def casas_test(x, y, num_classes, model):
"""
Test Trained Logistic Regression Model
:param x: numpy.array training data
:type y: numpy.array
:param y: numpy.array training labels
:param num_classes: integer number of enabled classes
:param model: an model to be trained (in this case: Logistic Regression Object)
:return: numpy.array (confusion matrix)
"""
assert (type(model) == StackedDenoisingAutoencoder)
x_tensor = theano.shared(np.asarray(x, dtype=theano.config.floatX),
borrow=True)
result = model.classify(x_tensor)
predicted_y = result[0]
confusion_matrix = get_confusion_matrix(num_classes=num_classes,
label=y,
predicted=predicted_y)
return confusion_matrix
model.do_pretraining(
data=x_tensor[train_index],
num_data=len(train_index),
batch_size=10,
learning_rate_array=pre_training_learning_rate,
num_epochs=pre_training_epochs)
model.do_fine_tuning(
data=x_tensor[train_index],
label=y_tensor[train_index],
num_data=len(train_index),
batch_size=10,
learning_rate_array=fine_tuning_training_rate,
num_epochs=fine_tuning_epochs)
(predicted_y, ) = model.classify(x_tensor[test_index])
confusion_matrix = get_confusion_matrix(
num_classes=feature.num_enabled_activities,
label=feature.y[test_index],
predicted=predicted_y)
(overall_performance, per_class_performance) = \
get_performance_array(num_classes=feature.num_enabled_activities,
confusion_matrix=confusion_matrix)
learning_result.add_record(
model.export_to_dict(),
key=record_key,
confusion_matrix=confusion_matrix,
overall_performance=overall_performance,
per_class_performance=per_class_performance)
else:
model.load_from_dict(record['model'])
confusion_matrix = record['confusion_matrix']
overall_performance = record['overall_performance']
per_class_performance = record['per_class_performance']
learning_rate_array=[
0.005, 0.005, 0.005, 0.005, 0.005, 0.005,
0.005, 0.005
],
num_epochs=36)
# model.do_log_layer_training_only(data=train_set_x_tensor, label=train_set_y_tensor,
# num_data=train_set_x.shape[0], batch_size=10,
# learning_rate_array=[0.005, 0.005])
model.save(model_file_name)
num_test = test_set_x.shape[0]
result = model.do_reconstruction(test_set_x_tensor)
reconstruction = result[0]
result = model.classify(test_set_x_tensor)
class_result = result[0]
confusion_matrix = get_confusion_matrix(10, test_set_y, class_result)
performance = get_performance_array(10, confusion_matrix)
sys.stdout.write('%22s\t' % ' ')
for performance_label in performance_index:
sys.stdout.write('%20s\t' % performance_label)
sys.stdout.write('\n')
num_performance = len(performance_index)
for i in range(10):
activity_label = ('%d' % i)
sys.stdout.write('%22s\t' % activity_label)
for j in range(num_performance):
sys.stdout.write('%20.5f\t' % (performance[i][j] * 100))
sys.stdout.write('\n')
check_array = np.zeros(num_test)
check_array[class_result == test_set_y] = 1