Graph regularization for document classification using natural graphs

View on TensorFlow.org Run in Google Colab View source on GitHub

Overview

Graph regularization is a specific technique under the broader paradigm of Neural Graph Learning (Bui et al., 2018). The core idea is to train neural network models with a graph-regularized objective, harnessing both labeled and unlabeled data.

In this tutorial, we will explore the use of graph regularization to classify documents that form a natural (organic) graph.

The general recipe for creating a graph-regularized model using the Neural Structured Learning (NSL) framework is as follows:

  1. Generate training data from the input graph and sample features. Nodes in the graph correspond to samples and edges in the graph correspond to similarity between pairs of samples. The resulting training data will contain neighbor features in addition to the original node features.
  2. Create a neural network as a base model using the Keras sequential, functional, or subclass API.
  3. Wrap the base model with the GraphRegularization wrapper class, which is provided by the NSL framework, to create a new graph Keras model. This new model will include a graph regularization loss as the regularization term in its training objective.
  4. Train and evaluate the graph Keras model.

Setup

Install the Neural Structured Learning package.

pip install --quiet neural-structured-learning

Dependencies and imports

import neural_structured_learning as nsl

import tensorflow as tf

# Resets notebook state
tf.keras.backend.clear_session()

print("Version: ", tf.__version__)
print("Eager mode: ", tf.executing_eagerly())
print(
    "GPU is",
    "available" if tf.config.list_physical_devices("GPU") else "NOT AVAILABLE")
Version:  2.2.0
Eager mode:  True
GPU is NOT AVAILABLE

Cora dataset

The Cora dataset is a citation graph where nodes represent machine learning papers and edges represent citations between pairs of papers. The task involved is document classification where the goal is to categorize each paper into one of 7 categories. In other words, this is a multi-class classification problem with 7 classes.

Graph

The original graph is directed. However, for the purpose of this example, we consider the undirected version of this graph. So, if paper A cites paper B, we also consider paper B to have cited A. Although this is not necessarily true, in this example, we consider citations as a proxy for similarity, which is usually a commutative property.

Features

Each paper in the input effectively contains 2 features:

  1. Words: A dense, multi-hot bag-of-words representation of the text in the paper. The vocabulary for the Cora dataset contains 1433 unique words. So, the length of this feature is 1433, and the value at position 'i' is 0/1 indicating whether word 'i' in the vocabulary exists in the given paper or not.

  2. Label: A single integer representing the class ID (category) of the paper.

Download the Cora dataset

wget --quiet -P /tmp https://linqs-data.soe.ucsc.edu/public/lbc/cora.tgz
tar -C /tmp -xvzf /tmp/cora.tgz
cora/
cora/README
cora/cora.content
cora/cora.cites

Convert the Cora data to the NSL format

In order to preprocess the Cora dataset and convert it to the format required by Neural Structured Learning, we will run the 'preprocess_cora_dataset.py' script, which is included in the NSL github repository. This script does the following:

  1. Generate neighbor features using the original node features and the graph.
  2. Generate train and test data splits containing tf.train.Example instances.
  3. Persist the resulting train and test data in the TFRecord format.
!wget https://raw.githubusercontent.com/tensorflow/neural-structured-learning/master/neural_structured_learning/examples/preprocess/cora/preprocess_cora_dataset.py

!python preprocess_cora_dataset.py \
--input_cora_content=/tmp/cora/cora.content \
--input_cora_graph=/tmp/cora/cora.cites \
--max_nbrs=5 \
--output_train_data=/tmp/cora/train_merged_examples.tfr \
--output_test_data=/tmp/cora/test_examples.tfr
--2020-05-08 04:32:07--  https://raw.githubusercontent.com/tensorflow/neural-structured-learning/master/neural_structured_learning/examples/preprocess/cora/preprocess_cora_dataset.py
Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 151.101.0.133, 151.101.64.133, 151.101.128.133, ...
Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|151.101.0.133|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 11641 (11K) [text/plain]
Saving to: ‘preprocess_cora_dataset.py’

preprocess_cora_dat 100%[===================>]  11.37K  --.-KB/s    in 0s      

2020-05-08 04:32:08 (131 MB/s) - ‘preprocess_cora_dataset.py’ saved [11641/11641]

Reading graph file: /tmp/cora/cora.cites...
Done reading 5429 edges from: /tmp/cora/cora.cites (0.01 seconds).
Making all edges bi-directional...
Done (0.01 seconds). Total graph nodes: 2708
Joining seed and neighbor tf.train.Examples with graph edges...
Done creating and writing 2155 merged tf.train.Examples (1.39 seconds).
Out-degree histogram: [(1, 386), (2, 468), (3, 452), (4, 309), (5, 540)]
Output training data written to TFRecord file: /tmp/cora/train_merged_examples.tfr.
Output test data written to TFRecord file: /tmp/cora/test_examples.tfr.
Total running time: 0.04 minutes.

Global variables

The file paths to the train and test data are based on the command line flag values used to invoke the 'preprocess_cora_dataset.py' script above.

### Experiment dataset
TRAIN_DATA_PATH = '/tmp/cora/train_merged_examples.tfr'
TEST_DATA_PATH = '/tmp/cora/test_examples.tfr'

### Constants used to identify neighbor features in the input.
NBR_FEATURE_PREFIX = 'NL_nbr_'
NBR_WEIGHT_SUFFIX = '_weight'

Hyperparameters

We will use an instance of HParams to include various hyperparameters and constants used for training and evaluation. We briefly describe each of them below:

  • num_classes: There are a total 7 different classes

  • max_seq_length: This is the size of the vocabulary and all instances in the input have a dense multi-hot, bag-of-words representation. In other words, a value of 1 for a word indicates that the word is present in the input and a value of 0 indicates that it is not.

  • distance_type: This is the distance metric used to regularize the sample with its neighbors.

  • graph_regularization_multiplier: This controls the relative weight of the graph regularization term in the overall loss function.

  • num_neighbors: The number of neighbors used for graph regularization. This value has to be less than or equal to the max_nbrs command-line argument used above when running preprocess_cora_dataset.py.

  • num_fc_units: The number of fully connected layers in our neural network.

  • train_epochs: The number of training epochs.

  • batch_size: Batch size used for training and evaluation.

  • dropout_rate: Controls the rate of dropout following each fully connected layer

  • eval_steps: The number of batches to process before deeming evaluation is complete. If set to None, all instances in the test set are evaluated.

class HParams(object):
  """Hyperparameters used for training."""
  def __init__(self):
    ### dataset parameters
    self.num_classes = 7
    self.max_seq_length = 1433
    ### neural graph learning parameters
    self.distance_type = nsl.configs.DistanceType.L2
    self.graph_regularization_multiplier = 0.1
    self.num_neighbors = 1
    ### model architecture
    self.num_fc_units = [50, 50]
    ### training parameters
    self.train_epochs = 100
    self.batch_size = 128
    self.dropout_rate = 0.5
    ### eval parameters
    self.eval_steps = None  # All instances in the test set are evaluated.

HPARAMS = HParams()

Load train and test data

As described earlier in this notebook, the input training and test data have been created by the 'preprocess_cora_dataset.py'. We will load them into two tf.data.Dataset objects -- one for train and one for test.

In the input layer of our model, we will extract not just the 'words' and the 'label' features from each sample, but also corresponding neighbor features based on the hparams.num_neighbors value. Instances with fewer neighbors than hparams.num_neighbors will be assigned dummy values for those non-existent neighbor features.

def parse_example(example_proto):
  """Extracts relevant fields from the `example_proto`.

  Args:
    example_proto: An instance of `tf.train.Example`.

  Returns:
    A pair whose first value is a dictionary containing relevant features
    and whose second value contains the ground truth label.
  """
  # The 'words' feature is a multi-hot, bag-of-words representation of the
  # original raw text. A default value is required for examples that don't
  # have the feature.
  feature_spec = {
      'words':
          tf.io.FixedLenFeature([HPARAMS.max_seq_length],
                                tf.int64,
                                default_value=tf.constant(
                                    0,
                                    dtype=tf.int64,
                                    shape=[HPARAMS.max_seq_length])),
      'label':
          tf.io.FixedLenFeature((), tf.int64, default_value=-1),
  }
  # We also extract corresponding neighbor features in a similar manner to
  # the features above.
  for i in range(HPARAMS.num_neighbors):
    nbr_feature_key = '{}{}_{}'.format(NBR_FEATURE_PREFIX, i, 'words')
    nbr_weight_key = '{}{}{}'.format(NBR_FEATURE_PREFIX, i, NBR_WEIGHT_SUFFIX)
    feature_spec[nbr_feature_key] = tf.io.FixedLenFeature(
        [HPARAMS.max_seq_length],
        tf.int64,
        default_value=tf.constant(
            0, dtype=tf.int64, shape=[HPARAMS.max_seq_length]))

    # We assign a default value of 0.0 for the neighbor weight so that
    # graph regularization is done on samples based on their exact number
    # of neighbors. In other words, non-existent neighbors are discounted.
    feature_spec[nbr_weight_key] = tf.io.FixedLenFeature(
        [1], tf.float32, default_value=tf.constant([0.0]))

  features = tf.io.parse_single_example(example_proto, feature_spec)
  label = features.pop('label')
  return features, label


def make_dataset(file_path, training=False):
  """Creates a `tf.data.TFRecordDataset`.

  Args:
    file_path: Name of the file in the `.tfrecord` format containing
      `tf.train.Example` objects.
    training: Boolean indicating if we are in training mode.

  Returns:
    An instance of `tf.data.TFRecordDataset` containing the `tf.train.Example`
    objects.
  """
  dataset = tf.data.TFRecordDataset([file_path])
  if training:
    dataset = dataset.shuffle(10000)
  dataset = dataset.map(parse_example)
  dataset = dataset.batch(HPARAMS.batch_size)
  return dataset


train_dataset = make_dataset(TRAIN_DATA_PATH, training=True)
test_dataset = make_dataset(TEST_DATA_PATH)

Let's peek into the train dataset to look at its contents.

for feature_batch, label_batch in train_dataset.take(1):
  print('Feature list:', list(feature_batch.keys()))
  print('Batch of inputs:', feature_batch['words'])
  nbr_feature_key = '{}{}_{}'.format(NBR_FEATURE_PREFIX, 0, 'words')
  nbr_weight_key = '{}{}{}'.format(NBR_FEATURE_PREFIX, 0, NBR_WEIGHT_SUFFIX)
  print('Batch of neighbor inputs:', feature_batch[nbr_feature_key])
  print('Batch of neighbor weights:',
        tf.reshape(feature_batch[nbr_weight_key], [-1]))
  print('Batch of labels:', label_batch)
Feature list: ['words', 'NL_nbr_0_words', 'NL_nbr_0_weight']
Batch of inputs: tf.Tensor(
[[0 0 0 ... 0 0 0]
 [0 0 1 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 ...
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]], shape=(128, 1433), dtype=int64)
Batch of neighbor inputs: tf.Tensor(
[[0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 ...
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]], shape=(128, 1433), dtype=int64)
Batch of neighbor weights: tf.Tensor(
[1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.

 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1.], shape=(128,), dtype=float32)
Batch of labels: tf.Tensor(
[2 2 1 0 1 4 4 3 6 2 3 4 1 0 2 0 6 1 3 6 6 0 1 2 3 6 5 0 0 2 2 1 3 2 1 0 6
 4 2 5 4 2 3 0 5 6 1 6 4 4 2 1 2 0 2 4 2 2 2 2 6 3 3 3 0 0 2 1 2 5 6 6 1 6
 3 2 2 3 0 2 0 6 5 2 3 1 0 1 1 6 3 6 0 1 2 5 0 6 1 2 3 4 5 3 3 0 3 2 6 2 2
 2 4 3 2 1 1 2 2 1 2 2 2 0 4 1 2 2], shape=(128,), dtype=int64)

Let's peek into the test dataset to look at its contents.

for feature_batch, label_batch in test_dataset.take(1):
  print('Feature list:', list(feature_batch.keys()))
  print('Batch of inputs:', feature_batch['words'])
  nbr_feature_key = '{}{}_{}'.format(NBR_FEATURE_PREFIX, 0, 'words')
  nbr_weight_key = '{}{}{}'.format(NBR_FEATURE_PREFIX, 0, NBR_WEIGHT_SUFFIX)
  print('Batch of neighbor inputs:', feature_batch[nbr_feature_key])
  print('Batch of neighbor weights:',
        tf.reshape(feature_batch[nbr_weight_key], [-1]))
  print('Batch of labels:', label_batch)
Feature list: ['words', 'NL_nbr_0_words', 'NL_nbr_0_weight']
Batch of inputs: tf.Tensor(
[[0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 ...
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]], shape=(128, 1433), dtype=int64)
Batch of neighbor inputs: tf.Tensor(
[[0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 ...
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]], shape=(128, 1433), dtype=int64)
Batch of neighbor weights: tf.Tensor(
[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.

 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0.], shape=(128,), dtype=float32)
Batch of labels: tf.Tensor(
[5 2 0 6 0 3 2 3 1 3 4 5 6 4 4 0 2 2 6 5 0 1 3 3 3 6 2 0 0 5 0 3 0 4 2 3 1
 2 3 6 4 2 6 0 2 1 6 2 6 3 5 2 2 3 0 1 1 3 1 1 2 4 2 6 3 0 6 2 2 5 0 1 3 6
 1 6 6 3 6 1 1 6 2 2 2 0 3 6 1 6 0 3 2 5 6 6 2 6 2 4 1 3 3 1 5 2 2 5 1 2 6
 2 1 3 1 3 2 4 2 0 3 2 1 2 2 1 6 1], shape=(128,), dtype=int64)

Model definition

In order to demonstrate the use of graph regularization, we build a base model for this problem first. We will use a simple feed-forward neural network with 2 hidden layers and dropout in between. We illustrate the creation of the base model using all model types supported by the tf.Keras framework -- sequential, functional, and subclass.

Sequential base model

def make_mlp_sequential_model(hparams):
  """Creates a sequential multi-layer perceptron model."""
  model = tf.keras.Sequential()
  model.add(
      tf.keras.layers.InputLayer(
          input_shape=(hparams.max_seq_length,), name='words'))
  # Input is already one-hot encoded in the integer format. We cast it to
  # floating point format here.
  model.add(
      tf.keras.layers.Lambda(lambda x: tf.keras.backend.cast(x, tf.float32)))
  for num_units in hparams.num_fc_units:
    model.add(tf.keras.layers.Dense(num_units, activation='relu'))
    # For sequential models, by default, Keras ensures that the 'dropout' layer
    # is invoked only during training.
    model.add(tf.keras.layers.Dropout(hparams.dropout_rate))
  model.add(tf.keras.layers.Dense(hparams.num_classes, activation='softmax'))
  return model

Functional base model

def make_mlp_functional_model(hparams):
  """Creates a functional API-based multi-layer perceptron model."""
  inputs = tf.keras.Input(
      shape=(hparams.max_seq_length,), dtype='int64', name='words')

  # Input is already one-hot encoded in the integer format. We cast it to
  # floating point format here.
  cur_layer = tf.keras.layers.Lambda(
      lambda x: tf.keras.backend.cast(x, tf.float32))(
          inputs)

  for num_units in hparams.num_fc_units:
    cur_layer = tf.keras.layers.Dense(num_units, activation='relu')(cur_layer)
    # For functional models, by default, Keras ensures that the 'dropout' layer
    # is invoked only during training.
    cur_layer = tf.keras.layers.Dropout(hparams.dropout_rate)(cur_layer)

  outputs = tf.keras.layers.Dense(
      hparams.num_classes, activation='softmax')(
          cur_layer)

  model = tf.keras.Model(inputs, outputs=outputs)
  return model

Subclass base model

def make_mlp_subclass_model(hparams):
  """Creates a multi-layer perceptron subclass model in Keras."""

  class MLP(tf.keras.Model):
    """Subclass model defining a multi-layer perceptron."""

    def __init__(self):
      super(MLP, self).__init__()
      # Input is already one-hot encoded in the integer format. We create a
      # layer to cast it to floating point format here.
      self.cast_to_float_layer = tf.keras.layers.Lambda(
          lambda x: tf.keras.backend.cast(x, tf.float32))
      self.dense_layers = [
          tf.keras.layers.Dense(num_units, activation='relu')
          for num_units in hparams.num_fc_units
      ]
      self.dropout_layer = tf.keras.layers.Dropout(hparams.dropout_rate)
      self.output_layer = tf.keras.layers.Dense(
          hparams.num_classes, activation='softmax')

    def call(self, inputs, training=False):
      cur_layer = self.cast_to_float_layer(inputs['words'])
      for dense_layer in self.dense_layers:
        cur_layer = dense_layer(cur_layer)
        cur_layer = self.dropout_layer(cur_layer, training=training)

      outputs = self.output_layer(cur_layer)

      return outputs

  return MLP()

Create base model(s)

# Create a base MLP model using the functional API.
# Alternatively, you can also create a sequential or subclass base model using
# the make_mlp_sequential_model() or make_mlp_subclass_model() functions
# respectively, defined above. Note that if a subclass model is used, its
# summary cannot be generated until it is built.
base_model_tag, base_model = 'FUNCTIONAL', make_mlp_functional_model(HPARAMS)
base_model.summary()
Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
words (InputLayer)           [(None, 1433)]            0         
_________________________________________________________________
lambda (Lambda)              (None, 1433)              0         
_________________________________________________________________
dense (Dense)                (None, 50)                71700     
_________________________________________________________________
dropout (Dropout)            (None, 50)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 50)                2550      
_________________________________________________________________
dropout_1 (Dropout)          (None, 50)                0         
_________________________________________________________________
dense_2 (Dense)              (None, 7)                 357       
=================================================================
Total params: 74,607
Trainable params: 74,607
Non-trainable params: 0
_________________________________________________________________

Train base MLP model

# Compile and train the base MLP model
base_model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',
    metrics=['accuracy'])
base_model.fit(train_dataset, epochs=HPARAMS.train_epochs, verbose=1)
Epoch 1/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.1503 - loss: 1.9464
Epoch 2/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.2719 - loss: 1.8673
Epoch 3/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.3183 - loss: 1.7866
Epoch 4/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.3527 - loss: 1.6846
Epoch 5/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.4107 - loss: 1.5674
Epoch 6/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.4747 - loss: 1.4340
Epoch 7/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.5527 - loss: 1.2856
Epoch 8/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.6260 - loss: 1.1231
Epoch 9/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.6589 - loss: 1.0077
Epoch 10/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.7095 - loss: 0.8896
Epoch 11/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.7462 - loss: 0.8060
Epoch 12/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.7680 - loss: 0.7238
Epoch 13/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8051 - loss: 0.6424
Epoch 14/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8111 - loss: 0.6268
Epoch 15/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.8306 - loss: 0.5431
Epoch 16/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8362 - loss: 0.5137
Epoch 17/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8742 - loss: 0.4398
Epoch 18/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8640 - loss: 0.4422
Epoch 19/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8910 - loss: 0.3881
Epoch 20/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8840 - loss: 0.3820
Epoch 21/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8914 - loss: 0.3524
Epoch 22/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9077 - loss: 0.3332
Epoch 23/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9118 - loss: 0.3046
Epoch 24/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9142 - loss: 0.2894
Epoch 25/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9211 - loss: 0.2594
Epoch 26/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9239 - loss: 0.2507
Epoch 27/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9336 - loss: 0.2318
Epoch 28/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9378 - loss: 0.2230
Epoch 29/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9341 - loss: 0.2277
Epoch 30/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9443 - loss: 0.2001
Epoch 31/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9434 - loss: 0.2014
Epoch 32/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9527 - loss: 0.1754
Epoch 33/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9485 - loss: 0.1712
Epoch 34/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9513 - loss: 0.1704
Epoch 35/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9517 - loss: 0.1784
Epoch 36/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9555 - loss: 0.1582
Epoch 37/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9550 - loss: 0.1608
Epoch 38/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9582 - loss: 0.1506
Epoch 39/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9647 - loss: 0.1380
Epoch 40/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9652 - loss: 0.1264
Epoch 41/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9629 - loss: 0.1370
Epoch 42/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9698 - loss: 0.1178
Epoch 43/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9657 - loss: 0.1267
Epoch 44/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9754 - loss: 0.1112
Epoch 45/100
17/17 [==============================] - 0s 15ms/step - accuracy: 0.9684 - loss: 0.1142
Epoch 46/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9657 - loss: 0.1224
Epoch 47/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9684 - loss: 0.1117
Epoch 48/100
17/17 [==============================] - 0s 15ms/step - accuracy: 0.9754 - loss: 0.1044
Epoch 49/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9745 - loss: 0.0965
Epoch 50/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9722 - loss: 0.0986
Epoch 51/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9745 - loss: 0.1020
Epoch 52/100
17/17 [==============================] - 0s 15ms/step - accuracy: 0.9763 - loss: 0.0932
Epoch 53/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9824 - loss: 0.0877
Epoch 54/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9773 - loss: 0.0883
Epoch 55/100
17/17 [==============================] - 0s 15ms/step - accuracy: 0.9800 - loss: 0.0872
Epoch 56/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9805 - loss: 0.0797
Epoch 57/100
17/17 [==============================] - 0s 17ms/step - accuracy: 0.9763 - loss: 0.0834
Epoch 58/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9782 - loss: 0.0875
Epoch 59/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9777 - loss: 0.0754
Epoch 60/100
17/17 [==============================] - 0s 16ms/step - accuracy: 0.9763 - loss: 0.0813
Epoch 61/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9824 - loss: 0.0696
Epoch 62/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9759 - loss: 0.0827
Epoch 63/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9810 - loss: 0.0721
Epoch 64/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9787 - loss: 0.0696
Epoch 65/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9814 - loss: 0.0702
Epoch 66/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9791 - loss: 0.0792
Epoch 67/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9814 - loss: 0.0633
Epoch 68/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9791 - loss: 0.0670
Epoch 69/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9796 - loss: 0.0667
Epoch 70/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9810 - loss: 0.0674
Epoch 71/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9847 - loss: 0.0570
Epoch 72/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9810 - loss: 0.0650
Epoch 73/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9824 - loss: 0.0656
Epoch 74/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9800 - loss: 0.0668
Epoch 75/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9847 - loss: 0.0598
Epoch 76/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9847 - loss: 0.0562
Epoch 77/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9852 - loss: 0.0552
Epoch 78/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9782 - loss: 0.0686
Epoch 79/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9819 - loss: 0.0568
Epoch 80/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9852 - loss: 0.0555
Epoch 81/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9824 - loss: 0.0603
Epoch 82/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9838 - loss: 0.0543
Epoch 83/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9865 - loss: 0.0539
Epoch 84/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9824 - loss: 0.0554
Epoch 85/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9847 - loss: 0.0486
Epoch 86/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9879 - loss: 0.0497
Epoch 87/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9870 - loss: 0.0516
Epoch 88/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9907 - loss: 0.0391
Epoch 89/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9861 - loss: 0.0472
Epoch 90/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9879 - loss: 0.0437
Epoch 91/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9884 - loss: 0.0496
Epoch 92/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9884 - loss: 0.0458
Epoch 93/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9856 - loss: 0.0457
Epoch 94/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9856 - loss: 0.0457
Epoch 95/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9893 - loss: 0.0473
Epoch 96/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9828 - loss: 0.0542
Epoch 97/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9819 - loss: 0.0512
Epoch 98/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9879 - loss: 0.0412
Epoch 99/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9903 - loss: 0.0382
Epoch 100/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9879 - loss: 0.0446

<tensorflow.python.keras.callbacks.History at 0x7f02a81b0940>

Evaluate base MLP model

# Helper function to print evaluation metrics.
def print_metrics(model_desc, eval_metrics):
  """Prints evaluation metrics.

  Args:
    model_desc: A description of the model.
    eval_metrics: A dictionary mapping metric names to corresponding values. It
      must contain the loss and accuracy metrics.
  """
  print('\n')
  print('Eval accuracy for ', model_desc, ': ', eval_metrics['accuracy'])
  print('Eval loss for ', model_desc, ': ', eval_metrics['loss'])
  if 'graph_loss' in eval_metrics:
    print('Eval graph loss for ', model_desc, ': ', eval_metrics['graph_loss'])
eval_results = dict(
    zip(base_model.metrics_names,
        base_model.evaluate(test_dataset, steps=HPARAMS.eval_steps)))
print_metrics('Base MLP model', eval_results)
5/5 [==============================] - 0s 7ms/step - accuracy: 0.7993 - loss: 1.2641


Eval accuracy for  Base MLP model :  0.7992766499519348
Eval loss for  Base MLP model :  1.2641242742538452

Train MLP model with graph regularization

Incorporating graph regularization into the loss term of an existing tf.Keras.Model requires just a few lines of code. The base model is wrapped to create a new tf.Keras subclass model, whose loss includes graph regularization.

To assess the incremental benefit of graph regularization, we will create a new base model instance. This is because base_model has already been trained for a few iterations, and reusing this trained model to create a graph-regularized model will not be a fair comparison for base_model.

# Build a new base MLP model.
base_reg_model_tag, base_reg_model = 'FUNCTIONAL', make_mlp_functional_model(
    HPARAMS)
# Wrap the base MLP model with graph regularization.
graph_reg_config = nsl.configs.make_graph_reg_config(
    max_neighbors=HPARAMS.num_neighbors,
    multiplier=HPARAMS.graph_regularization_multiplier,
    distance_type=HPARAMS.distance_type,
    sum_over_axis=-1)
graph_reg_model = nsl.keras.GraphRegularization(base_reg_model,
                                                graph_reg_config)
graph_reg_model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',
    metrics=['accuracy'])
graph_reg_model.fit(train_dataset, epochs=HPARAMS.train_epochs, verbose=1)
Epoch 1/100

/tmpfs/src/tf_docs_env/lib/python3.5/site-packages/tensorflow/python/framework/indexed_slices.py:434: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.
  "Converting sparse IndexedSlices to a dense Tensor of unknown shape. "

17/17 [==============================] - 0s 12ms/step - accuracy: 0.1782 - graph_loss: 0.0085 - loss: 1.9311
Epoch 2/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.2937 - graph_loss: 0.0114 - loss: 1.8450
Epoch 3/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.3285 - graph_loss: 0.0252 - loss: 1.7494
Epoch 4/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.3759 - graph_loss: 0.0436 - loss: 1.6522
Epoch 5/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.4501 - graph_loss: 0.0725 - loss: 1.5018
Epoch 6/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.5239 - graph_loss: 0.1129 - loss: 1.3702
Epoch 7/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.5907 - graph_loss: 0.1473 - loss: 1.2299
Epoch 8/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.6227 - graph_loss: 0.1880 - loss: 1.1188
Epoch 9/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.6770 - graph_loss: 0.2255 - loss: 0.9983
Epoch 10/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.7179 - graph_loss: 0.2383 - loss: 0.8992
Epoch 11/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.7378 - graph_loss: 0.2511 - loss: 0.8309
Epoch 12/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.7698 - graph_loss: 0.2753 - loss: 0.7562
Epoch 13/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8125 - graph_loss: 0.2816 - loss: 0.6599
Epoch 14/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.8181 - graph_loss: 0.2922 - loss: 0.6251
Epoch 15/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8510 - graph_loss: 0.3001 - loss: 0.5664
Epoch 16/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8510 - graph_loss: 0.3100 - loss: 0.5340
Epoch 17/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.8561 - graph_loss: 0.3106 - loss: 0.5018
Epoch 18/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.8840 - graph_loss: 0.3278 - loss: 0.4664
Epoch 19/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.8747 - graph_loss: 0.3241 - loss: 0.4350
Epoch 20/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.8877 - graph_loss: 0.3371 - loss: 0.4220
Epoch 21/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.8831 - graph_loss: 0.3250 - loss: 0.3960
Epoch 22/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9049 - graph_loss: 0.3311 - loss: 0.3583
Epoch 23/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9142 - graph_loss: 0.3266 - loss: 0.3405
Epoch 24/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9100 - graph_loss: 0.3387 - loss: 0.3384
Epoch 25/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9160 - graph_loss: 0.3296 - loss: 0.3212
Epoch 26/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9183 - graph_loss: 0.3396 - loss: 0.3033
Epoch 27/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9179 - graph_loss: 0.3475 - loss: 0.3040
Epoch 28/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9253 - graph_loss: 0.3360 - loss: 0.2897
Epoch 29/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9425 - graph_loss: 0.3272 - loss: 0.2565
Epoch 30/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9309 - graph_loss: 0.3435 - loss: 0.2599
Epoch 31/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9415 - graph_loss: 0.3488 - loss: 0.2485
Epoch 32/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9448 - graph_loss: 0.3417 - loss: 0.2366
Epoch 33/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9517 - graph_loss: 0.3295 - loss: 0.2225
Epoch 34/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9564 - graph_loss: 0.3320 - loss: 0.2057
Epoch 35/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9476 - graph_loss: 0.3488 - loss: 0.2230
Epoch 36/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9490 - graph_loss: 0.3258 - loss: 0.2093
Epoch 37/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9513 - graph_loss: 0.3373 - loss: 0.1933
Epoch 38/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9559 - graph_loss: 0.3419 - loss: 0.2003
Epoch 39/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9647 - graph_loss: 0.3314 - loss: 0.1779
Epoch 40/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9698 - graph_loss: 0.3425 - loss: 0.1674
Epoch 41/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9601 - graph_loss: 0.3417 - loss: 0.1835
Epoch 42/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9550 - graph_loss: 0.3483 - loss: 0.1750
Epoch 43/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9675 - graph_loss: 0.3378 - loss: 0.1626
Epoch 44/100
17/17 [==============================] - 0s 15ms/step - accuracy: 0.9624 - graph_loss: 0.3393 - loss: 0.1706
Epoch 45/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9735 - graph_loss: 0.3378 - loss: 0.1584
Epoch 46/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9661 - graph_loss: 0.3287 - loss: 0.1499
Epoch 47/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9689 - graph_loss: 0.3324 - loss: 0.1446
Epoch 48/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9726 - graph_loss: 0.3418 - loss: 0.1368
Epoch 49/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9694 - graph_loss: 0.3306 - loss: 0.1474
Epoch 50/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9671 - graph_loss: 0.3344 - loss: 0.1470
Epoch 51/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9703 - graph_loss: 0.3435 - loss: 0.1451
Epoch 52/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9800 - graph_loss: 0.3319 - loss: 0.1280
Epoch 53/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9684 - graph_loss: 0.3546 - loss: 0.1420
Epoch 54/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9777 - graph_loss: 0.3396 - loss: 0.1313
Epoch 55/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9777 - graph_loss: 0.3425 - loss: 0.1241
Epoch 56/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9726 - graph_loss: 0.3353 - loss: 0.1291
Epoch 57/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9694 - graph_loss: 0.3352 - loss: 0.1344
Epoch 58/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9759 - graph_loss: 0.3416 - loss: 0.1236
Epoch 59/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9689 - graph_loss: 0.3377 - loss: 0.1397
Epoch 60/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9787 - graph_loss: 0.3457 - loss: 0.1140
Epoch 61/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9763 - graph_loss: 0.3474 - loss: 0.1215
Epoch 62/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9791 - graph_loss: 0.3448 - loss: 0.1165
Epoch 63/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9824 - graph_loss: 0.3379 - loss: 0.1070
Epoch 64/100
17/17 [==============================] - 0s 15ms/step - accuracy: 0.9754 - graph_loss: 0.3450 - loss: 0.1132
Epoch 65/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9819 - graph_loss: 0.3344 - loss: 0.1060
Epoch 66/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9852 - graph_loss: 0.3424 - loss: 0.1110
Epoch 67/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9810 - graph_loss: 0.3427 - loss: 0.1030
Epoch 68/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9824 - graph_loss: 0.3436 - loss: 0.1063
Epoch 69/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9856 - graph_loss: 0.3395 - loss: 0.1015
Epoch 70/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9805 - graph_loss: 0.3413 - loss: 0.1049
Epoch 71/100
17/17 [==============================] - 0s 15ms/step - accuracy: 0.9805 - graph_loss: 0.3432 - loss: 0.1018
Epoch 72/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9819 - graph_loss: 0.3473 - loss: 0.0988
Epoch 73/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9796 - graph_loss: 0.3466 - loss: 0.0966
Epoch 74/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9796 - graph_loss: 0.3348 - loss: 0.1009
Epoch 75/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9805 - graph_loss: 0.3414 - loss: 0.0978
Epoch 76/100
17/17 [==============================] - 0s 15ms/step - accuracy: 0.9828 - graph_loss: 0.3351 - loss: 0.0971
Epoch 77/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9833 - graph_loss: 0.3387 - loss: 0.0980
Epoch 78/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9791 - graph_loss: 0.3391 - loss: 0.1020
Epoch 79/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9833 - graph_loss: 0.3418 - loss: 0.0967
Epoch 80/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9861 - graph_loss: 0.3392 - loss: 0.0894
Epoch 81/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9838 - graph_loss: 0.3394 - loss: 0.0927
Epoch 82/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9833 - graph_loss: 0.3426 - loss: 0.0942
Epoch 83/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9819 - graph_loss: 0.3396 - loss: 0.0951
Epoch 84/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9875 - graph_loss: 0.3486 - loss: 0.0907
Epoch 85/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9856 - graph_loss: 0.3471 - loss: 0.0893
Epoch 86/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9842 - graph_loss: 0.3463 - loss: 0.0931
Epoch 87/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9856 - graph_loss: 0.3361 - loss: 0.0885
Epoch 88/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9842 - graph_loss: 0.3374 - loss: 0.0882
Epoch 89/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9852 - graph_loss: 0.3439 - loss: 0.0863
Epoch 90/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9856 - graph_loss: 0.3380 - loss: 0.0925
Epoch 91/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9889 - graph_loss: 0.3348 - loss: 0.0808
Epoch 92/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9870 - graph_loss: 0.3340 - loss: 0.0829
Epoch 93/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9903 - graph_loss: 0.3417 - loss: 0.0763
Epoch 94/100
17/17 [==============================] - 0s 14ms/step - accuracy: 0.9889 - graph_loss: 0.3424 - loss: 0.0802
Epoch 95/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9875 - graph_loss: 0.3377 - loss: 0.0814
Epoch 96/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9833 - graph_loss: 0.3424 - loss: 0.0860
Epoch 97/100
17/17 [==============================] - 0s 15ms/step - accuracy: 0.9842 - graph_loss: 0.3386 - loss: 0.0888
Epoch 98/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9856 - graph_loss: 0.3405 - loss: 0.0876
Epoch 99/100
17/17 [==============================] - 0s 13ms/step - accuracy: 0.9842 - graph_loss: 0.3333 - loss: 0.0837
Epoch 100/100
17/17 [==============================] - 0s 12ms/step - accuracy: 0.9856 - graph_loss: 0.3340 - loss: 0.0827

<tensorflow.python.keras.callbacks.History at 0x7f02a00727b8>

Evaluate MLP model with graph regularization

eval_results = dict(
    zip(graph_reg_model.metrics_names,
        graph_reg_model.evaluate(test_dataset, steps=HPARAMS.eval_steps)))
print_metrics('MLP + graph regularization', eval_results)
5/5 [==============================] - 0s 12ms/step - accuracy: 0.8192 - graph_loss: 0.0000e+00 - loss: 1.1746


Eval accuracy for  MLP + graph regularization :  0.8191681504249573
Eval loss for  MLP + graph regularization :  1.1746000051498413
Eval graph loss for  MLP + graph regularization :  0.0

The graph-regularized model's accuracy is about 2-3% higher than that of the base model (base_model).

Conclusion

We have demonstrated the use of graph regularization for document classification on a natural citation graph (Cora) using the Neural Structured Learning (NSL) framework. Our advanced tutorial involves synthesizing graphs based on sample embeddings before training a neural network with graph regularization. This approach is useful if the input does not contain an explicit graph.

We encourage users to experiment further by varying the amount of supervision as well as trying different neural architectures for graph regularization.