def test_binary_classification(self):
train_features = np.array([[0.0286274, 0.41107054, 0.30557073],
[0.18646135, 0.71026038, 0.87030804],
[0.46904668, 0.96190886, 0.85772885],
[0.40327128, 0.5739354, 0.21895921],
[0.53548, 0.9645708, 0.56493308],
[0.80917639, 0.78891976, 0.96257564],
[0.10951679, 0.75733494, 0.10935291]])
train_labels = np.array([0, 0, 0, 1, 0, 0, 1])
gp_repurposer = GpRepurposer(self.source_model,
self.source_model_layers)
gp_repurposer.target_model = gp_repurposer._train_model_from_features(
train_features, train_labels, {'l1': np.arange(0, 3)})
self.assertTrue(
len(gp_repurposer.target_model) == 1,
"Number of GP models expected: 1. Got: {}".format(
len(gp_repurposer.target_model)))
# Validate predicted probabilities
test_features = np.array([[0.63747595, 0.86516482, 0.21255967],
[0.33403457, 0.43162212, 0.77119909],
[0.1678248, 0.41870605, 0.37232554]])
test_labels = np.array([1, 0, 0])
expected_probabilities = np.array([[0.48597323, 0.51402677],
[0.67488224, 0.32511776],
[0.55386502, 0.44613498]])
predicted_probabilities = gp_repurposer._predict_probability_from_features(
test_features)
self.assertTrue(
np.allclose(predicted_probabilities, expected_probabilities))
# Validate predicted labels
predicted_labels = gp_repurposer._predict_label_from_features(
test_features)
self.assertTrue(np.array_equal(predicted_labels, test_labels))
def test_predict_label_from_features(self):
gp_repurposer = GpRepurposer(self.source_model, self.source_model_layers, apply_l2_norm=True)
gp_repurposer.target_model = gp_repurposer._train_model_from_features(
self.train_features[:self.num_data_points_to_train],
self.train_labels[:self.num_data_points_to_train],
{'l1': self.train_feature_indices})
predicted_labels = gp_repurposer._predict_label_from_features(self.test_features
[:self.num_data_points_to_predict])
self._validate_prediction_results(predicted_labels, test_predict_probability=False,
expected_accuracy=self.expected_accuracy_from_features,
num_predictions=self.num_data_points_to_predict)
