras_model,
test_features=_test_features,
test_labels=_test_labels,
stats_path='%s/al_stats_temp.csv' % cur_work_space,
max_round=flags.max_round,
num_select_per_round=flags.num_select_per_round,
pre_train=flags.pre_train,
