def model_signal_sensitivity(models, signals):
factor = -1 * np.log(0.01)
results = list()
signal_size = [int(signal['gaussian']['height']) for signal in signals]
for model in models:
model['p_values'] = list()
print(model['name'])
print('------------------------')
test_bgs = np.load(PATH_TEST_BACKGROUNDS).reshape(
model['shape']) / factor
predict_bgs = model['ae'].predict(test_bgs)
bgs_losses = loss_distribution(test_bgs, predict_bgs)
for signal in signals:
test_signal = np.load(get_path_signal(signal)).reshape(
model['shape']) / factor
predict_signal = model['ae'].predict(test_signal)
signal_losses = loss_distribution(test_signal, predict_signal)
p_value = model_efficiency_p_value(bgs_losses.numpy(),
signal_losses.numpy())
print(signal['id'], 'p_value: ', p_value)
model['p_values'].append(p_value)
results.append(model)
return signal_size, results
def create_loss_distribution(self, name, signal_id=1, file_name=''):
test_bgs_data, test_signal_data = self.load_test_data(signal_id=signal_id)
predict_bgs_test = self.predict(test_bgs_data)
predict_signal_test = self.predict(test_signal_data)
test_bgs_distribution = model_utils.loss_distribution(test_bgs_data,
predict_bgs_test)
test_signal_distribution = model_utils.loss_distribution(test_signal_data,
predict_signal_test)
model_utils.plot_histogram(test_bgs_distribution.numpy(), test_signal_distribution.numpy(), name, file_name=file_name)
def create_loss_distribution(self, name, m_5=6000, k=1000, file_name=''):
test_bgs_data, test_signal_data = self.load_test_data(m_5=m_5, k=k)
predict_bgs_test = self.predict(test_bgs_data)
predict_signal_test = self.predict(test_signal_data)
test_bgs_distribution = model_utils.loss_distribution(test_bgs_data,
predict_bgs_test.reshape(test_bgs_data.shape))
test_signal_distribution = model_utils.loss_distribution(test_signal_data,
predict_signal_test.reshape(test_signal_data.shape))
model_utils.plot_histogram(test_bgs_distribution.numpy(),
test_signal_distribution.numpy(),
name,
file_name=file_name)