Help protect the Great Barrier Reef with TensorFlow on Kaggle Join Challenge

Graph regularization for document classification using natural graphs

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

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:  2.8.0-rc0
Eager mode:  True
GPU is NOT AVAILABLE
2022-01-05 12:39:27.704660: E tensorflow/stream_executor/cuda/cuda_driver.cc:271] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected

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.cites
cora/cora.content

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

--2022-01-05 12:39:28--  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)... 185.199.108.133, 185.199.109.133, 185.199.110.133, ...
Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 11640 (11K) [text/plain]
Saving to: ‘preprocess_cora_dataset.py’

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

2022-01-05 12:39:28 (78.9 MB/s) - ‘preprocess_cora_dataset.py’ saved [11640/11640]

2022-01-05 12:39:31.378912: E tensorflow/stream_executor/cuda/cuda_driver.cc:271] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
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 (1.36 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 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.
  """

  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 during training.
    if training:
      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

  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: ['NL_nbr_0_weight', 'NL_nbr_0_words', 'words']
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 ... 1 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 6 2 0 6 1 3 5 0 1 2 3 6 1 1 0 3 5 2 3 1 4 1 6 1 3 2 2 2 0 3 2 1 3 3 2
 3 3 2 3 2 2 0 2 2 6 0 2 1 1 0 5 2 1 4 2 1 2 4 0 2 5 4 3 6 3 2 1 6 2 4 2 2
 6 4 6 4 3 5 2 2 2 4 2 2 2 1 2 2 2 4 2 3 6 2 0 6 6 0 2 6 2 1 2 0 1 1 3 2 0
 2 0 2 1 1 3 5 2 1 2 5 1 6 2 4 6 4], 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'])
  print('Batch of labels:', label_batch)

Feature list: ['words']
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 labels: tf.Tensor(
[5 2 2 2 1 2 6 3 2 3 6 1 3 6 4 4 2 3 3 0 2 0 5 2 1 0 6 3 6 4 2 2 3 0 4 2 2
 2 2 3 2 2 2 0 2 2 2 2 4 2 3 4 0 2 6 2 1 4 2 0 0 1 4 2 6 0 5 2 2 3 2 5 2 5
 2 3 2 2 2 2 2 6 6 3 2 4 2 6 3 2 2 6 2 4 2 2 1 3 4 6 0 0 2 4 2 1 3 6 6 2 6
 6 6 1 4 6 4 3 6 6 0 0 2 6 2 4 0 0], 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))
  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)(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)

    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=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
    metrics=['accuracy'])
base_model.fit(train_dataset, epochs=HPARAMS.train_epochs, verbose=1)

Epoch 1/100
/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/engine/functional.py:559: UserWarning: Input dict contained keys ['NL_nbr_0_weight', 'NL_nbr_0_words'] which did not match any model input. They will be ignored by the model.
  inputs = self._flatten_to_reference_inputs(inputs)
17/17 [==============================] - 1s 18ms/step - loss: 1.9521 - accuracy: 0.1838
Epoch 2/100
17/17 [==============================] - 0s 3ms/step - loss: 1.8590 - accuracy: 0.3044
Epoch 3/100
17/17 [==============================] - 0s 3ms/step - loss: 1.7770 - accuracy: 0.3601
Epoch 4/100
17/17 [==============================] - 0s 3ms/step - loss: 1.6655 - accuracy: 0.3898
Epoch 5/100
17/17 [==============================] - 0s 3ms/step - loss: 1.5386 - accuracy: 0.4543
Epoch 6/100
17/17 [==============================] - 0s 3ms/step - loss: 1.3856 - accuracy: 0.5077
Epoch 7/100
17/17 [==============================] - 0s 3ms/step - loss: 1.2736 - accuracy: 0.5531
Epoch 8/100
17/17 [==============================] - 0s 3ms/step - loss: 1.1636 - accuracy: 0.5889
Epoch 9/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0654 - accuracy: 0.6385
Epoch 10/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9703 - accuracy: 0.6761
Epoch 11/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8689 - accuracy: 0.7104
Epoch 12/100
17/17 [==============================] - 0s 3ms/step - loss: 0.7704 - accuracy: 0.7494
Epoch 13/100
17/17 [==============================] - 0s 3ms/step - loss: 0.7157 - accuracy: 0.7810
Epoch 14/100
17/17 [==============================] - 0s 3ms/step - loss: 0.6296 - accuracy: 0.8186
Epoch 15/100
17/17 [==============================] - 0s 3ms/step - loss: 0.5932 - accuracy: 0.8167
Epoch 16/100
17/17 [==============================] - 0s 3ms/step - loss: 0.5526 - accuracy: 0.8464
Epoch 17/100
17/17 [==============================] - 0s 3ms/step - loss: 0.5112 - accuracy: 0.8445
Epoch 18/100
17/17 [==============================] - 0s 3ms/step - loss: 0.4624 - accuracy: 0.8613
Epoch 19/100
17/17 [==============================] - 0s 3ms/step - loss: 0.4163 - accuracy: 0.8696
Epoch 20/100
17/17 [==============================] - 0s 3ms/step - loss: 0.3808 - accuracy: 0.8849
Epoch 21/100
17/17 [==============================] - 0s 3ms/step - loss: 0.3564 - accuracy: 0.8933
Epoch 22/100
17/17 [==============================] - 0s 3ms/step - loss: 0.3453 - accuracy: 0.9002
Epoch 23/100
17/17 [==============================] - 0s 3ms/step - loss: 0.3226 - accuracy: 0.9114
Epoch 24/100
17/17 [==============================] - 0s 3ms/step - loss: 0.3058 - accuracy: 0.9151
Epoch 25/100
17/17 [==============================] - 0s 3ms/step - loss: 0.2798 - accuracy: 0.9146
Epoch 26/100
17/17 [==============================] - 0s 3ms/step - loss: 0.2638 - accuracy: 0.9248
Epoch 27/100
17/17 [==============================] - 0s 3ms/step - loss: 0.2538 - accuracy: 0.9290
Epoch 28/100
17/17 [==============================] - 0s 3ms/step - loss: 0.2356 - accuracy: 0.9411
Epoch 29/100
17/17 [==============================] - 0s 3ms/step - loss: 0.2080 - accuracy: 0.9425
Epoch 30/100
17/17 [==============================] - 0s 3ms/step - loss: 0.2172 - accuracy: 0.9364
Epoch 31/100
17/17 [==============================] - 0s 3ms/step - loss: 0.2259 - accuracy: 0.9225
Epoch 32/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1944 - accuracy: 0.9480
Epoch 33/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1892 - accuracy: 0.9434
Epoch 34/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1718 - accuracy: 0.9592
Epoch 35/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1826 - accuracy: 0.9508
Epoch 36/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1585 - accuracy: 0.9559
Epoch 37/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1605 - accuracy: 0.9545
Epoch 38/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1529 - accuracy: 0.9550
Epoch 39/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1411 - accuracy: 0.9615
Epoch 40/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1366 - accuracy: 0.9624
Epoch 41/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1431 - accuracy: 0.9578
Epoch 42/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1241 - accuracy: 0.9619
Epoch 43/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1310 - accuracy: 0.9661
Epoch 44/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1284 - accuracy: 0.9652
Epoch 45/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1215 - accuracy: 0.9633
Epoch 46/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1130 - accuracy: 0.9722
Epoch 47/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1074 - accuracy: 0.9722
Epoch 48/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1143 - accuracy: 0.9694
Epoch 49/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1015 - accuracy: 0.9740
Epoch 50/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1077 - accuracy: 0.9698
Epoch 51/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1035 - accuracy: 0.9684
Epoch 52/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1076 - accuracy: 0.9694
Epoch 53/100
17/17 [==============================] - 0s 3ms/step - loss: 0.1000 - accuracy: 0.9689
Epoch 54/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0967 - accuracy: 0.9749
Epoch 55/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0994 - accuracy: 0.9703
Epoch 56/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0943 - accuracy: 0.9740
Epoch 57/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0923 - accuracy: 0.9735
Epoch 58/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0848 - accuracy: 0.9800
Epoch 59/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0836 - accuracy: 0.9782
Epoch 60/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0913 - accuracy: 0.9735
Epoch 61/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0823 - accuracy: 0.9773
Epoch 62/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0753 - accuracy: 0.9810
Epoch 63/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0746 - accuracy: 0.9777
Epoch 64/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0861 - accuracy: 0.9731
Epoch 65/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0765 - accuracy: 0.9787
Epoch 66/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0750 - accuracy: 0.9791
Epoch 67/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0725 - accuracy: 0.9814
Epoch 68/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0762 - accuracy: 0.9791
Epoch 69/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0645 - accuracy: 0.9842
Epoch 70/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0606 - accuracy: 0.9861
Epoch 71/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0775 - accuracy: 0.9805
Epoch 72/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0655 - accuracy: 0.9800
Epoch 73/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0629 - accuracy: 0.9833
Epoch 74/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0625 - accuracy: 0.9824
Epoch 75/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0607 - accuracy: 0.9838
Epoch 76/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0578 - accuracy: 0.9824
Epoch 77/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0568 - accuracy: 0.9842
Epoch 78/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0595 - accuracy: 0.9833
Epoch 79/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0615 - accuracy: 0.9842
Epoch 80/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0555 - accuracy: 0.9852
Epoch 81/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0517 - accuracy: 0.9870
Epoch 82/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0541 - accuracy: 0.9856
Epoch 83/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0533 - accuracy: 0.9884
Epoch 84/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0509 - accuracy: 0.9838
Epoch 85/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0600 - accuracy: 0.9828
Epoch 86/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0617 - accuracy: 0.9800
Epoch 87/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0599 - accuracy: 0.9800
Epoch 88/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0502 - accuracy: 0.9870
Epoch 89/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0416 - accuracy: 0.9907
Epoch 90/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0542 - accuracy: 0.9842
Epoch 91/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0490 - accuracy: 0.9847
Epoch 92/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0374 - accuracy: 0.9916
Epoch 93/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0467 - accuracy: 0.9893
Epoch 94/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0426 - accuracy: 0.9879
Epoch 95/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0543 - accuracy: 0.9861
Epoch 96/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0420 - accuracy: 0.9870
Epoch 97/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0461 - accuracy: 0.9861
Epoch 98/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0425 - accuracy: 0.9898
Epoch 99/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0406 - accuracy: 0.9907
Epoch 100/100
17/17 [==============================] - 0s 3ms/step - loss: 0.0486 - accuracy: 0.9847
<keras.callbacks.History at 0x7f6f9d5eacd0>

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 5ms/step - loss: 1.4192 - accuracy: 0.7939


Eval accuracy for  Base MLP model :  0.7938517332077026
Eval loss for  Base MLP model :  1.4192423820495605

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=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
    metrics=['accuracy'])
graph_reg_model.fit(train_dataset, epochs=HPARAMS.train_epochs, verbose=1)

Epoch 1/100
/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/framework/indexed_slices.py:446: UserWarning: Converting sparse IndexedSlices(IndexedSlices(indices=Tensor("gradient_tape/GraphRegularization/graph_loss/Reshape_1:0", shape=(None,), dtype=int32), values=Tensor("gradient_tape/GraphRegularization/graph_loss/Reshape:0", shape=(None, 7), dtype=float32), dense_shape=Tensor("gradient_tape/GraphRegularization/graph_loss/Cast:0", shape=(2,), dtype=int32))) to a dense Tensor of unknown shape. This may consume a large amount of memory.
  "shape. This may consume a large amount of memory." % value)
17/17 [==============================] - 2s 4ms/step - loss: 1.9798 - accuracy: 0.1601 - scaled_graph_loss: 0.0373
Epoch 2/100
17/17 [==============================] - 0s 3ms/step - loss: 1.9024 - accuracy: 0.2979 - scaled_graph_loss: 0.0254
Epoch 3/100
17/17 [==============================] - 0s 3ms/step - loss: 1.8623 - accuracy: 0.3160 - scaled_graph_loss: 0.0317
Epoch 4/100
17/17 [==============================] - 0s 3ms/step - loss: 1.8042 - accuracy: 0.3443 - scaled_graph_loss: 0.0498
Epoch 5/100
17/17 [==============================] - 0s 3ms/step - loss: 1.7552 - accuracy: 0.3582 - scaled_graph_loss: 0.0696
Epoch 6/100
17/17 [==============================] - 0s 3ms/step - loss: 1.7012 - accuracy: 0.4084 - scaled_graph_loss: 0.0866
Epoch 7/100
17/17 [==============================] - 0s 3ms/step - loss: 1.6578 - accuracy: 0.4515 - scaled_graph_loss: 0.1114
Epoch 8/100
17/17 [==============================] - 0s 3ms/step - loss: 1.6058 - accuracy: 0.5039 - scaled_graph_loss: 0.1300
Epoch 9/100
17/17 [==============================] - 0s 3ms/step - loss: 1.5498 - accuracy: 0.5434 - scaled_graph_loss: 0.1508
Epoch 10/100
17/17 [==============================] - 0s 3ms/step - loss: 1.5098 - accuracy: 0.6019 - scaled_graph_loss: 0.1651
Epoch 11/100
17/17 [==============================] - 0s 3ms/step - loss: 1.4746 - accuracy: 0.6302 - scaled_graph_loss: 0.1844
Epoch 12/100
17/17 [==============================] - 0s 3ms/step - loss: 1.4315 - accuracy: 0.6520 - scaled_graph_loss: 0.1917
Epoch 13/100
17/17 [==============================] - 0s 3ms/step - loss: 1.3932 - accuracy: 0.6770 - scaled_graph_loss: 0.2024
Epoch 14/100
17/17 [==============================] - 0s 3ms/step - loss: 1.3645 - accuracy: 0.7183 - scaled_graph_loss: 0.2145
Epoch 15/100
17/17 [==============================] - 0s 3ms/step - loss: 1.3265 - accuracy: 0.7369 - scaled_graph_loss: 0.2324
Epoch 16/100
17/17 [==============================] - 0s 3ms/step - loss: 1.3045 - accuracy: 0.7555 - scaled_graph_loss: 0.2358
Epoch 17/100
17/17 [==============================] - 0s 3ms/step - loss: 1.2836 - accuracy: 0.7652 - scaled_graph_loss: 0.2404
Epoch 18/100
17/17 [==============================] - 0s 3ms/step - loss: 1.2456 - accuracy: 0.7898 - scaled_graph_loss: 0.2469
Epoch 19/100
17/17 [==============================] - 0s 3ms/step - loss: 1.2348 - accuracy: 0.8074 - scaled_graph_loss: 0.2615
Epoch 20/100
17/17 [==============================] - 0s 3ms/step - loss: 1.2000 - accuracy: 0.8074 - scaled_graph_loss: 0.2542
Epoch 21/100
17/17 [==============================] - 0s 3ms/step - loss: 1.1994 - accuracy: 0.8260 - scaled_graph_loss: 0.2729
Epoch 22/100
17/17 [==============================] - 0s 3ms/step - loss: 1.1825 - accuracy: 0.8269 - scaled_graph_loss: 0.2676
Epoch 23/100
17/17 [==============================] - 0s 3ms/step - loss: 1.1598 - accuracy: 0.8455 - scaled_graph_loss: 0.2742
Epoch 24/100
17/17 [==============================] - 0s 3ms/step - loss: 1.1543 - accuracy: 0.8534 - scaled_graph_loss: 0.2797
Epoch 25/100
17/17 [==============================] - 0s 3ms/step - loss: 1.1456 - accuracy: 0.8552 - scaled_graph_loss: 0.2714
Epoch 26/100
17/17 [==============================] - 0s 3ms/step - loss: 1.1154 - accuracy: 0.8566 - scaled_graph_loss: 0.2796
Epoch 27/100
17/17 [==============================] - 0s 3ms/step - loss: 1.1150 - accuracy: 0.8687 - scaled_graph_loss: 0.2850
Epoch 28/100
17/17 [==============================] - 0s 3ms/step - loss: 1.1154 - accuracy: 0.8626 - scaled_graph_loss: 0.2772
Epoch 29/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0806 - accuracy: 0.8733 - scaled_graph_loss: 0.2756
Epoch 30/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0828 - accuracy: 0.8626 - scaled_graph_loss: 0.2907
Epoch 31/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0724 - accuracy: 0.8886 - scaled_graph_loss: 0.2834
Epoch 32/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0589 - accuracy: 0.8826 - scaled_graph_loss: 0.2881
Epoch 33/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0490 - accuracy: 0.8872 - scaled_graph_loss: 0.2972
Epoch 34/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0550 - accuracy: 0.8923 - scaled_graph_loss: 0.2935
Epoch 35/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0397 - accuracy: 0.8840 - scaled_graph_loss: 0.2795
Epoch 36/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0360 - accuracy: 0.8891 - scaled_graph_loss: 0.2966
Epoch 37/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0235 - accuracy: 0.8961 - scaled_graph_loss: 0.2890
Epoch 38/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0219 - accuracy: 0.8984 - scaled_graph_loss: 0.2965
Epoch 39/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0168 - accuracy: 0.9044 - scaled_graph_loss: 0.3023
Epoch 40/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0148 - accuracy: 0.9035 - scaled_graph_loss: 0.2984
Epoch 41/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9956 - accuracy: 0.9118 - scaled_graph_loss: 0.2888
Epoch 42/100
17/17 [==============================] - 0s 3ms/step - loss: 1.0019 - accuracy: 0.9021 - scaled_graph_loss: 0.2877
Epoch 43/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9956 - accuracy: 0.9049 - scaled_graph_loss: 0.2912
Epoch 44/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9986 - accuracy: 0.9026 - scaled_graph_loss: 0.3040
Epoch 45/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9939 - accuracy: 0.9067 - scaled_graph_loss: 0.3016
Epoch 46/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9828 - accuracy: 0.9058 - scaled_graph_loss: 0.2877
Epoch 47/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9629 - accuracy: 0.9137 - scaled_graph_loss: 0.2844
Epoch 48/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9645 - accuracy: 0.9146 - scaled_graph_loss: 0.2933
Epoch 49/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9752 - accuracy: 0.9165 - scaled_graph_loss: 0.3013
Epoch 50/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9552 - accuracy: 0.9179 - scaled_graph_loss: 0.2865
Epoch 51/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9539 - accuracy: 0.9193 - scaled_graph_loss: 0.3044
Epoch 52/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9443 - accuracy: 0.9183 - scaled_graph_loss: 0.3010
Epoch 53/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9559 - accuracy: 0.9244 - scaled_graph_loss: 0.2987
Epoch 54/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9497 - accuracy: 0.9225 - scaled_graph_loss: 0.2979
Epoch 55/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9674 - accuracy: 0.9183 - scaled_graph_loss: 0.3034
Epoch 56/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9537 - accuracy: 0.9174 - scaled_graph_loss: 0.2834
Epoch 57/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9341 - accuracy: 0.9188 - scaled_graph_loss: 0.2939
Epoch 58/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9392 - accuracy: 0.9225 - scaled_graph_loss: 0.2998
Epoch 59/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9240 - accuracy: 0.9313 - scaled_graph_loss: 0.3022
Epoch 60/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9368 - accuracy: 0.9267 - scaled_graph_loss: 0.2979
Epoch 61/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9306 - accuracy: 0.9234 - scaled_graph_loss: 0.2952
Epoch 62/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9197 - accuracy: 0.9230 - scaled_graph_loss: 0.2916
Epoch 63/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9360 - accuracy: 0.9206 - scaled_graph_loss: 0.2947
Epoch 64/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9181 - accuracy: 0.9299 - scaled_graph_loss: 0.2996
Epoch 65/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9105 - accuracy: 0.9341 - scaled_graph_loss: 0.2981
Epoch 66/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9014 - accuracy: 0.9323 - scaled_graph_loss: 0.2897
Epoch 67/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9059 - accuracy: 0.9364 - scaled_graph_loss: 0.3083
Epoch 68/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9053 - accuracy: 0.9309 - scaled_graph_loss: 0.2976
Epoch 69/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9099 - accuracy: 0.9258 - scaled_graph_loss: 0.3069
Epoch 70/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9025 - accuracy: 0.9355 - scaled_graph_loss: 0.2890
Epoch 71/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8849 - accuracy: 0.9281 - scaled_graph_loss: 0.2933
Epoch 72/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8959 - accuracy: 0.9323 - scaled_graph_loss: 0.2918
Epoch 73/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9074 - accuracy: 0.9248 - scaled_graph_loss: 0.3065
Epoch 74/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8845 - accuracy: 0.9369 - scaled_graph_loss: 0.2874
Epoch 75/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8873 - accuracy: 0.9401 - scaled_graph_loss: 0.2996
Epoch 76/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8942 - accuracy: 0.9327 - scaled_graph_loss: 0.3086
Epoch 77/100
17/17 [==============================] - 0s 3ms/step - loss: 0.9052 - accuracy: 0.9253 - scaled_graph_loss: 0.2986
Epoch 78/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8811 - accuracy: 0.9336 - scaled_graph_loss: 0.2948
Epoch 79/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8896 - accuracy: 0.9276 - scaled_graph_loss: 0.2919
Epoch 80/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8853 - accuracy: 0.9313 - scaled_graph_loss: 0.2944
Epoch 81/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8875 - accuracy: 0.9323 - scaled_graph_loss: 0.2925
Epoch 82/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8639 - accuracy: 0.9323 - scaled_graph_loss: 0.2967
Epoch 83/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8820 - accuracy: 0.9332 - scaled_graph_loss: 0.3047
Epoch 84/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8752 - accuracy: 0.9346 - scaled_graph_loss: 0.2942
Epoch 85/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8651 - accuracy: 0.9374 - scaled_graph_loss: 0.3066
Epoch 86/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8765 - accuracy: 0.9332 - scaled_graph_loss: 0.2881
Epoch 87/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8691 - accuracy: 0.9420 - scaled_graph_loss: 0.3030
Epoch 88/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8631 - accuracy: 0.9374 - scaled_graph_loss: 0.2916
Epoch 89/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8651 - accuracy: 0.9392 - scaled_graph_loss: 0.3032
Epoch 90/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8632 - accuracy: 0.9420 - scaled_graph_loss: 0.3019
Epoch 91/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8600 - accuracy: 0.9425 - scaled_graph_loss: 0.2965
Epoch 92/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8569 - accuracy: 0.9346 - scaled_graph_loss: 0.2977
Epoch 93/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8704 - accuracy: 0.9374 - scaled_graph_loss: 0.3083
Epoch 94/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8562 - accuracy: 0.9406 - scaled_graph_loss: 0.2883
Epoch 95/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8545 - accuracy: 0.9415 - scaled_graph_loss: 0.3030
Epoch 96/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8592 - accuracy: 0.9332 - scaled_graph_loss: 0.2927
Epoch 97/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8503 - accuracy: 0.9397 - scaled_graph_loss: 0.2927
Epoch 98/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8434 - accuracy: 0.9462 - scaled_graph_loss: 0.2937
Epoch 99/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8578 - accuracy: 0.9374 - scaled_graph_loss: 0.3064
Epoch 100/100
17/17 [==============================] - 0s 3ms/step - loss: 0.8504 - accuracy: 0.9411 - scaled_graph_loss: 0.3043
<keras.callbacks.History at 0x7f70041be650>

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 5ms/step - loss: 0.8884 - accuracy: 0.7957


Eval accuracy for  MLP + graph regularization :  0.7956600189208984
Eval loss for  MLP + graph regularization :  0.8883611559867859

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.