In [0]:
#@title Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
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:
Keras
sequential,
functional, or subclass API.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.Keras
model.Install the Neural Structured Learning package.
In [0]:
!pip install --quiet neural-structured-learning
In [0]:
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")
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.
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.
Each paper in the input effectively contains 2 features:
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.
Label: A single integer representing the class ID (category) of the paper.
In [0]:
!wget --quiet -P /tmp https://linqs-data.soe.ucsc.edu/public/lbc/cora.tgz
!tar -C /tmp -xvzf /tmp/cora.tgz
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:
tf.train.Example
instances.TFRecord
format.
In [0]:
!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
In [0]:
### 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'
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.
In [0]:
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()
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.
In [0]:
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.
In [0]:
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)
Let's peek into the test dataset to look at its contents.
In [0]:
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)
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.
In [0]:
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
In [0]:
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
In [0]:
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()
In [0]:
# 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()
In [0]:
# 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)
In [0]:
# 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'])
In [0]:
eval_results = dict(
zip(base_model.metrics_names,
base_model.evaluate(test_dataset, steps=HPARAMS.eval_steps)))
print_metrics('Base MLP model', eval_results)
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
.
In [0]:
# Build a new base MLP model.
base_reg_model_tag, base_reg_model = 'FUNCTIONAL', make_mlp_functional_model(
HPARAMS)
In [0]:
# 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)
In [0]:
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)
The graph-regularized model's accuracy is about 2-3% higher than that of the
base model (base_model
).
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.