During non-eager execution there was a bit of inconsistency with the way we

handled the placeholders where as the actually fed inputs when the inputs were
scalars. The standardize_single_array method expands dimensions so that (None,)
inputs look like (None, 1) instead but the placeholder shapes were the old
(None,).

PiperOrigin-RevId: 255033566

tensorflow/python/keras/engine/feature_columns_integration_test.py

@@ -94,6 +94,45 @@ class FeatureColumnsIntegrationTest(keras_parameterized.TestCase):
     model.evaluate(ds, steps=1)
     model.predict(ds, steps=1)
 
+  @keras_parameterized.run_all_keras_modes(always_skip_v1=True)
+  def test_sequential_model_with_crossed_column(self):
+    feature_columns = []
+    age_buckets = fc.bucketized_column(
+        fc.numeric_column('age'),
+        boundaries=[18, 25, 30, 35, 40, 45, 50, 55, 60, 65])
+    feature_columns.append(age_buckets)
+
+    # indicator cols
+    thal = fc.categorical_column_with_vocabulary_list(
+        'thal', ['fixed', 'normal', 'reversible'])
+
+    crossed_feature = fc.crossed_column([age_buckets, thal],
+                                        hash_bucket_size=1000)
+    crossed_feature = fc.indicator_column(crossed_feature)
+    feature_columns.append(crossed_feature)
+
+    feature_layer = fc.DenseFeatures(feature_columns)
+
+    model = keras.models.Sequential([
+        feature_layer,
+        keras.layers.Dense(128, activation='relu'),
+        keras.layers.Dense(128, activation='relu'),
+        keras.layers.Dense(1, activation='sigmoid')
+    ])
+
+    age_data = np.random.randint(10, 100, size=100)
+    thal_data = np.random.choice(['fixed', 'normal', 'reversible'], size=100)
+    inp_x = {'age': age_data, 'thal': thal_data}
+    inp_y = np.random.randint(0, 1, size=100)
+    ds = dataset_ops.Dataset.from_tensor_slices((inp_x, inp_y)).batch(5)
+    model.compile(optimizer='adam',
+                  loss='binary_crossentropy',
+                  metrics=['accuracy'],)
+    model.fit(ds, epochs=1)
+    model.fit(ds, epochs=1)
+    model.evaluate(ds)
+    model.predict(ds)
+
   @keras_parameterized.run_all_keras_modes
   def test_subclassed_model_with_feature_columns(self):
     col_a = fc.numeric_column('a')

tensorflow/python/keras/engine/training_test.py

@@ -250,7 +250,7 @@ class TrainingTest(keras_parameterized.TestCase):
     model = keras.Sequential([ReturnTraining()])
     model.compile('sgd', 'mse')
     hist = model.fit(x=np.array([0.]), y=np.array([0.]))
-    self.assertAllClose(hist.history['loss'][0], (10000,))
+    self.assertAllClose(hist.history['loss'][0], 10000)
 
   @keras_parameterized.run_with_all_model_types(exclude_models='sequential')
   @keras_parameterized.run_all_keras_modes

tensorflow/python/keras/engine/training_utils.py

@@ -1619,12 +1619,16 @@ class ModelInputs(object):
         # we have. The user should call `model._set_inputs(placeholders)`
         # to specify custom placeholders if the need arises.
         shape = (None,) + tuple(v.shape[1:])
+        if shape == (None,):
+          shape = (None, 1)
         dtype = dtypes.as_dtype(v.dtype)
         if dtype.is_floating:
           dtype = K.floatx()
         v = K.placeholder(shape=shape, name=k, dtype=dtype)
       elif isinstance(v, tensor_spec.TensorSpec):
         shape = (None,) + tuple(v.shape.as_list()[1:])
+        if shape == (None,):
+          shape = (None, 1)
         v = K.placeholder(shape=shape, name=k, dtype=v.dtype)
 
       self._flattened_inputs[i] = v