Minor changes.

Author: otiliastr

neural_structured_learning/research/gam/experiments/run_train_mnist.py

@@ -138,7 +138,7 @@
     'max_num_iter_agr', 100000,
     'Maximum number of iterations to train the agreement model for.')
 flags.DEFINE_integer(
-    'num_iter_after_best_val_agr', 5000,
+    'num_iter_after_best_val_agr', 2000,
     'Minimum number of iterations to train the agreement model for after '
     'the best validation accuracy is improved.')
 flags.DEFINE_integer(
@@ -194,7 +194,7 @@
 flags.DEFINE_integer(
     'batch_size_cls', 512, 'Batch size for classification model.')
 flags.DEFINE_float(
-    'gradient_clip', 10,
+    'gradient_clip', None,
     'The gradient clipping global norm value. If None, no clipping is done.')
 flags.DEFINE_integer(
     'logging_step_cls', 200,
@@ -252,7 +252,7 @@
     'penalize_neg_agr', True,
     'Whether to encourage differences when agreement is negative.')
 flags.DEFINE_bool(
-    'use_l2_cls', True,
+    'use_l2_cls', False,
     'Whether to use L2 loss for the classifier, not cross entropy.')
 flags.DEFINE_bool(
     'first_iter_original', True,
@@ -424,10 +424,10 @@ def main(argv):
   # Put together parameters to create a model name.
   model_name = FLAGS.model_cls + (('_' + FLAGS.hidden_cls)
                                   if FLAGS.model_cls == 'mlp' else '')
-  model_name += '-' + FLAGS.model_agr + (('_' + FLAGS.hidden_agr)
-                                         if FLAGS.model_agr == 'mlp' else '')
-  model_name += ('-aggr_' + FLAGS.aggregation_agr_inputs + '_' +
-                 FLAGS.hidden_aggreg)
+  model_name += '-' + FLAGS.model_agr
+  model_name += ('_' + FLAGS.hidden_agr) if FLAGS.model_agr == 'mlp' else ''
+  model_name += '-aggr_' + FLAGS.aggregation_agr_inputs
+  model_name += ('_' + FLAGS.hidden_aggreg) if FLAGS.hidden_aggreg else ''
   model_name += ('-add_%d-conf_%.2f-iter_cls_%d-iter_agr_%d-batch_cls_%d' %
                  (FLAGS.num_samples_to_label, FLAGS.min_confidence_new_label,
                   FLAGS.max_num_iter_cls, FLAGS.max_num_iter_agr,
@@ -436,7 +436,7 @@ def main(argv):
   model_name += '-perfectCls' if FLAGS.use_perfect_classifier else ''
   model_name += '-keepProp' if FLAGS.keep_label_proportions else ''
   model_name += '-PenNegAgr' if FLAGS.penalize_neg_agr else ''
-  model_name += '-inductive' if FLAGS.inductive else ''
+  model_name += '-transduct' if not FLAGS.inductive else ''
   model_name += '-L2Loss' if FLAGS.use_l2_cls else '-CELoss'
   model_name += '-seed_' + str(FLAGS.seed)
   model_name += FLAGS.experiment_suffix

neural_structured_learning/research/gam/models/cnn.py

@@ -174,7 +174,7 @@ def _construct_layers(self, inputs):
 
     return local4, reg_params
 
-  def get_encoding_and_params(self, inputs, is_train, **kwargs):
+  def get_encoding_and_params(self, inputs, **unused_kwargs):
     """Creates the model hidden representations and prediction ops.
 
     For this model, the hidden representation is the last layer
@@ -183,9 +183,7 @@ def get_encoding_and_params(self, inputs, is_train, **kwargs):
     Args:
       inputs: A tensor containing the model inputs. The first dimension is the
         batch size.
-      is_train: A placeholder representing a boolean value that specifies if
-        this model will be used for training or for test.
-      **kwargs: Other keyword arguments.
+      **unused_kwargs: Other unused keyword arguments.
 
     Returns:
       encoding: A tensor containing an encoded batch of samples. The first
@@ -256,12 +254,10 @@ def get_predictions_and_params(self, encoding, is_train, **kwargs):
         shape=(input_size, self.output_dim),
         initializer=tf.truncated_normal_initializer(
             stddev=1.0/float(input_size),
-            dtype=tf.float32),
-        use_resource=True)
+            dtype=tf.float32))
       biases = tf.get_variable(
         'b_outputs',
-        initializer=tf.zeros([self.output_dim], dtype=tf.float32),
-        use_resource=True)
+        initializer=tf.zeros([self.output_dim], dtype=tf.float32))
       predictions = tf.add(tf.matmul(encoding, weights), biases,
                            name='predictions')
 

neural_structured_learning/research/gam/models/mlp.py

@@ -92,7 +92,7 @@ def _construct_layers(self, inputs):
       hidden = self.activation(tf.nn.xw_plus_b(hidden, weights, biases))
     return hidden, reg_params
 
-  def get_encoding_and_params(self, inputs, is_train, **kwargs):
+  def get_encoding_and_params(self, inputs, **unused_kwargs):
     """Creates the model hidden representations and prediction ops.
 
     For this model, the hidden representation is the last layer of the MLP,
@@ -101,9 +101,7 @@ def get_encoding_and_params(self, inputs, is_train, **kwargs):
     Args:
       inputs: A tensor containing the model inputs. The first dimension is the
         batch size.
-      is_train: A placeholder representing a boolean value that specifies if
-        this model will be used for training or for test.
-      **kwargs: Other keyword arguments.
+      **unused_kwargs: Other unused keyword arguments.
 
     Returns:
       encoding: A tensor containing an encoded batch of samples. The first

neural_structured_learning/research/gam/trainer/trainer_agreement.py

@@ -754,10 +754,6 @@ def predict(self, session, src_features, tgt_features, **unused_kwargs):
         features of the first element of the pair.
       tgt_features: An array of shape (num_samples, num_features) containing the
         features of the second element of the pair.
-      src_indices: An array of integers containing the index of each sample in
-        self.data of the samples in src_features.
-      tgt_indices: An array of integers containing the index of each sample in
-        self.data of the samples in tgt_features.
 
     Returns:
       An array containing the predicted agreement value for each pair of