File size: 9,398 Bytes
b4be21e
 
 
 
 
 
 
 
16fad40
b4be21e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16fad40
 
 
 
 
 
 
 
 
 
 
 
51ad6eb
16fad40
 
 
 
81e6cca
392c4ff
 
81e6cca
 
b4be21e
 
 
81e6cca
b4be21e
 
 
81e6cca
 
b4be21e
81e6cca
 
298bd94
 
 
b4be21e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8cb6aa
b4be21e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
81e6cca
b4be21e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
298bd94
b4be21e
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
import streamlit as st
import tensorflow as tf
import tensorflow_gnn as tfgnn
from tensorflow_gnn import runner
from tensorflow_gnn.experimental import sampler
from tensorflow_gnn.models import mt_albis
import functools
import os
from typing import Mapping

# Set environment variable for legacy Keras
os.environ['TF_USE_LEGACY_KERAS'] = '1'

# Set Streamlit title
st.title("Solving OGBN-MAG end-to-end with TF-GNN")

st.write("Setting up the environment...")
tf.get_logger().setLevel('ERROR')
st.write(f"Running TF-GNN {tfgnn.__version__} under TensorFlow {tf.__version__}.")

NUM_TRAINING_SAMPLES = 629571
NUM_VALIDATION_SAMPLES = 64879

GRAPH_TENSOR_FILE = 'gs://download.tensorflow.org/data/ogbn-mag/sampled/v2/graph_tensor.example.pb'
SCHEMA_FILE = 'gs://download.tensorflow.org/data/ogbn-mag/sampled/v2/graph_schema.pbtxt'

# Load the graph schema and graph tensor
st.write("Loading graph schema and tensor...")
graph_schema = tfgnn.read_schema(SCHEMA_FILE)
serialized_ogbn_mag_graph_tensor_string = tf.io.read_file(GRAPH_TENSOR_FILE)

full_ogbn_mag_graph_tensor = tfgnn.parse_single_example(
    tfgnn.create_graph_spec_from_schema_pb(graph_schema, indices_dtype=tf.int64),
    serialized_ogbn_mag_graph_tensor_string)

st.write("Graph tensor loaded successfully.")

# Define sampling sizes
train_sampling_sizes = {
    "cites": 8,
    "rev_writes": 8,
    "writes": 8,
    "affiliated_with": 8,
    "has_topic": 8,
}
validation_sample_sizes = train_sampling_sizes.copy()

# Create sampling model
def create_sampling_model(full_graph_tensor: tfgnn.GraphTensor, sizes: Mapping[str, int]) -> tf.keras.Model:
    def edge_sampler(sampling_op: tfgnn.sampler.SamplingOp):
        edge_set_name = sampling_op.edge_set_name
        sample_size = sizes[edge_set_name]
        return sampler.InMemUniformEdgesSampler.from_graph_tensor(
            full_graph_tensor, edge_set_name, sample_size=sample_size
        )

    def get_features(node_set_name: tfgnn.NodeSetName):
        return sampler.InMemIndexToFeaturesAccessor.from_graph_tensor(
            full_graph_tensor, node_set_name
        )

    # Spell out the sampling procedure in python
    sampling_spec_builder = tfgnn.sampler.SamplingSpecBuilder(graph_schema)
    seed = sampling_spec_builder.seed("paper")
    papers_cited_from_seed = seed.sample(sizes["cites"], "cites")
    authors_of_papers = papers_cited_from_seed.join([seed]).sample(sizes["rev_writes"], "rev_writes")
    papers_by_authors = authors_of_papers.sample(sizes["writes"], "writes")
    institutions = authors_of_papers.sample(sizes["affiliated_with"], "affiliated_with")
    fields_of_study = seed.join([papers_cited_from_seed, papers_by_authors]).sample(sizes["has_topic"], "has_topic")
    sampling_spec = sampling_spec_builder.build()

    model = sampler.create_sampling_model_from_spec(
        graph_schema, sampling_spec, edge_sampler, get_features,
        seed_node_dtype=tf.int64)

    return model

# Create the sampling model
st.write("Creating sampling model...")
sampling_model = create_sampling_model(full_ogbn_mag_graph_tensor, train_sampling_sizes)

st.write("Sampling model created successfully.")

# Define seed dataset function
def seed_dataset(years: tf.Tensor, split_name: str) -> tf.data.Dataset:
    """Seed dataset as indices of papers within split years."""
    if split_name == "train":
        mask = years <= 2017  # 629,571 examples
    elif split_name == "validation":
        mask = years == 2018  # 64,879 examples
    elif split_name == "test":
        mask = years == 2019  # 41,939 examples
    else:
        raise ValueError(f"Unknown split_name: '{split_name}'")
    seed_indices = tf.squeeze(tf.where(mask), axis=-1)
    return tf.data.Dataset.from_tensor_slices(seed_indices)

# Define SubgraphDatasetProvider
class SubgraphDatasetProvider(runner.DatasetProvider):
    """Dataset Provider based on Sampler V2."""

    def __init__(self, full_graph_tensor: tfgnn.GraphTensor, sizes: Mapping[str, int], split_name: str):
        super().__init__()
        self._years = tf.squeeze(full_graph_tensor.node_sets["paper"]["year"], axis=-1)
        self._sampling_model = create_sampling_model(full_graph_tensor, sizes)
        self._split_name = split_name
        self.input_graph_spec = self._sampling_model.output.spec

    def get_dataset(self, context: tf.distribute.InputContext) -> tf.data.Dataset:
        """Creates TF dataset."""
        self._seed_dataset = seed_dataset(self._years, self._split_name)
        ds = self._seed_dataset.shard(
            num_shards=context.num_input_pipelines, index=context.input_pipeline_id)
        if self._split_name == "train":
            ds = ds.shuffle(NUM_TRAINING_SAMPLES).repeat()
        ds = ds.batch(128)
        ds = ds.map(
            functools.partial(self.sample),
            num_parallel_calls=tf.data.AUTOTUNE,
            deterministic=False,
        )
        return ds.unbatch().prefetch(tf.data.AUTOTUNE)

    def sample(self, seeds: tf.Tensor) -> tfgnn.GraphTensor:
        seeds = tf.cast(seeds, tf.int64)
        batch_size = tf.size(seeds)
        seeds_ragged = tf.RaggedTensor.from_row_lengths(
            seeds, tf.ones([batch_size], tf.int64),
        )
        return self._sampling_model(seeds_ragged)

# Create dataset providers
st.write("Creating dataset providers...")
train_ds_provider = SubgraphDatasetProvider(full_ogbn_mag_graph_tensor, train_sampling_sizes, "train")
valid_ds_provider = SubgraphDatasetProvider(full_ogbn_mag_graph_tensor, validation_sample_sizes, "validation")
example_input_graph_spec = train_ds_provider.input_graph_spec._unbatch()

st.write("Dataset providers created successfully.")

# Define the model function
node_state_dim = 128
num_graph_updates = 4
message_dim = 128
state_dropout_rate = 0.2
l2_regularization = 1e-5

def set_initial_node_states(node_set: tfgnn.NodeSet, node_set_name: str):
    if node_set_name == "field_of_study":
        return tf.keras.layers.Embedding(50_000, 32)(node_set["hashed_id"])
    if node_set_name == "institution":
        return tf.keras.layers.Embedding(6_500, 16)(node_set["hashed_id"])
    if node_set_name == "paper":
        return tf.keras.layers.Dense(node_state_dim, activation="relu")(node_set["feat"])
    if node_set_name == "author":
        return node_set["empty_state"]
    raise KeyError(f"Unexpected node_set_name='{node_set_name}'")

def model_fn(graph_tensor_spec: tfgnn.GraphTensorSpec):
    inputs = tf.keras.layers.Input(type_spec=graph_tensor_spec)
    graph = tfgnn.keras.layers.MapFeatures(node_sets_fn=set_initial_node_states)(inputs)
    
    for _ in range(num_graph_updates):
        graph = mt_albis.MtAlbisGraphUpdate(
            units=node_state_dim,
            message_dim=message_dim,
            attention_type="none",
            simple_conv_reduce_type="mean|sum",
            normalization_type="layer",
            next_state_type="residual",
            state_dropout_rate=state_dropout_rate,
            l2_regularization=l2_regularization
        )(graph)
    
    paper_state = tfgnn.keras.layers.Readout(node_set_name="paper", feature_name="state")(graph)
    paper_state = tf.keras.layers.Dense(349, activation="softmax")(paper_state)
    return tf.keras.Model(inputs, paper_state)

# Check for TPU/ GPU and set strategy
st.write("Setting up strategy for distributed training...")
if tf.config.list_physical_devices("TPU"):
    st.write("Using TPUStrategy")
    strategy = runner.TPUStrategy("local")
    train_padding = runner.FitOrSkipPadding(example_input_graph_spec, train_ds_provider)
    valid_padding = runner.TightPadding(example_input_graph_spec, valid_ds_provider)
elif tf.config.list_physical_devices("GPU"):
    st.write("Using MirroredStrategy for GPUs")
    strategy = tf.distribute.MirroredStrategy()
    train_padding = None
    valid_padding = None
else:
    st.write("Using default strategy")
    strategy = tf.distribute.get_strategy()
    train_padding = None
    valid_padding = None

st.write(f"Found {strategy.num_replicas_in_sync} replicas in sync")

# Define task
st.write("Defining the task...")
task = runner.NodeMulticlassClassification(
    num_classes=349,
    label_feature_name="paper_venue")

# Set hyperparameters
st.write("Setting hyperparameters...")
global_batch_size = 128
epochs = 10
initial_learning_rate = 0.001

steps_per_epoch = NUM_TRAINING_SAMPLES // global_batch_size
validation_steps = NUM_VALIDATION_SAMPLES // global_batch_size
learning_rate = tf.keras.optimizers.schedules.CosineDecay(
    initial_learning_rate, steps_per_epoch * epochs)
optimizer_fn = functools.partial(tf.keras.optimizers.Adam, learning_rate=learning_rate)

# Define trainer
st.write("Setting up the trainer...")
trainer = runner.KerasTrainer(
    strategy=strategy,
    model_dir="/tmp/gnn_model/",
    callbacks=None,
    steps_per_epoch=steps_per_epoch,
    validation_steps=validation_steps,
    restore_best_weights=False,
    checkpoint_every_n_steps="never",
    summarize_every_n_steps="never",
    backup_and_restore=False,
)

# Run training
st.write("Training the model...")
runner.run(
    task=task,
    model_fn=model_fn,
    trainer=trainer,
    optimizer_fn=optimizer_fn,
    epochs=epochs,
    global_batch_size=global_batch_size,
    train_ds_provider=train_ds_provider,
    valid_ds_provider=valid_ds_provider,
    gtspec=example_input_graph_spec,
)

st.write("Training completed successfully.")