Fixes for bare-column and column generators as inputs

1) Passing a single feature column (not wrapped in a list) should succeed, or fail clearly (In this change I wrap it in a list so it succeeds)

Currently, since concrete `_FeatureColumn` classes are iterable (inherit from named_tuple), the code attempts to interpret the _FeatureColumn object as a list of feature columns.

This can give confusing errors:

    fc.input_layer({'a':[1]},fc.numeric_column('a'))

    ValueError: Items of feature_columns must be a _FeatureColumn. Given (type <class 'str'>): a.

    fc.input_layer(
        {'a':[1]},
        fc.indicator_column(
            fc.categorical_column_with_identity('a', 100)))

    ValueError: Items of feature_columns must be a _DenseColumn. You can wrap a categorical column with an embedding_column or indicator_column. Given: _IdentityCategoricalColumn(key='a', num_buckets=100, default_value=None)

2) Passing an `iterator` should fail, or convert it to a list (so it can be iterated multiple times).

In this change I convert it to a list

Currently it throws an unclear error:

    features = {'a':[1],'b':[2]}
    columns = (fc.numeric_column(key) for key in features)
    fc.input_layer(features,columns)

    ValueError: List argument 'values' to 'ConcatV2' Op with length 0 shorter than minimum length 2.

PiperOrigin-RevId: 169314383

tensorflow/python/feature_column/feature_column.py

@@ -204,7 +204,7 @@ def input_layer(features,
   Raises:
     ValueError: if an item in `feature_columns` is not a `_DenseColumn`.
   """
-  _check_feature_columns(feature_columns)
+  feature_columns = _clean_feature_columns(feature_columns)
   for column in feature_columns:
     if not isinstance(column, _DenseColumn):
       raise ValueError(
@@ -294,7 +294,7 @@ def linear_model(features,
     ValueError: if an item in `feature_columns` is neither a `_DenseColumn`
       nor `_CategoricalColumn`.
   """
-  _check_feature_columns(feature_columns)
+  feature_columns = _clean_feature_columns(feature_columns)
   for column in feature_columns:
     if not isinstance(column, (_DenseColumn, _CategoricalColumn)):
       raise ValueError('Items of feature_columns must be either a _DenseColumn '
@@ -367,7 +367,7 @@ def _transform_features(features, feature_columns):
   Returns:
     A `dict` mapping `_FeatureColumn` to `Tensor` and `SparseTensor` values.
   """
-  _check_feature_columns(feature_columns)
+  feature_columns = _clean_feature_columns(feature_columns)
   outputs = {}
   with ops.name_scope(
       None, default_name='transform_features', values=features.values()):
@@ -1647,10 +1647,17 @@ def _to_sparse_input(input_tensor, ignore_value=None):
     return sparse_tensor_lib.SparseTensor(indices, values, dense_shape)
 
 
-def _check_feature_columns(feature_columns):
-  """Verifies feature_columns input."""
+def _clean_feature_columns(feature_columns):
+  """Verifies and normalizes `feature_columns` input."""
+  if isinstance(feature_columns, _FeatureColumn):
+    feature_columns = [feature_columns]
+
+  if isinstance(feature_columns, collections.Iterator):
+    feature_columns = list(feature_columns)
+
   if isinstance(feature_columns, dict):
     raise ValueError('Expected feature_columns to be iterable, found dict.')
+
   for column in feature_columns:
     if not isinstance(column, _FeatureColumn):
       raise ValueError('Items of feature_columns must be a _FeatureColumn. '
@@ -1668,6 +1675,8 @@ def _check_feature_columns(feature_columns):
                                                name_to_column[column.name]))
     name_to_column[column.name] = column
 
+  return feature_columns
+
 
 class _NumericColumn(_DenseColumn,
                      collections.namedtuple('_NumericColumn', [

tensorflow/python/feature_column/feature_column_test.py

@@ -1654,6 +1654,21 @@ def test_does_not_support_dict_columns(self):
       fc.input_layer(
           features={'a': [[0]]}, feature_columns={'a': fc.numeric_column('a')})
 
+  def test_bare_column(self):
+    with ops.Graph().as_default():
+      features = features = {'a': [0.]}
+      net = fc.input_layer(features, fc.numeric_column('a'))
+      with _initialized_session():
+        self.assertAllClose([[0.]], net.eval())
+
+  def test_column_generator(self):
+    with ops.Graph().as_default():
+      features = features = {'a': [0.], 'b': [1.]}
+      columns = (fc.numeric_column(key) for key in features)
+      net = fc.input_layer(features, columns)
+      with _initialized_session():
+        self.assertAllClose([[0., 1.]], net.eval())
+
   def test_raises_if_duplicate_name(self):
     with self.assertRaisesRegexp(
         ValueError, 'Duplicate feature column name found for columns'):