tensorflow_estimator/python/estimator/canned/linear_testing_utils.py [1321:1429]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - self._linear_classifier_fn = linear_classifier_fn self._fc_lib = fc_lib def setUp(self): self._model_dir = tempfile.mkdtemp() def tearDown(self): if self._model_dir: shutil.rmtree(self._model_dir) def _test_evaluation_for_simple_data(self, n_classes): label = 1 age = 1. # For binary case, the expected weight has shape (1,1). For multi class # case, the shape is (1, n_classes). In order to test the weights, set # weights as 2.0 * range(n_classes). age_weight = [[-11.0]] if n_classes == 2 else (np.reshape( -11.0 * np.array(list(range(n_classes)), dtype=np.float32), (1, n_classes))) bias = [-30.0] if n_classes == 2 else [-30.0] * n_classes with tf.Graph().as_default(): tf.Variable(age_weight, name=AGE_WEIGHT_NAME) tf.Variable(bias, name=BIAS_NAME) tf.Variable( 100, name=tf.compat.v1.GraphKeys.GLOBAL_STEP, dtype=tf.dtypes.int64) save_variables_to_ckpt(self._model_dir) est = self._linear_classifier_fn( feature_columns=(self._fc_lib.numeric_column('age'),), n_classes=n_classes, model_dir=self._model_dir) eval_metrics = est.evaluate( input_fn=lambda: ({ 'age': ((age,),) }, ((label,),)), steps=1) if n_classes == 2: # Binary classes: loss = sum(corss_entropy(41)) = 41. expected_metrics = { metric_keys.MetricKeys.LOSS: 41., tf.compat.v1.GraphKeys.GLOBAL_STEP: 100, metric_keys.MetricKeys.LOSS_MEAN: 41., metric_keys.MetricKeys.ACCURACY: 0., metric_keys.MetricKeys.PRECISION: 0., metric_keys.MetricKeys.RECALL: 0., metric_keys.MetricKeys.PREDICTION_MEAN: 0., metric_keys.MetricKeys.LABEL_MEAN: 1., metric_keys.MetricKeys.ACCURACY_BASELINE: 1, metric_keys.MetricKeys.AUC: 0., metric_keys.MetricKeys.AUC_PR: 1., } else: # Multi classes: loss = 1 * -log ( soft_max(logits)[label] ) logits = age_weight * age + bias logits_exp = np.exp(logits) softmax = logits_exp / logits_exp.sum() expected_loss = -1 * math.log(softmax[0, label]) expected_metrics = { metric_keys.MetricKeys.LOSS: expected_loss, metric_keys.MetricKeys.LOSS_MEAN: expected_loss, tf.compat.v1.GraphKeys.GLOBAL_STEP: 100, metric_keys.MetricKeys.ACCURACY: 0., } self.assertAllClose( sorted_key_dict(expected_metrics), sorted_key_dict(eval_metrics), rtol=1e-3) def test_binary_classes_evaluation_for_simple_data(self): self._test_evaluation_for_simple_data(n_classes=2) def test_multi_classes_evaluation_for_simple_data(self): self._test_evaluation_for_simple_data(n_classes=4) def _test_evaluation_batch(self, n_classes): """Tests evaluation for batch_size==2.""" label = [1, 0] age = [17., 18.] # For binary case, the expected weight has shape (1,1). For multi class # case, the shape is (1, n_classes). In order to test the weights, set # weights as 2.0 * range(n_classes). age_weight = [[2.0]] if n_classes == 2 else (np.reshape( 2.0 * np.array(list(range(n_classes)), dtype=np.float32), (1, n_classes))) bias = [-35.0] if n_classes == 2 else [-35.0] * n_classes initial_global_step = 100 with tf.Graph().as_default(): tf.Variable(age_weight, name=AGE_WEIGHT_NAME) tf.Variable(bias, name=BIAS_NAME) tf.Variable( initial_global_step, name=tf.compat.v1.GraphKeys.GLOBAL_STEP, dtype=tf.dtypes.int64) save_variables_to_ckpt(self._model_dir) est = self._linear_classifier_fn( feature_columns=(self._fc_lib.numeric_column('age'),), n_classes=n_classes, model_dir=self._model_dir) eval_metrics = est.evaluate( input_fn=lambda: ({ 'age': (age) }, (label)), steps=1) if n_classes == 2: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - tensorflow_estimator/python/estimator/canned/v1/linear_testing_utils_v1.py [1467:1575]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - self._linear_classifier_fn = linear_classifier_fn self._fc_lib = fc_lib def setUp(self): self._model_dir = tempfile.mkdtemp() def tearDown(self): if self._model_dir: shutil.rmtree(self._model_dir) def _test_evaluation_for_simple_data(self, n_classes): label = 1 age = 1. # For binary case, the expected weight has shape (1,1). For multi class # case, the shape is (1, n_classes). In order to test the weights, set # weights as 2.0 * range(n_classes). age_weight = [[-11.0]] if n_classes == 2 else (np.reshape( -11.0 * np.array(list(range(n_classes)), dtype=np.float32), (1, n_classes))) bias = [-30.0] if n_classes == 2 else [-30.0] * n_classes with tf.Graph().as_default(): tf.Variable(age_weight, name=AGE_WEIGHT_NAME) tf.Variable(bias, name=BIAS_NAME) tf.Variable( 100, name=tf.compat.v1.GraphKeys.GLOBAL_STEP, dtype=tf.dtypes.int64) save_variables_to_ckpt(self._model_dir) est = self._linear_classifier_fn( feature_columns=(self._fc_lib.numeric_column('age'),), n_classes=n_classes, model_dir=self._model_dir) eval_metrics = est.evaluate( input_fn=lambda: ({ 'age': ((age,),) }, ((label,),)), steps=1) if n_classes == 2: # Binary classes: loss = sum(corss_entropy(41)) = 41. expected_metrics = { metric_keys.MetricKeys.LOSS: 41., tf.compat.v1.GraphKeys.GLOBAL_STEP: 100, metric_keys.MetricKeys.LOSS_MEAN: 41., metric_keys.MetricKeys.ACCURACY: 0., metric_keys.MetricKeys.PRECISION: 0., metric_keys.MetricKeys.RECALL: 0., metric_keys.MetricKeys.PREDICTION_MEAN: 0., metric_keys.MetricKeys.LABEL_MEAN: 1., metric_keys.MetricKeys.ACCURACY_BASELINE: 1, metric_keys.MetricKeys.AUC: 0., metric_keys.MetricKeys.AUC_PR: 1., } else: # Multi classes: loss = 1 * -log ( soft_max(logits)[label] ) logits = age_weight * age + bias logits_exp = np.exp(logits) softmax = logits_exp / logits_exp.sum() expected_loss = -1 * math.log(softmax[0, label]) expected_metrics = { metric_keys.MetricKeys.LOSS: expected_loss, metric_keys.MetricKeys.LOSS_MEAN: expected_loss, tf.compat.v1.GraphKeys.GLOBAL_STEP: 100, metric_keys.MetricKeys.ACCURACY: 0., } self.assertAllClose( sorted_key_dict(expected_metrics), sorted_key_dict(eval_metrics), rtol=1e-3) def test_binary_classes_evaluation_for_simple_data(self): self._test_evaluation_for_simple_data(n_classes=2) def test_multi_classes_evaluation_for_simple_data(self): self._test_evaluation_for_simple_data(n_classes=4) def _test_evaluation_batch(self, n_classes): """Tests evaluation for batch_size==2.""" label = [1, 0] age = [17., 18.] # For binary case, the expected weight has shape (1,1). For multi class # case, the shape is (1, n_classes). In order to test the weights, set # weights as 2.0 * range(n_classes). age_weight = [[2.0]] if n_classes == 2 else (np.reshape( 2.0 * np.array(list(range(n_classes)), dtype=np.float32), (1, n_classes))) bias = [-35.0] if n_classes == 2 else [-35.0] * n_classes initial_global_step = 100 with tf.Graph().as_default(): tf.Variable(age_weight, name=AGE_WEIGHT_NAME) tf.Variable(bias, name=BIAS_NAME) tf.Variable( initial_global_step, name=tf.compat.v1.GraphKeys.GLOBAL_STEP, dtype=tf.dtypes.int64) save_variables_to_ckpt(self._model_dir) est = self._linear_classifier_fn( feature_columns=(self._fc_lib.numeric_column('age'),), n_classes=n_classes, model_dir=self._model_dir) eval_metrics = est.evaluate( input_fn=lambda: ({ 'age': (age) }, (label)), steps=1) if n_classes == 2: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -