Reenable Precision and Recall with TF1

keras/metrics.py

@@ -1412,12 +1412,6 @@ class Precision(Metric):
             raise RuntimeError(
                 '`top_k` argument for `Precision` metric is currently supported '
                 'only with TensorFlow backend.')
-            import tensorflow as tf
-            if not tf.__version__.startswith('2.'):
-                raise RuntimeError(
-                    '`top_k` argument for `Precision` metric is currently '
-                    'supported only with TensorFlow 2.0. Your version '
-                    'of TensorFlow is ' + str(tf.__version__))
 
         self.top_k = top_k
         self.class_id = class_id
@@ -1528,12 +1522,6 @@ class Recall(Metric):
             raise RuntimeError(
                 '`top_k` argument for `Recall` metric is currently supported only '
                 'with TensorFlow backend.')
-            import tensorflow as tf
-            if not tf.__version__.startswith('2.'):
-                raise RuntimeError(
-                    '`top_k` argument for `Precision` metric is currently '
-                    'supported only with TensorFlow 2.0. Your version '
-                    'of TensorFlow is ' + str(tf.__version__))
 
         self.top_k = top_k
         self.class_id = class_id
@@ -1897,12 +1885,6 @@ class MeanIoU(BaseMeanIoU):
             raise RuntimeError(
                 '`MeanIoU` metric is currently supported only '
                 'with TensorFlow backend and TF version >= 2.0.0.')
-            import tensorflow as tf
-            if not tf.__version__.startswith('2.'):
-                raise RuntimeError(
-                    '`top_k` argument for `Precision` metric is currently '
-                    'supported only with TensorFlow 2.0. Your version '
-                    'of TensorFlow is ' + str(tf.__version__))
         super(MeanIoU, self).__init__(num_classes, name=name, dtype=dtype)
 
 

tests/keras/metrics_confusion_matrix_test.py

@@ -594,8 +594,6 @@ class TestAUC(object):
             metrics.AUC(summation_method='Invalid')
 
 
-@pytest.mark.skipif(not tf.__version__.startswith('2.'),
-                    reason='Requires TF 2')
 class TestPrecisionTest(object):
 
     def test_config(self):
@@ -745,8 +743,6 @@ class TestPrecisionTest(object):
         assert np.isclose(0, K.eval(p_obj.false_positives))
 
 
-@pytest.mark.skipif(not tf.__version__.startswith('2.'),
-                    reason='Requires TF 2')
 class TestRecall(object):
 
     def test_config(self):

tests/keras/metrics_training_test.py

@@ -77,6 +77,10 @@ def test_sensitivity_metrics():
 
 @pytest.mark.skipif(K.backend() != 'tensorflow', reason='requires tensorflow')
 def test_mean_iou():
+    import tensorflow as tf
+    if not tf.__version__.startswith('2.'):
+        return
+
     model = Sequential([Dense(1, input_shape=(3,))])
     model.compile('rmsprop', 'mse', metrics=[metrics.MeanIoU(2)])
     x = np.random.random((10, 3))