Merge pull request tensorflow#4700 from jhseu/branch_134847453

Branch 134847453

tensorflow/BUILD

@@ -54,10 +54,7 @@ config_setting(
 
 package_group(
     name = "internal",
-    packages = [
-        "//learning/vis/...",
-        "//tensorflow/...",
-    ],
+    packages = ["//tensorflow/..."],
 )
 
 sh_binary(

tensorflow/cc/saved_model/constants.h

@@ -28,11 +28,11 @@ constexpr char kSavedModelFilenamePbTxt[] = "saved_model.pbtxt";
 constexpr char kSavedModelVariablesDirectory[] = "variables";
 
 // SavedModel variables filename.
-constexpr char kSavedModelVariablesFilename[] = "saved_model_variables";
+constexpr char kSavedModelVariablesFilename[] = "variables";
 
 // SavedModel sharded variables filename.
 constexpr char kSavedModelVariablesShardedFilename[] =
-    "saved_model_variables-\?\?\?\?\?-of-\?\?\?\?\?";
+    "variables-\?\?\?\?\?-of-\?\?\?\?\?";
 
 // Commonly used tags.
 constexpr char kSavedModelTagServe[] = "serve";

tensorflow/cc/saved_model/testdata/half_plus_two/variables/checkpoint

@@ -1,2 +1,2 @@
-model_checkpoint_path: "/tmp/saved_model/half_plus_two/saved_model_variables"
-all_model_checkpoint_paths: "/tmp/saved_model/half_plus_two/saved_model_variables"
+model_checkpoint_path: "/tmp/saved_model/half_plus_two/variables/variables"
+all_model_checkpoint_paths: "/tmp/saved_model/half_plus_two/variables/variables"

tensorflow/cc/saved_model/testdata/half_plus_two_pbtxt/variables/checkpoint

@@ -0,0 +1,2 @@
+model_checkpoint_path: "/tmp/saved_model/half_plus_two_pbtxt/variables/variables-?????-of-00001"
+all_model_checkpoint_paths: "/tmp/saved_model/half_plus_two_pbtxt/variables/variables-?????-of-00001"

tensorflow/cc/saved_model/testdata/half_plus_two_sharded/variables/checkpoint

@@ -1,2 +1,2 @@
-model_checkpoint_path: "/tmp/saved_model/half_plus_two/variables/saved_model_variables-?????-of-00001"
-all_model_checkpoint_paths: "/tmp/saved_model/half_plus_two/variables/saved_model_variables-?????-of-00001"
+model_checkpoint_path: "/tmp/saved_model/half_plus_two/variables/variables-?????-of-00001"
+all_model_checkpoint_paths: "/tmp/saved_model/half_plus_two/variables/variables-?????-of-00001"

tensorflow/contrib/BUILD

@@ -38,6 +38,7 @@ py_library(
         "//tensorflow/contrib/tensor_forest/hybrid:ops_lib",
         "//tensorflow/contrib/tensorboard",
         "//tensorflow/contrib/testing:testing_py",
+        "//tensorflow/contrib/tfprof",
         "//tensorflow/contrib/training:training_py",
         "//tensorflow/contrib/util:util_py",
     ],

tensorflow/contrib/__init__.py

@@ -41,5 +41,6 @@
 from tensorflow.contrib import tensor_forest
 from tensorflow.contrib import tensorboard
 from tensorflow.contrib import testing
+from tensorflow.contrib import tfprof
 from tensorflow.contrib import training
 from tensorflow.contrib import util

tensorflow/contrib/learn/python/learn/estimators/classifier.py

@@ -62,7 +62,7 @@ class Classifier(estimator.Estimator):
   PROBABILITY_OUTPUT = 'probabilities'
 
   def __init__(self, model_fn, n_classes, model_dir=None, config=None,
-               params=None):
+               params=None, feature_engineering_fn=None):
     """Constructor for Classifier.
 
     Args:
@@ -73,16 +73,20 @@ def __init__(self, model_fn, n_classes, model_dir=None, config=None,
         continue training a previously saved model.
       config: Configuration object (optional)
       params: `dict` of hyper parameters that will be passed into `model_fn`.
+      feature_engineering_fn: Feature engineering function. Takes features and
+                        targets which are the output of `input_fn` and
+                        returns features and targets which will be fed
+                        into the model.
     """
     self._n_classes = n_classes
     self._logits_fn = model_fn
     if params:
       model_fn = self._classifier_model_with_params
     else:
       model_fn = self._classifier_model
-    super(Classifier, self).__init__(model_fn=model_fn,
-                                     model_dir=model_dir, config=config,
-                                     params=params)
+    super(Classifier, self).__init__(
+        model_fn=model_fn, model_dir=model_dir, config=config, params=params,
+        feature_engineering_fn=feature_engineering_fn)
 
   def evaluate(self,
                x=None,

tensorflow/contrib/learn/python/learn/estimators/dnn.py

@@ -348,7 +348,8 @@ def __init__(self,
                dropout=None,
                gradient_clip_norm=None,
                enable_centered_bias=None,
-               config=None):
+               config=None,
+               feature_engineering_fn=None):
     """Initializes a DNNClassifier instance.
 
     Args:
@@ -379,6 +380,10 @@ def __init__(self,
         bias variable for each class. Rest of the model structure learns the
         residual after centered bias.
       config: `RunConfig` object to configure the runtime settings.
+      feature_engineering_fn: Feature engineering function. Takes features and
+                        targets which are the output of `input_fn` and
+                        returns features and targets which will be fed
+                        into the model.
 
     Returns:
       A `DNNClassifier` estimator.
@@ -415,7 +420,8 @@ def __init__(self,
             "gradient_clip_norm": gradient_clip_norm,
             "enable_centered_bias": enable_centered_bias,
             "num_ps_replicas": num_ps_replicas,
-        })
+        },
+        feature_engineering_fn=feature_engineering_fn)
 
   def fit(self, x=None, y=None, input_fn=None, steps=None, batch_size=None,
           monitors=None, max_steps=None):
@@ -561,8 +567,10 @@ def model_dir(self):
     return self._model_dir
 
   @property
-  @deprecated("2016-10-13", "This method inspects the private state of the "
-              "object, and should not be used")
+  @deprecated("2016-10-30",
+              "This method will be removed after the deprecation date. "
+              "To inspect variables, use get_variable_names() and "
+              "get_variable_value().")
   def weights_(self):
     hiddenlayer_weights = [checkpoints.load_variable(
         self._model_dir, name=("dnn/hiddenlayer_%d/weights" % i))
@@ -572,8 +580,10 @@ def weights_(self):
     return hiddenlayer_weights + logits_weights
 
   @property
-  @deprecated("2016-10-13", "This method inspects the private state of the "
-              "object, and should not be used")
+  @deprecated("2016-10-30",
+              "This method will be removed after the deprecation date. "
+              "To inspect variables, use get_variable_names() and "
+              "get_variable_value().")
   def bias_(self):
     hiddenlayer_bias = [checkpoints.load_variable(
         self._model_dir, name=("dnn/hiddenlayer_%d/biases" % i))
@@ -655,7 +665,8 @@ def __init__(self,
                dropout=None,
                gradient_clip_norm=None,
                enable_centered_bias=None,
-               config=None):
+               config=None,
+               feature_engineering_fn=None):
     """Initializes a `DNNRegressor` instance.
 
     Args:
@@ -684,6 +695,10 @@ def __init__(self,
         bias variable for each class. Rest of the model structure learns the
         residual after centered bias.
       config: `RunConfig` object to configure the runtime settings.
+      feature_engineering_fn: Feature engineering function. Takes features and
+                        targets which are the output of `input_fn` and
+                        returns features and targets which will be fed
+                        into the model.
 
     Returns:
       A `DNNRegressor` estimator.
@@ -701,7 +716,8 @@ def __init__(self,
         dnn_dropout=dropout,
         gradient_clip_norm=gradient_clip_norm,
         enable_centered_bias=enable_centered_bias,
-        config=config)
+        config=config,
+        feature_engineering_fn=feature_engineering_fn)
     self.feature_columns = feature_columns
     self.optimizer = optimizer
     self.activation_fn = activation_fn
@@ -710,9 +726,17 @@ def __init__(self,
     self._feature_columns_inferred = False
 
   @property
+  @deprecated("2016-10-30",
+              "This method will be removed after the deprecation date. "
+              "To inspect variables, use get_variable_names() and "
+              "get_variable_value().")
   def weights_(self):
     return self.dnn_weights_
 
   @property
+  @deprecated("2016-10-30",
+              "This method will be removed after the deprecation date. "
+              "To inspect variables, use get_variable_names() and "
+              "get_variable_value().")
   def bias_(self):
     return self.dnn_bias_

tensorflow/contrib/learn/python/learn/estimators/dnn_linear_combined.py

@@ -79,7 +79,8 @@ def __init__(self,  # _joint_linear_weights pylint: disable=invalid-name
                dnn_dropout=None,
                gradient_clip_norm=None,
                enable_centered_bias=True,
-               config=None):
+               config=None,
+               feature_engineering_fn=None):
     """Initializes a _DNNLinearCombinedBaseEstimator instance.
 
     Args:
@@ -114,6 +115,10 @@ def __init__(self,  # _joint_linear_weights pylint: disable=invalid-name
         bias variable for each class. Rest of the model structure learns the
         residual after centered bias.
       config: RunConfig object to configure the runtime settings.
+      feature_engineering_fn: Feature engineering function. Takes features and
+                        targets which are the output of `input_fn` and
+                        returns features and targets which will be fed
+                        into the model.
 
     Raises:
       ValueError: If both linear_feature_columns and dnn_features_columns are
@@ -147,24 +152,43 @@ def __init__(self,  # _joint_linear_weights pylint: disable=invalid-name
     self._centered_bias_weight_collection = "centered_bias"
     self._enable_centered_bias = enable_centered_bias
     self._target_column = target_column
+    self._feature_engineering_fn = (
+        feature_engineering_fn or
+        (lambda features, targets: (features, targets)))
 
   @property
+  @deprecated("2016-10-30",
+              "This method will be removed after the deprecation date. "
+              "To inspect variables, use get_variable_names() and "
+              "get_variable_value().")
   def linear_weights_(self):
     """Returns weights per feature of the linear part."""
     return self._linear_model.get_weights(model_dir=self._model_dir)
 
   @property
+  @deprecated("2016-10-30",
+              "This method will be removed after the deprecation date. "
+              "To inspect variables, use get_variable_names() and "
+              "get_variable_value().")
   def linear_bias_(self):
     """Returns bias of the linear part."""
     return (self._linear_model.get_bias(model_dir=self._model_dir) +
             self.get_variable_value("centered_bias_weight"))
 
   @property
+  @deprecated("2016-10-30",
+              "This method will be removed after the deprecation date. "
+              "To inspect variables, use get_variable_names() and "
+              "get_variable_value().")
   def dnn_weights_(self):
     """Returns weights of deep neural network part."""
     return self._dnn_model.get_weights(model_dir=self._model_dir)
 
   @property
+  @deprecated("2016-10-30",
+              "This method will be removed after the deprecation date. "
+              "To inspect variables, use get_variable_names() and "
+              "get_variable_value().")
   def dnn_bias_(self):
     """Returns bias of deep neural network part."""
     return (self._dnn_model.get_bias(model_dir=self._model_dir) +
@@ -185,6 +209,7 @@ def _get_train_ops(self, features, targets):
     assert global_step
 
     features = self._get_feature_dict(features)
+    features, targets = self._feature_engineering_fn(features, targets)
     logits = self._logits(features, is_training=True)
     if self._enable_centered_bias:
       centered_bias_step = [self._centered_bias_step(targets, features)]
@@ -209,12 +234,14 @@ def _get_train_ops(self, features, targets):
   def _get_eval_ops(self, features, targets, metrics=None):
     """See base class."""
     features = self._get_feature_dict(features)
+    features, targets = self._feature_engineering_fn(features, targets)
     logits = self._logits(features)
     return self._target_column.get_eval_ops(features, logits, targets, metrics)
 
   def _get_predict_ops(self, features):
     """See base class."""
     features = self._get_feature_dict(features)
+    features, _ = self._feature_engineering_fn(features, None)
     logits = self._logits(features)
     return self._target_column.logits_to_predictions(logits, proba=True)
 
@@ -366,7 +393,8 @@ def __init__(self,  # _joint_linear_weights pylint: disable=invalid-name
                dnn_dropout=None,
                gradient_clip_norm=None,
                enable_centered_bias=None,
-               config=None):
+               config=None,
+               feature_engineering_fn=None):
     """Constructs a DNNLinearCombinedClassifier instance.
 
     Args:
@@ -403,6 +431,10 @@ def __init__(self,  # _joint_linear_weights pylint: disable=invalid-name
         bias variable for each class. Rest of the model structure learns the
         residual after centered bias.
       config: RunConfig object to configure the runtime settings.
+      feature_engineering_fn: Feature engineering function. Takes features and
+                        targets which are the output of `input_fn` and
+                        returns features and targets which will be fed
+                        into the model.
 
     Raises:
       ValueError: If `n_classes` < 2.
@@ -433,7 +465,8 @@ def __init__(self,  # _joint_linear_weights pylint: disable=invalid-name
         gradient_clip_norm=gradient_clip_norm,
         enable_centered_bias=enable_centered_bias,
         target_column=target_column,
-        config=config)
+        config=config,
+        feature_engineering_fn=feature_engineering_fn)
 
   @deprecated_arg_values(
       estimator.AS_ITERABLE_DATE, estimator.AS_ITERABLE_INSTRUCTIONS,
@@ -559,7 +592,8 @@ def __init__(self,  # _joint_linear_weights pylint: disable=invalid-name
                gradient_clip_norm=None,
                enable_centered_bias=None,
                target_dimension=1,
-               config=None):
+               config=None,
+               feature_engineering_fn=None):
     """Initializes a DNNLinearCombinedRegressor instance.
 
     Args:
@@ -596,6 +630,10 @@ def __init__(self,  # _joint_linear_weights pylint: disable=invalid-name
         residual after centered bias.
       target_dimension: TODO(zakaria): dimension of the target for multilabels.
       config: RunConfig object to configure the runtime settings.
+      feature_engineering_fn: Feature engineering function. Takes features and
+                        targets which are the output of `input_fn` and
+                        returns features and targets which will be fed
+                        into the model.
 
     Raises:
       ValueError: If both linear_feature_columns and dnn_features_columns are
@@ -620,4 +658,5 @@ def __init__(self,  # _joint_linear_weights pylint: disable=invalid-name
         gradient_clip_norm=gradient_clip_norm,
         enable_centered_bias=enable_centered_bias,
         target_column=target_column,
-        config=config)
+        config=config,
+        feature_engineering_fn=feature_engineering_fn)

tensorflow/contrib/learn/python/learn/estimators/dnn_linear_combined_test.py

@@ -1003,5 +1003,45 @@ def _input_fn(num_epochs=None):
     self.assertLess(scores['loss'], 0.2)
 
 
+class FeatureEngineeringFunctionTest(tf.test.TestCase):
+  """Tests feature_engineering_fn."""
+
+  def testNoneFeatureEngineeringFn(self):
+    def input_fn():
+      # Create 4 rows of (y = x)
+      target = tf.constant([[100.], [3.], [2.], [2.]])
+      features = {'x': tf.constant([[100.], [3.], [2.], [2.]])}
+      return features, target
+
+    def feature_engineering_fn(features, targets):
+      _, _ = features, targets
+      target = tf.constant([[1000.], [30.], [20.], [20.]])
+      features = {'x': tf.constant([[1000.], [30.], [20.], [20.]])}
+      return features, target
+
+    estimator_with_fe_fn = tf.contrib.learn.DNNLinearCombinedRegressor(
+        linear_feature_columns=[tf.contrib.layers.real_valued_column('x')],
+        dnn_feature_columns=[tf.contrib.layers.real_valued_column('x')],
+        dnn_hidden_units=[3, 3],
+        config=tf.contrib.learn.RunConfig(tf_random_seed=1),
+        feature_engineering_fn=feature_engineering_fn)
+    estimator_with_fe_fn.fit(input_fn=input_fn, steps=100)
+
+    estimator_without_fe_fn = tf.contrib.learn.DNNLinearCombinedRegressor(
+        linear_feature_columns=[tf.contrib.layers.real_valued_column('x')],
+        dnn_feature_columns=[tf.contrib.layers.real_valued_column('x')],
+        dnn_hidden_units=[3, 3],
+        config=tf.contrib.learn.RunConfig(tf_random_seed=1))
+    estimator_without_fe_fn.fit(input_fn=input_fn, steps=100)
+
+     # predictions = y
+    prediction_with_fe_fn = next(
+        estimator_with_fe_fn.predict(input_fn=input_fn, as_iterable=True))
+    self.assertAlmostEqual(1000., prediction_with_fe_fn, delta=1.0)
+    prediction_without_fe_fn = next(
+        estimator_without_fe_fn.predict(input_fn=input_fn, as_iterable=True))
+    self.assertAlmostEqual(100., prediction_without_fe_fn, delta=1.0)
+
+
 if __name__ == '__main__':
   tf.test.main()