This tutorial describes graph regularization from the Neural Structured Learning framework and demonstrates an end-to-end workflow for sentiment classification in a TFX pipeline.
 View on TensorFlow.org
|
 Run in Google Colab
|
 View on GitHub
|
 Download notebook
|
 See TF Hub model
|
Overview
This notebook classifies movie reviews as positive or negative using the text of the review. This is an example of binary classification, an important and widely applicable kind of machine learning problem.
We will demonstrate the use of graph regularization in this notebook by building a graph from the given input. The general recipe for building a graph-regularized model using the Neural Structured Learning (NSL) framework when the input does not contain an explicit graph is as follows:
- Create embeddings for each text sample in the input. This can be done using pre-trained models such as word2vec, Swivel, BERT etc.
- Build a graph based on these embeddings by using a similarity metric such as the 'L2' distance, 'cosine' distance, etc. Nodes in the graph correspond to samples and edges in the graph correspond to similarity between pairs of samples.
- Generate training data from the above synthesized graph and sample features. The resulting training data will contain neighbor features in addition to the original node features.
- Create a neural network as a base model using Estimators.
- Wrap the base model with the
add_graph_regularizationwrapper function, which is provided by the NSL framework, to create a new graph Estimator model. This new model will include a graph regularization loss as the regularization term in its training objective. - Train and evaluate the graph Estimator model.
In this tutorial, we integrate the above workflow in a TFX pipeline using several custom TFX components as well as a custom graph-regularized trainer component.
Below is the schematic for our TFX pipeline. Orange boxes represent off-the-shelf TFX components and pink boxes represent custom TFX components.
Upgrade Pip
To avoid upgrading Pip in a system when running locally, check to make sure that we're running in Colab. Local systems can of course be upgraded separately.
try:
import colab
!pip install --upgrade pip
except:
pass
Install Required Packages
!pip install -q -U \
tfx==1.2.0 \
neural-structured-learning \
tensorflow-hub \
tensorflow-datasets
Did you restart the runtime?
If you are using Google Colab, the first time that you run the cell above, you must restart the runtime (Runtime > Restart runtime ...). This is because of the way that Colab loads packages.
Dependencies and imports
import apache_beam as beam
import gzip as gzip_lib
import numpy as np
import os
import pprint
import shutil
import tempfile
import urllib
import uuid
pp = pprint.PrettyPrinter()
import tensorflow as tf
import neural_structured_learning as nsl
import tfx
from tfx.components.evaluator.component import Evaluator
from tfx.components.example_gen.import_example_gen.component import ImportExampleGen
from tfx.components.example_validator.component import ExampleValidator
from tfx.components.model_validator.component import ModelValidator
from tfx.components.pusher.component import Pusher
from tfx.components.schema_gen.component import SchemaGen
from tfx.components.statistics_gen.component import StatisticsGen
from tfx.components.trainer import executor as trainer_executor
from tfx.components.trainer.component import Trainer
from tfx.components.transform.component import Transform
from tfx.dsl.components.base import executor_spec
from tfx.orchestration.experimental.interactive.interactive_context import InteractiveContext
from tfx.proto import evaluator_pb2
from tfx.proto import example_gen_pb2
from tfx.proto import pusher_pb2
from tfx.proto import trainer_pb2
from tfx.types import artifact
from tfx.types import artifact_utils
from tfx.types import channel
from tfx.types import standard_artifacts
from tfx.types.standard_artifacts import Examples
from tfx.dsl.component.experimental.annotations import InputArtifact
from tfx.dsl.component.experimental.annotations import OutputArtifact
from tfx.dsl.component.experimental.annotations import Parameter
from tfx.dsl.component.experimental.decorators import component
from tensorflow_metadata.proto.v0 import anomalies_pb2
from tensorflow_metadata.proto.v0 import schema_pb2
from tensorflow_metadata.proto.v0 import statistics_pb2
import tensorflow_data_validation as tfdv
import tensorflow_transform as tft
import tensorflow_model_analysis as tfma
import tensorflow_hub as hub
import tensorflow_datasets as tfds
print("TF Version: ", tf.__version__)
print("Eager mode: ", tf.executing_eagerly())
print(
"GPU is",
"available" if tf.config.list_physical_devices("GPU") else "NOT AVAILABLE")
print("NSL Version: ", nsl.__version__)
print("TFX Version: ", tfx.__version__)
print("TFDV version: ", tfdv.__version__)
print("TFT version: ", tft.__version__)
print("TFMA version: ", tfma.__version__)
print("Hub version: ", hub.__version__)
print("Beam version: ", beam.__version__)
TF Version: 2.5.3 Eager mode: True GPU is available NSL Version: 1.3.1 TFX Version: 1.2.0 TFDV version: 1.2.0 TFT version: 1.2.0 TFMA version: 0.33.0 Hub version: 0.12.0 Beam version: 2.37.0
IMDB dataset
The IMDB dataset contains the text of 50,000 movie reviews from the Internet Movie Database. These are split into 25,000 reviews for training and 25,000 reviews for testing. The training and testing sets are balanced, meaning they contain an equal number of positive and negative reviews. Moreover, there are 50,000 additional unlabeled movie reviews.
Download preprocessed IMDB dataset
The following code downloads the IMDB dataset (or uses a cached copy if it has already been downloaded) using TFDS. To speed up this notebook we will use only 10,000 labeled reviews and 10,000 unlabeled reviews for training, and 10,000 test reviews for evaluation.
train_set, eval_set = tfds.load(
"imdb_reviews:1.0.0",
split=["train[:10000]+unsupervised[:10000]", "test[:10000]"],
shuffle_files=False)
Let's look at a few reviews from the training set:
for tfrecord in train_set.take(4):
print("Review: {}".format(tfrecord["text"].numpy().decode("utf-8")[:300]))
print("Label: {}\n".format(tfrecord["label"].numpy()))
Review: This was an absolutely terrible movie. Don't be lured in by Christopher Walken or Michael Ironside. Both are great actors, but this must simply be their worst role in history. Even their great acting could not redeem this movie's ridiculous storyline. This movie is an early nineties US propaganda pi Label: 0 Review: I have been known to fall asleep during films, but this is usually due to a combination of things including, really tired, being warm and comfortable on the sette and having just eaten a lot. However on this occasion I fell asleep because the film was rubbish. The plot development was constant. Cons Label: 0 Review: Mann photographs the Alberta Rocky Mountains in a superb fashion, and Jimmy Stewart and Walter Brennan give enjoyable performances as they always seem to do. <br /><br />But come on Hollywood - a Mountie telling the people of Dawson City, Yukon to elect themselves a marshal (yes a marshal!) and to e Label: 0 Review: This is the kind of film for a snowy Sunday afternoon when the rest of the world can go ahead with its own business as you descend into a big arm-chair and mellow for a couple of hours. Wonderful performances from Cher and Nicolas Cage (as always) gently row the plot along. There are no rapids to cr Label: 1
def _dict_to_example(instance):
"""Decoded CSV to tf example."""
feature = {}
for key, value in instance.items():
if value is None:
feature[key] = tf.train.Feature()
elif value.dtype == np.integer:
feature[key] = tf.train.Feature(
int64_list=tf.train.Int64List(value=value.tolist()))
elif value.dtype == np.float32:
feature[key] = tf.train.Feature(
float_list=tf.train.FloatList(value=value.tolist()))
else:
feature[key] = tf.train.Feature(
bytes_list=tf.train.BytesList(value=value.tolist()))
return tf.train.Example(features=tf.train.Features(feature=feature))
examples_path = tempfile.mkdtemp(prefix="tfx-data")
train_path = os.path.join(examples_path, "train.tfrecord")
eval_path = os.path.join(examples_path, "eval.tfrecord")
for path, dataset in [(train_path, train_set), (eval_path, eval_set)]:
with tf.io.TFRecordWriter(path) as writer:
for example in dataset:
writer.write(
_dict_to_example({
"label": np.array([example["label"].numpy()]),
"text": np.array([example["text"].numpy()]),
}).SerializeToString())
/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/ipykernel_launcher.py:7: DeprecationWarning: Converting `np.integer` or `np.signedinteger` to a dtype is deprecated. The current result is `np.dtype(np.int_)` which is not strictly correct. Note that the result depends on the system. To ensure stable results use may want to use `np.int64` or `np.int32`. import sys
Run TFX Components Interactively
In the cells that follow you will construct TFX components and run each one interactively within the InteractiveContext to obtain ExecutionResult objects. This mirrors the process of an orchestrator running components in a TFX DAG based on when the dependencies for each component are met.
context = InteractiveContext()
WARNING:absl:InteractiveContext pipeline_root argument not provided: using temporary directory /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t as root for pipeline outputs. WARNING:absl:InteractiveContext metadata_connection_config not provided: using SQLite ML Metadata database at /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/metadata.sqlite.
The ExampleGen Component
In any ML development process the first step when starting code development is to ingest the training and test datasets. The ExampleGen component brings data into the TFX pipeline.
Create an ExampleGen component and run it.
input_config = example_gen_pb2.Input(splits=[
example_gen_pb2.Input.Split(name='train', pattern='train.tfrecord'),
example_gen_pb2.Input.Split(name='eval', pattern='eval.tfrecord')
])
example_gen = ImportExampleGen(input_base=examples_path, input_config=input_config)
context.run(example_gen, enable_cache=True)
WARNING:apache_beam.runners.interactive.interactive_environment:Dependencies required for Interactive Beam PCollection visualization are not available, please use: `pip install apache-beam[interactive]` to install necessary dependencies to enable all data visualization features. WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter. WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
for artifact in example_gen.outputs['examples'].get():
print(artifact)
print('\nexample_gen.outputs is a {}'.format(type(example_gen.outputs)))
print(example_gen.outputs)
print(example_gen.outputs['examples'].get()[0].split_names)
Artifact(artifact: id: 1
type_id: 14
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/ImportExampleGen/examples/1"
properties {
key: "split_names"
value {
string_value: "[\"train\", \"eval\"]"
}
}
custom_properties {
key: "file_format"
value {
string_value: "tfrecords_gzip"
}
}
custom_properties {
key: "input_fingerprint"
value {
string_value: "split:train,num_files:1,total_bytes:27706811,xor_checksum:1648545250,sum_checksum:1648545250\nsplit:eval,num_files:1,total_bytes:13374744,xor_checksum:1648545256,sum_checksum:1648545256"
}
}
custom_properties {
key: "payload_format"
value {
string_value: "FORMAT_TF_EXAMPLE"
}
}
custom_properties {
key: "span"
value {
int_value: 0
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 14
name: "Examples"
properties {
key: "span"
value: INT
}
properties {
key: "split_names"
value: STRING
}
properties {
key: "version"
value: INT
}
)
example_gen.outputs is a <class 'dict'>
{'examples': Channel(
type_name: Examples
artifacts: [Artifact(artifact: id: 1
type_id: 14
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/ImportExampleGen/examples/1"
properties {
key: "split_names"
value {
string_value: "[\"train\", \"eval\"]"
}
}
custom_properties {
key: "file_format"
value {
string_value: "tfrecords_gzip"
}
}
custom_properties {
key: "input_fingerprint"
value {
string_value: "split:train,num_files:1,total_bytes:27706811,xor_checksum:1648545250,sum_checksum:1648545250\nsplit:eval,num_files:1,total_bytes:13374744,xor_checksum:1648545256,sum_checksum:1648545256"
}
}
custom_properties {
key: "payload_format"
value {
string_value: "FORMAT_TF_EXAMPLE"
}
}
custom_properties {
key: "span"
value {
int_value: 0
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 14
name: "Examples"
properties {
key: "span"
value: INT
}
properties {
key: "split_names"
value: STRING
}
properties {
key: "version"
value: INT
}
)]
additional_properties: {}
additional_custom_properties: {}
)}
["train", "eval"]
The component's outputs include 2 artifacts:
- the training examples (10,000 labeled reviews + 10,000 unlabeled reviews)
- the eval examples (10,000 labeled reviews)
The IdentifyExamples Custom Component
To use NSL, we will need each instance to have a unique ID. We create a custom component that adds such a unique ID to all instances across all splits. We leverage Apache Beam to be able to easily scale to large datasets if needed.
def make_example_with_unique_id(example, id_feature_name):
"""Adds a unique ID to the given `tf.train.Example` proto.
This function uses Python's 'uuid' module to generate a universally unique
identifier for each example.
Args:
example: An instance of a `tf.train.Example` proto.
id_feature_name: The name of the feature in the resulting `tf.train.Example`
that will contain the unique identifier.
Returns:
A new `tf.train.Example` proto that includes a unique identifier as an
additional feature.
"""
result = tf.train.Example()
result.CopyFrom(example)
unique_id = uuid.uuid4()
result.features.feature.get_or_create(
id_feature_name).bytes_list.MergeFrom(
tf.train.BytesList(value=[str(unique_id).encode('utf-8')]))
return result
@component
def IdentifyExamples(orig_examples: InputArtifact[Examples],
identified_examples: OutputArtifact[Examples],
id_feature_name: Parameter[str],
component_name: Parameter[str]) -> None:
# Get a list of the splits in input_data
splits_list = artifact_utils.decode_split_names(
split_names=orig_examples.split_names)
# For completeness, encode the splits names and payload_format.
# We could also just use input_data.split_names.
identified_examples.split_names = artifact_utils.encode_split_names(
splits=splits_list)
# TODO(b/168616829): Remove populating payload_format after tfx 0.25.0.
identified_examples.set_string_custom_property(
"payload_format",
orig_examples.get_string_custom_property("payload_format"))
for split in splits_list:
input_dir = artifact_utils.get_split_uri([orig_examples], split)
output_dir = artifact_utils.get_split_uri([identified_examples], split)
os.mkdir(output_dir)
with beam.Pipeline() as pipeline:
(pipeline
| 'ReadExamples' >> beam.io.ReadFromTFRecord(
os.path.join(input_dir, '*'),
coder=beam.coders.coders.ProtoCoder(tf.train.Example))
| 'AddUniqueId' >> beam.Map(make_example_with_unique_id, id_feature_name)
| 'WriteIdentifiedExamples' >> beam.io.WriteToTFRecord(
file_path_prefix=os.path.join(output_dir, 'data_tfrecord'),
coder=beam.coders.coders.ProtoCoder(tf.train.Example),
file_name_suffix='.gz'))
return
identify_examples = IdentifyExamples(
orig_examples=example_gen.outputs['examples'],
component_name=u'IdentifyExamples',
id_feature_name=u'id')
context.run(identify_examples, enable_cache=False)
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter. WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
The StatisticsGen Component
The StatisticsGen component computes descriptive statistics for your dataset. The statistics that it generates can be visualized for review, and are used for example validation and to infer a schema.
Create a StatisticsGen component and run it.
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(
examples=identify_examples.outputs["identified_examples"])
context.run(statistics_gen, enable_cache=True)
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
The SchemaGen Component
The SchemaGen component generates a schema for your data based on the statistics from StatisticsGen. It tries to infer the data types of each of your features, and the ranges of legal values for categorical features.
Create a SchemaGen component and run it.
# Generates schema based on statistics files.
schema_gen = SchemaGen(
statistics=statistics_gen.outputs['statistics'], infer_feature_shape=False)
context.run(schema_gen, enable_cache=True)
WARNING: Logging before InitGoogleLogging() is written to STDERR I0329 09:14:36.508610 8995 rdbms_metadata_access_object.cc:686] No property is defined for the Type
The generated artifact is just a schema.pbtxt containing a text representation of a schema_pb2.Schema protobuf:
train_uri = schema_gen.outputs['schema'].get()[0].uri
schema_filename = os.path.join(train_uri, 'schema.pbtxt')
schema = tfx.utils.io_utils.parse_pbtxt_file(
file_name=schema_filename, message=schema_pb2.Schema())
It can be visualized using tfdv.display_schema() (we will look at this in more detail in a subsequent lab):
tfdv.display_schema(schema)
The ExampleValidator Component
The ExampleValidator performs anomaly detection, based on the statistics from StatisticsGen and the schema from SchemaGen. It looks for problems such as missing values, values of the wrong type, or categorical values outside of the domain of acceptable values.
Create an ExampleValidator component and run it.
# Performs anomaly detection based on statistics and data schema.
validate_stats = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
context.run(validate_stats, enable_cache=False)
The SynthesizeGraph Component
Graph construction involves creating embeddings for text samples and then using a similarity function to compare the embeddings.
We will use pretrained Swivel embeddings to create embeddings in the
tf.train.Example format for each sample in the input. We will store the
resulting embeddings in the TFRecord format along with the sample's ID.
This is important and will allow us match sample embeddings with corresponding
nodes in the graph later.
Once we have the sample embeddings, we will use them to build a similarity graph, i.e, nodes in this graph will correspond to samples and edges in this graph will correspond to similarity between pairs of nodes.
Neural Structured Learning provides a graph building library to build a graph based on sample embeddings. It uses cosine similarity as the similarity measure to compare embeddings and build edges between them. It also allows us to specify a similarity threshold, which can be used to discard dissimilar edges from the final graph. In the following example, using 0.99 as the similarity threshold, we end up with a graph that has 111,066 bi-directional edges.
swivel_url = 'https://tfhub.dev/google/tf2-preview/gnews-swivel-20dim/1'
hub_layer = hub.KerasLayer(swivel_url, input_shape=[], dtype=tf.string)
def _bytes_feature(value):
"""Returns a bytes_list from a string / byte."""
return tf.train.Feature(bytes_list=tf.train.BytesList(value=value))
def _float_feature(value):
"""Returns a float_list from a float / double."""
return tf.train.Feature(float_list=tf.train.FloatList(value=value))
def create_embedding_example(example):
"""Create tf.Example containing the sample's embedding and its ID."""
sentence_embedding = hub_layer(tf.sparse.to_dense(example['text']))
# Flatten the sentence embedding back to 1-D.
sentence_embedding = tf.reshape(sentence_embedding, shape=[-1])
feature_dict = {
'id': _bytes_feature(tf.sparse.to_dense(example['id']).numpy()),
'embedding': _float_feature(sentence_embedding.numpy().tolist())
}
return tf.train.Example(features=tf.train.Features(feature=feature_dict))
def create_dataset(uri):
tfrecord_filenames = [os.path.join(uri, name) for name in os.listdir(uri)]
return tf.data.TFRecordDataset(tfrecord_filenames, compression_type='GZIP')
def create_embeddings(train_path, output_path):
dataset = create_dataset(train_path)
embeddings_path = os.path.join(output_path, 'embeddings.tfr')
feature_map = {
'label': tf.io.FixedLenFeature([], tf.int64),
'id': tf.io.VarLenFeature(tf.string),
'text': tf.io.VarLenFeature(tf.string)
}
with tf.io.TFRecordWriter(embeddings_path) as writer:
for tfrecord in dataset:
tensor_dict = tf.io.parse_single_example(tfrecord, feature_map)
embedding_example = create_embedding_example(tensor_dict)
writer.write(embedding_example.SerializeToString())
def build_graph(output_path, similarity_threshold):
embeddings_path = os.path.join(output_path, 'embeddings.tfr')
graph_path = os.path.join(output_path, 'graph.tsv')
graph_builder_config = nsl.configs.GraphBuilderConfig(
similarity_threshold=similarity_threshold,
lsh_splits=32,
lsh_rounds=15,
random_seed=12345)
nsl.tools.build_graph_from_config([embeddings_path], graph_path,
graph_builder_config)
"""Custom Artifact type"""
class SynthesizedGraph(tfx.types.artifact.Artifact):
"""Output artifact of the SynthesizeGraph component"""
TYPE_NAME = 'SynthesizedGraphPath'
PROPERTIES = {
'span': standard_artifacts.SPAN_PROPERTY,
'split_names': standard_artifacts.SPLIT_NAMES_PROPERTY,
}
@component
def SynthesizeGraph(identified_examples: InputArtifact[Examples],
synthesized_graph: OutputArtifact[SynthesizedGraph],
similarity_threshold: Parameter[float],
component_name: Parameter[str]) -> None:
# Get a list of the splits in input_data
splits_list = artifact_utils.decode_split_names(
split_names=identified_examples.split_names)
# We build a graph only based on the 'Split-train' split which includes both
# labeled and unlabeled examples.
train_input_examples_uri = os.path.join(identified_examples.uri,
'Split-train')
output_graph_uri = os.path.join(synthesized_graph.uri, 'Split-train')
os.mkdir(output_graph_uri)
print('Creating embeddings...')
create_embeddings(train_input_examples_uri, output_graph_uri)
print('Synthesizing graph...')
build_graph(output_graph_uri, similarity_threshold)
synthesized_graph.split_names = artifact_utils.encode_split_names(
splits=['Split-train'])
return
synthesize_graph = SynthesizeGraph(
identified_examples=identify_examples.outputs['identified_examples'],
component_name=u'SynthesizeGraph',
similarity_threshold=0.99)
context.run(synthesize_graph, enable_cache=False)
Creating embeddings... Synthesizing graph...
train_uri = synthesize_graph.outputs["synthesized_graph"].get()[0].uri
os.listdir(train_uri)
['Split-train']
graph_path = os.path.join(train_uri, "Split-train", "graph.tsv")
print("node 1\t\t\t\t\tnode 2\t\t\t\t\tsimilarity")
!head {graph_path}
print("...")
!tail {graph_path}
node 1 node 2 similarity 05af1f51-9800-49f2-8a74-f094259471a2 fe35e61f-651a-4afa-b8ce-bc267dc10c53 0.990020 fe35e61f-651a-4afa-b8ce-bc267dc10c53 05af1f51-9800-49f2-8a74-f094259471a2 0.990020 3f7fd5ae-86db-43bb-9271-a0c992adaf02 23e388d2-d608-4278-ae48-deb09bef4b56 0.992823 23e388d2-d608-4278-ae48-deb09bef4b56 3f7fd5ae-86db-43bb-9271-a0c992adaf02 0.992823 6a8c1fa0-277d-4cbc-8c7c-877c72e38db9 e815eacb-1cbf-4e7a-bb0c-0482e9478bf6 0.992471 e815eacb-1cbf-4e7a-bb0c-0482e9478bf6 6a8c1fa0-277d-4cbc-8c7c-877c72e38db9 0.992471 bcd26cbf-cb28-496a-a7e4-9844b2909b32 c9e8ebd2-fe67-4f36-bb78-878784b26f88 0.991234 c9e8ebd2-fe67-4f36-bb78-878784b26f88 bcd26cbf-cb28-496a-a7e4-9844b2909b32 0.991234 c9e8ebd2-fe67-4f36-bb78-878784b26f88 d19a3bc3-f978-4466-b61e-f59253b4a4ef 0.990838 d19a3bc3-f978-4466-b61e-f59253b4a4ef c9e8ebd2-fe67-4f36-bb78-878784b26f88 0.990838 ... 7b84d010-286f-4426-90ba-db851f0e5afa c3c03c2c-e22a-4106-a93f-3fbc757acba7 0.991327 c3c03c2c-e22a-4106-a93f-3fbc757acba7 7b84d010-286f-4426-90ba-db851f0e5afa 0.991327 fd2667ac-6066-4122-a408-9767169a58b3 97993e46-8bae-4daf-9f61-a801b2711b7e 0.991046 97993e46-8bae-4daf-9f61-a801b2711b7e fd2667ac-6066-4122-a408-9767169a58b3 0.991046 9565b48f-df21-408b-b62d-4f73074a8134 c81d6b23-1178-4081-b8cd-108e36accecd 0.991198 c81d6b23-1178-4081-b8cd-108e36accecd 9565b48f-df21-408b-b62d-4f73074a8134 0.991198 988901e1-244e-4cc3-96c4-0cf3915dd615 1012fca7-b391-4820-9944-d570eb0b6c34 0.990260 1012fca7-b391-4820-9944-d570eb0b6c34 988901e1-244e-4cc3-96c4-0cf3915dd615 0.990260 418fbbe2-9c6f-4496-a9ca-3b95ee880d96 4a09f662-3909-4bc2-bdb2-54072674cd52 0.991317 4a09f662-3909-4bc2-bdb2-54072674cd52 418fbbe2-9c6f-4496-a9ca-3b95ee880d96 0.991317
wc -l {graph_path}
222132 /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/SynthesizeGraph/synthesized_graph/6/Split-train/graph.tsv
The Transform Component
The Transform component performs data transformations and feature engineering. The results include an input TensorFlow graph which is used during both training and serving to preprocess the data before training or inference. This graph becomes part of the SavedModel that is the result of model training. Since the same input graph is used for both training and serving, the preprocessing will always be the same, and only needs to be written once.
The Transform component requires more code than many other components because of the arbitrary complexity of the feature engineering that you may need for the data and/or model that you're working with. It requires code files to be available which define the processing needed.
Each sample will include the following three features:
- id: The node ID of the sample.
- text_xf: An int64 list containing word IDs.
- label_xf: A singleton int64 identifying the target class of the review: 0=negative, 1=positive.
Let's define a module containing the preprocessing_fn() function that we will pass to the Transform component:
_transform_module_file = 'imdb_transform.py'
%%writefile {_transform_module_file}
import tensorflow as tf
import tensorflow_transform as tft
SEQUENCE_LENGTH = 100
VOCAB_SIZE = 10000
OOV_SIZE = 100
def tokenize_reviews(reviews, sequence_length=SEQUENCE_LENGTH):
reviews = tf.strings.lower(reviews)
reviews = tf.strings.regex_replace(reviews, r" '| '|^'|'$", " ")
reviews = tf.strings.regex_replace(reviews, "[^a-z' ]", " ")
tokens = tf.strings.split(reviews)[:, :sequence_length]
start_tokens = tf.fill([tf.shape(reviews)[0], 1], "<START>")
end_tokens = tf.fill([tf.shape(reviews)[0], 1], "<END>")
tokens = tf.concat([start_tokens, tokens, end_tokens], axis=1)
tokens = tokens[:, :sequence_length]
tokens = tokens.to_tensor(default_value="<PAD>")
pad = sequence_length - tf.shape(tokens)[1]
tokens = tf.pad(tokens, [[0, 0], [0, pad]], constant_values="<PAD>")
return tf.reshape(tokens, [-1, sequence_length])
def preprocessing_fn(inputs):
"""tf.transform's callback function for preprocessing inputs.
Args:
inputs: map from feature keys to raw not-yet-transformed features.
Returns:
Map from string feature key to transformed feature operations.
"""
outputs = {}
outputs["id"] = inputs["id"]
tokens = tokenize_reviews(_fill_in_missing(inputs["text"], ''))
outputs["text_xf"] = tft.compute_and_apply_vocabulary(
tokens,
top_k=VOCAB_SIZE,
num_oov_buckets=OOV_SIZE)
outputs["label_xf"] = _fill_in_missing(inputs["label"], -1)
return outputs
def _fill_in_missing(x, default_value):
"""Replace missing values in a SparseTensor.
Fills in missing values of `x` with the default_value.
Args:
x: A `SparseTensor` of rank 2. Its dense shape should have size at most 1
in the second dimension.
default_value: the value with which to replace the missing values.
Returns:
A rank 1 tensor where missing values of `x` have been filled in.
"""
if not isinstance(x, tf.sparse.SparseTensor):
return x
return tf.squeeze(
tf.sparse.to_dense(
tf.SparseTensor(x.indices, x.values, [x.dense_shape[0], 1]),
default_value),
axis=1)
Writing imdb_transform.py
Create and run the Transform component, referring to the files that were created above.
# Performs transformations and feature engineering in training and serving.
transform = Transform(
examples=identify_examples.outputs['identified_examples'],
schema=schema_gen.outputs['schema'],
module_file=_transform_module_file)
context.run(transform, enable_cache=True)
running bdist_wheel running build running build_py creating build creating build/lib copying imdb_transform.py -> build/lib installing to /tmp/tmpyr8869mo running install running install_lib copying build/lib/imdb_transform.py -> /tmp/tmpyr8869mo running install_egg_info running egg_info creating tfx_user_code_Transform.egg-info writing tfx_user_code_Transform.egg-info/PKG-INFO writing dependency_links to tfx_user_code_Transform.egg-info/dependency_links.txt writing top-level names to tfx_user_code_Transform.egg-info/top_level.txt writing manifest file 'tfx_user_code_Transform.egg-info/SOURCES.txt' reading manifest file 'tfx_user_code_Transform.egg-info/SOURCES.txt' writing manifest file 'tfx_user_code_Transform.egg-info/SOURCES.txt' Copying tfx_user_code_Transform.egg-info to /tmp/tmpyr8869mo/tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50-py3.7.egg-info running install_scripts creating /tmp/tmpyr8869mo/tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50.dist-info/WHEEL creating '/tmp/tmpo3tt6c48/tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50-py3-none-any.whl' and adding '/tmp/tmpyr8869mo' to it adding 'imdb_transform.py' adding 'tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50.dist-info/METADATA' adding 'tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50.dist-info/WHEEL' adding 'tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50.dist-info/top_level.txt' adding 'tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50.dist-info/RECORD' removing /tmp/tmpyr8869mo /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/setuptools/command/install.py:37: SetuptoolsDeprecationWarning: setup.py install is deprecated. Use build and pip and other standards-based tools. setuptools.SetuptoolsDeprecationWarning, I0329 09:16:07.937577 8995 rdbms_metadata_access_object.cc:686] No property is defined for the Type I0329 09:16:07.941587 8995 rdbms_metadata_access_object.cc:686] No property is defined for the Type Processing /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/_wheels/tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50-py3-none-any.whl Installing collected packages: tfx-user-code-Transform Successfully installed tfx-user-code-Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50 Processing /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/_wheels/tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50-py3-none-any.whl Installing collected packages: tfx-user-code-Transform Successfully installed tfx-user-code-Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50 WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow_transform/tf_utils.py:261: Tensor.experimental_ref (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version. Instructions for updating: Use ref() instead. WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow_transform/tf_utils.py:261: Tensor.experimental_ref (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version. Instructions for updating: Use ref() instead. Processing /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/_wheels/tfx_user_code_Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50-py3-none-any.whl WARNING:root:This output type hint will be ignored and not used for type-checking purposes. Typically, output type hints for a PTransform are single (or nested) types wrapped by a PCollection, PDone, or None. Got: Tuple[Dict[str, Union[NoneType, _Dataset]], Union[Dict[str, Dict[str, PCollection]], NoneType]] instead. Installing collected packages: tfx-user-code-Transform Successfully installed tfx-user-code-Transform-0.0+074f608d1f54105225e2fee77ebe4b6159a009eca01b5a0791099840a2185d50 WARNING:absl:Tables initialized inside a tf.function will be re-initialized on every invocation of the function. This re-initialization can have significant impact on performance. Consider lifting them out of the graph context using `tf.init_scope`.: compute_and_apply_vocabulary/apply_vocab/text_file_init/InitializeTableFromTextFileV2 WARNING:absl:Tables initialized inside a tf.function will be re-initialized on every invocation of the function. This re-initialization can have significant impact on performance. Consider lifting them out of the graph context using `tf.init_scope`.: compute_and_apply_vocabulary/apply_vocab/text_file_init/InitializeTableFromTextFileV2 WARNING:root:This output type hint will be ignored and not used for type-checking purposes. Typically, output type hints for a PTransform are single (or nested) types wrapped by a PCollection, PDone, or None. Got: Tuple[Dict[str, Union[NoneType, _Dataset]], Union[Dict[str, Dict[str, PCollection]], NoneType]] instead. WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter. 2022-03-29 09:16:17.653234: W tensorflow/python/util/util.cc:348] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them. INFO:tensorflow:Assets written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/transform_graph/7/.temp_path/tftransform_tmp/73787b44615f4ea980e98438660bb886/assets INFO:tensorflow:Assets written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/transform_graph/7/.temp_path/tftransform_tmp/73787b44615f4ea980e98438660bb886/assets INFO:tensorflow:Assets written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/transform_graph/7/.temp_path/tftransform_tmp/37681b13a11848b39218f3457a783faf/assets INFO:tensorflow:Assets written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/transform_graph/7/.temp_path/tftransform_tmp/37681b13a11848b39218f3457a783faf/assets
The Transform component has 2 types of outputs:
transform_graphis the graph that can perform the preprocessing operations (this graph will be included in the serving and evaluation models).transformed_examplesrepresents the preprocessed training and evaluation data.
transform.outputs
{'transform_graph': Channel(
type_name: TransformGraph
artifacts: [Artifact(artifact: id: 7
type_id: 25
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/transform_graph/7"
custom_properties {
key: "name"
value {
string_value: "transform_graph"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "Transform"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 25
name: "TransformGraph"
)]
additional_properties: {}
additional_custom_properties: {}
),
'transformed_examples': Channel(
type_name: Examples
artifacts: [Artifact(artifact: id: 8
type_id: 14
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/transformed_examples/7"
properties {
key: "split_names"
value {
string_value: "[\"train\", \"eval\"]"
}
}
custom_properties {
key: "name"
value {
string_value: "transformed_examples"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "Transform"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 14
name: "Examples"
properties {
key: "span"
value: INT
}
properties {
key: "split_names"
value: STRING
}
properties {
key: "version"
value: INT
}
)]
additional_properties: {}
additional_custom_properties: {}
),
'updated_analyzer_cache': Channel(
type_name: TransformCache
artifacts: [Artifact(artifact: id: 9
type_id: 26
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/updated_analyzer_cache/7"
custom_properties {
key: "name"
value {
string_value: "updated_analyzer_cache"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "Transform"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 26
name: "TransformCache"
)]
additional_properties: {}
additional_custom_properties: {}
),
'pre_transform_schema': Channel(
type_name: Schema
artifacts: [Artifact(artifact: id: 10
type_id: 19
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/pre_transform_schema/7"
custom_properties {
key: "name"
value {
string_value: "pre_transform_schema"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "Transform"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 19
name: "Schema"
)]
additional_properties: {}
additional_custom_properties: {}
),
'pre_transform_stats': Channel(
type_name: ExampleStatistics
artifacts: [Artifact(artifact: id: 11
type_id: 17
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/pre_transform_stats/7"
custom_properties {
key: "name"
value {
string_value: "pre_transform_stats"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "Transform"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
key: "span"
value: INT
}
properties {
key: "split_names"
value: STRING
}
)]
additional_properties: {}
additional_custom_properties: {}
),
'post_transform_schema': Channel(
type_name: Schema
artifacts: [Artifact(artifact: id: 12
type_id: 19
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/post_transform_schema/7"
custom_properties {
key: "name"
value {
string_value: "post_transform_schema"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "Transform"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 19
name: "Schema"
)]
additional_properties: {}
additional_custom_properties: {}
),
'post_transform_stats': Channel(
type_name: ExampleStatistics
artifacts: [Artifact(artifact: id: 13
type_id: 17
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/post_transform_stats/7"
custom_properties {
key: "name"
value {
string_value: "post_transform_stats"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "Transform"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
key: "span"
value: INT
}
properties {
key: "split_names"
value: STRING
}
)]
additional_properties: {}
additional_custom_properties: {}
),
'post_transform_anomalies': Channel(
type_name: ExampleAnomalies
artifacts: [Artifact(artifact: id: 14
type_id: 21
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/post_transform_anomalies/7"
custom_properties {
key: "name"
value {
string_value: "post_transform_anomalies"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "Transform"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 21
name: "ExampleAnomalies"
properties {
key: "span"
value: INT
}
properties {
key: "split_names"
value: STRING
}
)]
additional_properties: {}
additional_custom_properties: {}
)}
Take a peek at the transform_graph artifact: it points to a directory containing 3 subdirectories:
train_uri = transform.outputs['transform_graph'].get()[0].uri
os.listdir(train_uri)
['transform_fn', 'transformed_metadata', 'metadata']
The transform_fn subdirectory contains the actual preprocessing graph. The metadata subdirectory contains the schema of the original data. The transformed_metadata subdirectory contains the schema of the preprocessed data.
Take a look at some of the transformed examples and check that they are indeed processed as intended.
def pprint_examples(artifact, n_examples=3):
print("artifact:", artifact)
uri = os.path.join(artifact.uri, "Split-train")
print("uri:", uri)
tfrecord_filenames = [os.path.join(uri, name) for name in os.listdir(uri)]
print("tfrecord_filenames:", tfrecord_filenames)
dataset = tf.data.TFRecordDataset(tfrecord_filenames, compression_type="GZIP")
for tfrecord in dataset.take(n_examples):
serialized_example = tfrecord.numpy()
example = tf.train.Example.FromString(serialized_example)
pp.pprint(example)
pprint_examples(transform.outputs['transformed_examples'].get()[0])
artifact: Artifact(artifact: id: 8
type_id: 14
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/transformed_examples/7"
properties {
key: "split_names"
value {
string_value: "[\"train\", \"eval\"]"
}
}
custom_properties {
key: "name"
value {
string_value: "transformed_examples"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "Transform"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 14
name: "Examples"
properties {
key: "span"
value: INT
}
properties {
key: "split_names"
value: STRING
}
properties {
key: "version"
value: INT
}
)
uri: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/transformed_examples/7/Split-train
tfrecord_filenames: ['/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Transform/transformed_examples/7/Split-train/transformed_examples-00000-of-00001.gz']
features {
feature {
key: "id"
value {
bytes_list {
value: "8c54ccb1-1fa6-44ad-9ae7-7476adb48215"
}
}
}
feature {
key: "label_xf"
value {
int64_list {
value: 0
}
}
}
feature {
key: "text_xf"
value {
int64_list {
value: 13
value: 8
value: 14
value: 32
value: 338
value: 310
value: 15
value: 95
value: 27
value: 10001
value: 9
value: 31
value: 1173
value: 3153
value: 43
value: 495
value: 10060
value: 214
value: 26
value: 71
value: 142
value: 19
value: 8
value: 204
value: 339
value: 27
value: 74
value: 181
value: 238
value: 9
value: 440
value: 67
value: 74
value: 71
value: 94
value: 100
value: 22
value: 5442
value: 8
value: 1573
value: 607
value: 530
value: 8
value: 15
value: 6
value: 32
value: 378
value: 6292
value: 207
value: 2276
value: 388
value: 0
value: 84
value: 1023
value: 154
value: 65
value: 155
value: 52
value: 0
value: 10080
value: 7871
value: 65
value: 250
value: 74
value: 3202
value: 20
value: 10000
value: 3720
value: 10020
value: 10008
value: 1282
value: 3862
value: 3
value: 53
value: 3952
value: 110
value: 1879
value: 17
value: 3153
value: 14
value: 166
value: 19
value: 2
value: 1023
value: 1007
value: 9405
value: 9
value: 2
value: 15
value: 12
value: 14
value: 4504
value: 4
value: 109
value: 158
value: 1202
value: 7
value: 174
value: 505
value: 12
}
}
}
}
features {
feature {
key: "id"
value {
bytes_list {
value: "3be76372-c2fa-4693-9c00-dcc9075cdbf7"
}
}
}
feature {
key: "label_xf"
value {
int64_list {
value: 0
}
}
}
feature {
key: "text_xf"
value {
int64_list {
value: 13
value: 7
value: 23
value: 75
value: 494
value: 5
value: 748
value: 2155
value: 307
value: 91
value: 19
value: 8
value: 6
value: 499
value: 763
value: 5
value: 2
value: 1690
value: 4
value: 200
value: 593
value: 57
value: 1244
value: 120
value: 2364
value: 3
value: 4407
value: 21
value: 0
value: 10081
value: 3
value: 263
value: 42
value: 6947
value: 2
value: 169
value: 185
value: 21
value: 8
value: 5143
value: 7
value: 1339
value: 2155
value: 81
value: 0
value: 18
value: 14
value: 1468
value: 0
value: 86
value: 986
value: 14
value: 2259
value: 1790
value: 562
value: 3
value: 284
value: 200
value: 401
value: 5
value: 668
value: 19
value: 17
value: 58
value: 1934
value: 4
value: 45
value: 14
value: 4212
value: 113
value: 43
value: 135
value: 7
value: 753
value: 7
value: 224
value: 23
value: 1155
value: 179
value: 4
value: 0
value: 18
value: 19
value: 7
value: 191
value: 0
value: 2047
value: 4
value: 10
value: 3
value: 283
value: 42
value: 401
value: 5
value: 668
value: 4
value: 90
value: 234
value: 10023
value: 227
}
}
}
}
features {
feature {
key: "id"
value {
bytes_list {
value: "cf450121-a65b-4187-8d44-141c52389802"
}
}
}
feature {
key: "label_xf"
value {
int64_list {
value: 0
}
}
}
feature {
key: "text_xf"
value {
int64_list {
value: 13
value: 4577
value: 7158
value: 0
value: 10047
value: 3778
value: 3346
value: 9
value: 2
value: 758
value: 1915
value: 3
value: 2280
value: 1511
value: 3
value: 2003
value: 10020
value: 225
value: 786
value: 382
value: 16
value: 39
value: 203
value: 361
value: 5
value: 93
value: 11
value: 11
value: 19
value: 220
value: 21
value: 341
value: 2
value: 10000
value: 966
value: 0
value: 77
value: 4
value: 6677
value: 464
value: 10071
value: 5
value: 10042
value: 630
value: 2
value: 10044
value: 404
value: 2
value: 10044
value: 3
value: 5
value: 10008
value: 0
value: 1259
value: 630
value: 106
value: 10042
value: 6721
value: 10
value: 49
value: 21
value: 0
value: 2071
value: 20
value: 1292
value: 4
value: 0
value: 431
value: 11
value: 11
value: 166
value: 67
value: 2342
value: 5815
value: 12
value: 575
value: 21
value: 0
value: 1691
value: 537
value: 4
value: 0
value: 3605
value: 307
value: 0
value: 10054
value: 1563
value: 3115
value: 467
value: 4577
value: 3
value: 1069
value: 1158
value: 5
value: 23
value: 4279
value: 6677
value: 464
value: 20
value: 10004
}
}
}
}
The GraphAugmentation Component
Since we have the sample features and the synthesized graph, we can generate the augmented training data for Neural Structured Learning. The NSL framework provides a library to combine the graph and the sample features to produce the final training data for graph regularization. The resulting training data will include original sample features as well as features of their corresponding neighbors.
In this tutorial, we consider undirected edges and use a maximum of 3 neighbors per sample to augment training data with graph neighbors.
def split_train_and_unsup(input_uri):
'Separate the labeled and unlabeled instances.'
tmp_dir = tempfile.mkdtemp(prefix='tfx-data')
tfrecord_filenames = [
os.path.join(input_uri, filename) for filename in os.listdir(input_uri)
]
train_path = os.path.join(tmp_dir, 'train.tfrecord')
unsup_path = os.path.join(tmp_dir, 'unsup.tfrecord')
with tf.io.TFRecordWriter(train_path) as train_writer, \
tf.io.TFRecordWriter(unsup_path) as unsup_writer:
for tfrecord in tf.data.TFRecordDataset(
tfrecord_filenames, compression_type='GZIP'):
example = tf.train.Example()
example.ParseFromString(tfrecord.numpy())
if ('label_xf' not in example.features.feature or
example.features.feature['label_xf'].int64_list.value[0] == -1):
writer = unsup_writer
else:
writer = train_writer
writer.write(tfrecord.numpy())
return train_path, unsup_path
def gzip(filepath):
with open(filepath, 'rb') as f_in:
with gzip_lib.open(filepath + '.gz', 'wb') as f_out:
shutil.copyfileobj(f_in, f_out)
os.remove(filepath)
def copy_tfrecords(input_uri, output_uri):
for filename in os.listdir(input_uri):
input_filename = os.path.join(input_uri, filename)
output_filename = os.path.join(output_uri, filename)
shutil.copyfile(input_filename, output_filename)
@component
def GraphAugmentation(identified_examples: InputArtifact[Examples],
synthesized_graph: InputArtifact[SynthesizedGraph],
augmented_examples: OutputArtifact[Examples],
num_neighbors: Parameter[int],
component_name: Parameter[str]) -> None:
# Get a list of the splits in input_data
splits_list = artifact_utils.decode_split_names(
split_names=identified_examples.split_names)
train_input_uri = os.path.join(identified_examples.uri, 'Split-train')
eval_input_uri = os.path.join(identified_examples.uri, 'Split-eval')
train_graph_uri = os.path.join(synthesized_graph.uri, 'Split-train')
train_output_uri = os.path.join(augmented_examples.uri, 'Split-train')
eval_output_uri = os.path.join(augmented_examples.uri, 'Split-eval')
os.mkdir(train_output_uri)
os.mkdir(eval_output_uri)
# Separate the labeled and unlabeled examples from the 'Split-train' split.
train_path, unsup_path = split_train_and_unsup(train_input_uri)
output_path = os.path.join(train_output_uri, 'nsl_train_data.tfr')
pack_nbrs_args = dict(
labeled_examples_path=train_path,
unlabeled_examples_path=unsup_path,
graph_path=os.path.join(train_graph_uri, 'graph.tsv'),
output_training_data_path=output_path,
add_undirected_edges=True,
max_nbrs=num_neighbors)
print('nsl.tools.pack_nbrs arguments:', pack_nbrs_args)
nsl.tools.pack_nbrs(**pack_nbrs_args)
# Downstream components expect gzip'ed TFRecords.
gzip(output_path)
# The test examples are left untouched and are simply copied over.
copy_tfrecords(eval_input_uri, eval_output_uri)
augmented_examples.split_names = identified_examples.split_names
return
# Augments training data with graph neighbors.
graph_augmentation = GraphAugmentation(
identified_examples=transform.outputs['transformed_examples'],
synthesized_graph=synthesize_graph.outputs['synthesized_graph'],
component_name=u'GraphAugmentation',
num_neighbors=3)
context.run(graph_augmentation, enable_cache=False)
nsl.tools.pack_nbrs arguments: {'labeled_examples_path': '/tmp/tfx-data01uwdm_o/train.tfrecord', 'unlabeled_examples_path': '/tmp/tfx-data01uwdm_o/unsup.tfrecord', 'graph_path': '/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/SynthesizeGraph/synthesized_graph/6/Split-train/graph.tsv', 'output_training_data_path': '/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/GraphAugmentation/augmented_examples/8/Split-train/nsl_train_data.tfr', 'add_undirected_edges': True, 'max_nbrs': 3}
pprint_examples(graph_augmentation.outputs['augmented_examples'].get()[0], 6)
artifact: Artifact(artifact: id: 15
type_id: 14
uri: "/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/GraphAugmentation/augmented_examples/8"
properties {
key: "split_names"
value {
string_value: "[\"train\", \"eval\"]"
}
}
custom_properties {
key: "name"
value {
string_value: "augmented_examples"
}
}
custom_properties {
key: "producer_component"
value {
string_value: "GraphAugmentation"
}
}
custom_properties {
key: "state"
value {
string_value: "published"
}
}
custom_properties {
key: "tfx_version"
value {
string_value: "1.2.0"
}
}
state: LIVE
, artifact_type: id: 14
name: "Examples"
properties {
key: "span"
value: INT
}
properties {
key: "split_names"
value: STRING
}
properties {
key: "version"
value: INT
}
)
uri: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/GraphAugmentation/augmented_examples/8/Split-train
tfrecord_filenames: ['/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/GraphAugmentation/augmented_examples/8/Split-train/nsl_train_data.tfr.gz']
features {
feature {
key: "NL_num_nbrs"
value {
int64_list {
value: 0
}
}
}
feature {
key: "id"
value {
bytes_list {
value: "8c54ccb1-1fa6-44ad-9ae7-7476adb48215"
}
}
}
feature {
key: "label_xf"
value {
int64_list {
value: 0
}
}
}
feature {
key: "text_xf"
value {
int64_list {
value: 13
value: 8
value: 14
value: 32
value: 338
value: 310
value: 15
value: 95
value: 27
value: 10001
value: 9
value: 31
value: 1173
value: 3153
value: 43
value: 495
value: 10060
value: 214
value: 26
value: 71
value: 142
value: 19
value: 8
value: 204
value: 339
value: 27
value: 74
value: 181
value: 238
value: 9
value: 440
value: 67
value: 74
value: 71
value: 94
value: 100
value: 22
value: 5442
value: 8
value: 1573
value: 607
value: 530
value: 8
value: 15
value: 6
value: 32
value: 378
value: 6292
value: 207
value: 2276
value: 388
value: 0
value: 84
value: 1023
value: 154
value: 65
value: 155
value: 52
value: 0
value: 10080
value: 7871
value: 65
value: 250
value: 74
value: 3202
value: 20
value: 10000
value: 3720
value: 10020
value: 10008
value: 1282
value: 3862
value: 3
value: 53
value: 3952
value: 110
value: 1879
value: 17
value: 3153
value: 14
value: 166
value: 19
value: 2
value: 1023
value: 1007
value: 9405
value: 9
value: 2
value: 15
value: 12
value: 14
value: 4504
value: 4
value: 109
value: 158
value: 1202
value: 7
value: 174
value: 505
value: 12
}
}
}
}
features {
feature {
key: "NL_num_nbrs"
value {
int64_list {
value: 0
}
}
}
feature {
key: "id"
value {
bytes_list {
value: "3be76372-c2fa-4693-9c00-dcc9075cdbf7"
}
}
}
feature {
key: "label_xf"
value {
int64_list {
value: 0
}
}
}
feature {
key: "text_xf"
value {
int64_list {
value: 13
value: 7
value: 23
value: 75
value: 494
value: 5
value: 748
value: 2155
value: 307
value: 91
value: 19
value: 8
value: 6
value: 499
value: 763
value: 5
value: 2
value: 1690
value: 4
value: 200
value: 593
value: 57
value: 1244
value: 120
value: 2364
value: 3
value: 4407
value: 21
value: 0
value: 10081
value: 3
value: 263
value: 42
value: 6947
value: 2
value: 169
value: 185
value: 21
value: 8
value: 5143
value: 7
value: 1339
value: 2155
value: 81
value: 0
value: 18
value: 14
value: 1468
value: 0
value: 86
value: 986
value: 14
value: 2259
value: 1790
value: 562
value: 3
value: 284
value: 200
value: 401
value: 5
value: 668
value: 19
value: 17
value: 58
value: 1934
value: 4
value: 45
value: 14
value: 4212
value: 113
value: 43
value: 135
value: 7
value: 753
value: 7
value: 224
value: 23
value: 1155
value: 179
value: 4
value: 0
value: 18
value: 19
value: 7
value: 191
value: 0
value: 2047
value: 4
value: 10
value: 3
value: 283
value: 42
value: 401
value: 5
value: 668
value: 4
value: 90
value: 234
value: 10023
value: 227
}
}
}
}
features {
feature {
key: "NL_num_nbrs"
value {
int64_list {
value: 0
}
}
}
feature {
key: "id"
value {
bytes_list {
value: "cf450121-a65b-4187-8d44-141c52389802"
}
}
}
feature {
key: "label_xf"
value {
int64_list {
value: 0
}
}
}
feature {
key: "text_xf"
value {
int64_list {
value: 13
value: 4577
value: 7158
value: 0
value: 10047
value: 3778
value: 3346
value: 9
value: 2
value: 758
value: 1915
value: 3
value: 2280
value: 1511
value: 3
value: 2003
value: 10020
value: 225
value: 786
value: 382
value: 16
value: 39
value: 203
value: 361
value: 5
value: 93
value: 11
value: 11
value: 19
value: 220
value: 21
value: 341
value: 2
value: 10000
value: 966
value: 0
value: 77
value: 4
value: 6677
value: 464
value: 10071
value: 5
value: 10042
value: 630
value: 2
value: 10044
value: 404
value: 2
value: 10044
value: 3
value: 5
value: 10008
value: 0
value: 1259
value: 630
value: 106
value: 10042
value: 6721
value: 10
value: 49
value: 21
value: 0
value: 2071
value: 20
value: 1292
value: 4
value: 0
value: 431
value: 11
value: 11
value: 166
value: 67
value: 2342
value: 5815
value: 12
value: 575
value: 21
value: 0
value: 1691
value: 537
value: 4
value: 0
value: 3605
value: 307
value: 0
value: 10054
value: 1563
value: 3115
value: 467
value: 4577
value: 3
value: 1069
value: 1158
value: 5
value: 23
value: 4279
value: 6677
value: 464
value: 20
value: 10004
}
}
}
}
features {
feature {
key: "NL_num_nbrs"
value {
int64_list {
value: 0
}
}
}
feature {
key: "id"
value {
bytes_list {
value: "30741807-497c-44be-bc5e-845bd1143912"
}
}
}
feature {
key: "label_xf"
value {
int64_list {
value: 1
}
}
}
feature {
key: "text_xf"
value {
int64_list {
value: 13
value: 8
value: 6
value: 0
value: 251
value: 4
value: 18
value: 20
value: 2
value: 6783
value: 2295
value: 2338
value: 52
value: 0
value: 468
value: 4
value: 0
value: 189
value: 73
value: 153
value: 1294
value: 17
value: 90
value: 234
value: 935
value: 16
value: 25
value: 10024
value: 92
value: 2
value: 192
value: 4218
value: 3317
value: 3
value: 10098
value: 20
value: 2
value: 356
value: 4
value: 565
value: 334
value: 382
value: 36
value: 6989
value: 3
value: 6065
value: 2510
value: 16
value: 203
value: 7264
value: 2849
value: 0
value: 86
value: 346
value: 50
value: 26
value: 58
value: 10020
value: 5
value: 1464
value: 58
value: 2081
value: 2969
value: 42
value: 2
value: 2364
value: 3
value: 1402
value: 10062
value: 138
value: 147
value: 614
value: 115
value: 29
value: 90
value: 105
value: 2
value: 223
value: 18
value: 9
value: 160
value: 324
value: 3
value: 24
value: 12
value: 1252
value: 0
value: 2142
value: 10
value: 1832
value: 111
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
}
}
}
}
features {
feature {
key: "NL_num_nbrs"
value {
int64_list {
value: 0
}
}
}
feature {
key: "id"
value {
bytes_list {
value: "b81cf0f1-3bc6-461e-97d3-6bf7ba99570f"
}
}
}
feature {
key: "label_xf"
value {
int64_list {
value: 1
}
}
}
feature {
key: "text_xf"
value {
int64_list {
value: 13
value: 16
value: 423
value: 23
value: 1367
value: 30
value: 0
value: 363
value: 12
value: 153
value: 3174
value: 9
value: 8
value: 18
value: 26
value: 667
value: 338
value: 1372
value: 0
value: 86
value: 46
value: 9200
value: 282
value: 0
value: 10091
value: 4
value: 0
value: 694
value: 10028
value: 52
value: 362
value: 26
value: 202
value: 39
value: 216
value: 5
value: 27
value: 5822
value: 19
value: 52
value: 58
value: 362
value: 26
value: 202
value: 39
value: 474
value: 0
value: 10029
value: 4
value: 2
value: 243
value: 143
value: 386
value: 3
value: 0
value: 386
value: 579
value: 2
value: 132
value: 57
value: 725
value: 88
value: 140
value: 30
value: 27
value: 33
value: 1359
value: 29
value: 8
value: 567
value: 35
value: 106
value: 230
value: 60
value: 0
value: 3041
value: 5
value: 7879
value: 28
value: 281
value: 110
value: 111
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
value: 1
}
}
}
}
features {
feature {
key: "NL_num_nbrs"
value {
int64_list {
value: 0
}
}
}
feature {
key: "id"
value {
bytes_list {
value: "c7ee6142-99d7-4e1a-a764-f2bf35eab814"
}
}
}
feature {
key: "label_xf"
value {
int64_list {
value: 1
}
}
}
feature {
key: "text_xf"
value {
int64_list {
value: 13
value: 8
value: 6
value: 2
value: 18
value: 69
value: 140
value: 27
value: 83
value: 31
value: 1877
value: 905
value: 9
value: 10057
value: 31
value: 43
value: 2115
value: 36
value: 32
value: 2057
value: 6133
value: 10
value: 6
value: 32
value: 2474
value: 1614
value: 3
value: 2707
value: 990
value: 4
value: 10067
value: 9
value: 2
value: 1532
value: 242
value: 90
value: 3757
value: 3
value: 90
value: 10026
value: 0
value: 242
value: 6
value: 260
value: 31
value: 24
value: 4
value: 0
value: 84
value: 497
value: 177
value: 1151
value: 777
value: 9
value: 397
value: 552
value: 7726
value: 10051
value: 34
value: 14
value: 379
value: 33
value: 1829
value: 9
value: 123
value: 0
value: 916
value: 10028
value: 7
value: 64
value: 571
value: 12
value: 8
value: 18
value: 27
value: 687
value: 9
value: 30
value: 5609
value: 16
value: 25
value: 99
value: 117
value: 66
value: 2
value: 130
value: 21
value: 8
value: 842
value: 7726
value: 10051
value: 6
value: 338
value: 1107
value: 3
value: 24
value: 10020
value: 29
value: 53
value: 1476
}
}
}
}
The Trainer Component
The Trainer component trains models using TensorFlow.
Create a Python module containing a trainer_fn function, which must return an estimator. If you prefer creating a Keras model, you can do so and then convert it to an estimator using keras.model_to_estimator().
# Setup paths.
_trainer_module_file = 'imdb_trainer.py'
%%writefile {_trainer_module_file}
import neural_structured_learning as nsl
import tensorflow as tf
import tensorflow_model_analysis as tfma
import tensorflow_transform as tft
from tensorflow_transform.tf_metadata import schema_utils
NBR_FEATURE_PREFIX = 'NL_nbr_'
NBR_WEIGHT_SUFFIX = '_weight'
LABEL_KEY = 'label'
ID_FEATURE_KEY = 'id'
def _transformed_name(key):
return key + '_xf'
def _transformed_names(keys):
return [_transformed_name(key) for key in keys]
# Hyperparameters:
#
# We will use an instance of `HParams` to inclue various hyperparameters and
# constants used for training and evaluation. We briefly describe each of them
# below:
#
# - max_seq_length: This is the maximum number of words considered from each
# movie review in this example.
# - vocab_size: This is the size of the vocabulary considered for this
# example.
# - oov_size: This is the out-of-vocabulary size considered for this example.
# - 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 `num_neighbors`
# argument used above in the GraphAugmentation component when
# invoking `nsl.tools.pack_nbrs`.
# - num_fc_units: The number of units in the fully connected layer of the
# neural network.
class HParams(object):
"""Hyperparameters used for training."""
def __init__(self):
### dataset parameters
# The following 3 values should match those defined in the Transform
# Component.
self.max_seq_length = 100
self.vocab_size = 10000
self.oov_size = 100
### Neural Graph Learning parameters
self.distance_type = nsl.configs.DistanceType.L2
self.graph_regularization_multiplier = 0.1
# The following value has to be at most the value of 'num_neighbors' used
# in the GraphAugmentation component.
self.num_neighbors = 1
### Model Architecture
self.num_embedding_dims = 16
self.num_fc_units = 64
HPARAMS = HParams()
def optimizer_fn():
"""Returns an instance of `tf.Optimizer`."""
return tf.compat.v1.train.RMSPropOptimizer(
learning_rate=0.0001, decay=1e-6)
def build_train_op(loss, global_step):
"""Builds a train op to optimize the given loss using gradient descent."""
with tf.name_scope('train'):
optimizer = optimizer_fn()
train_op = optimizer.minimize(loss=loss, global_step=global_step)
return train_op
# Building the model:
#
# A neural network is created by stacking layers—this requires two main
# architectural decisions:
# * How many layers to use in the model?
# * How many *hidden units* to use for each layer?
#
# In this example, the input data consists of an array of word-indices. The
# labels to predict are either 0 or 1. We will use a feed-forward neural network
# as our base model in this tutorial.
def feed_forward_model(features, is_training, reuse=tf.compat.v1.AUTO_REUSE):
"""Builds a simple 2 layer feed forward neural network.
The layers are effectively stacked sequentially to build the classifier. The
first layer is an Embedding layer, which takes the integer-encoded vocabulary
and looks up the embedding vector for each word-index. These vectors are
learned as the model trains. The vectors add a dimension to the output array.
The resulting dimensions are: (batch, sequence, embedding). Next is a global
average pooling 1D layer, which reduces the dimensionality of its inputs from
3D to 2D. This fixed-length output vector is piped through a fully-connected
(Dense) layer with 16 hidden units. The last layer is densely connected with a
single output node. Using the sigmoid activation function, this value is a
float between 0 and 1, representing a probability, or confidence level.
Args:
features: A dictionary containing batch features returned from the
`input_fn`, that include sample features, corresponding neighbor features,
and neighbor weights.
is_training: a Python Boolean value or a Boolean scalar Tensor, indicating
whether to apply dropout.
reuse: a Python Boolean value for reusing variable scope.
Returns:
logits: Tensor of shape [batch_size, 1].
representations: Tensor of shape [batch_size, _] for graph regularization.
This is the representation of each example at the graph regularization
layer.
"""
with tf.compat.v1.variable_scope('ff', reuse=reuse):
inputs = features[_transformed_name('text')]
embeddings = tf.compat.v1.get_variable(
'embeddings',
shape=[
HPARAMS.vocab_size + HPARAMS.oov_size, HPARAMS.num_embedding_dims
])
embedding_layer = tf.nn.embedding_lookup(embeddings, inputs)
pooling_layer = tf.compat.v1.layers.AveragePooling1D(
pool_size=HPARAMS.max_seq_length, strides=HPARAMS.max_seq_length)(
embedding_layer)
# Shape of pooling_layer is now [batch_size, 1, HPARAMS.num_embedding_dims]
pooling_layer = tf.reshape(pooling_layer, [-1, HPARAMS.num_embedding_dims])
dense_layer = tf.compat.v1.layers.Dense(
16, activation='relu')(
pooling_layer)
output_layer = tf.compat.v1.layers.Dense(
1, activation='sigmoid')(
dense_layer)
# Graph regularization will be done on the penultimate (dense) layer
# because the output layer is a single floating point number.
return output_layer, dense_layer
# A note on hidden units:
#
# The above model has two intermediate or "hidden" layers, between the input and
# output, and excluding the Embedding layer. The number of outputs (units,
# nodes, or neurons) is the dimension of the representational space for the
# layer. In other words, the amount of freedom the network is allowed when
# learning an internal representation. If a model has more hidden units
# (a higher-dimensional representation space), and/or more layers, then the
# network can learn more complex representations. However, it makes the network
# more computationally expensive and may lead to learning unwanted
# patterns—patterns that improve performance on training data but not on the
# test data. This is called overfitting.
# This function will be used to generate the embeddings for samples and their
# corresponding neighbors, which will then be used for graph regularization.
def embedding_fn(features, mode):
"""Returns the embedding corresponding to the given features.
Args:
features: A dictionary containing batch features returned from the
`input_fn`, that include sample features, corresponding neighbor features,
and neighbor weights.
mode: Specifies if this is training, evaluation, or prediction. See
tf.estimator.ModeKeys.
Returns:
The embedding that will be used for graph regularization.
"""
is_training = (mode == tf.estimator.ModeKeys.TRAIN)
_, embedding = feed_forward_model(features, is_training)
return embedding
def feed_forward_model_fn(features, labels, mode, params, config):
"""Implementation of the model_fn for the base feed-forward model.
Args:
features: This is the first item returned from the `input_fn` passed to
`train`, `evaluate`, and `predict`. This should be a single `Tensor` or
`dict` of same.
labels: This is the second item returned from the `input_fn` passed to
`train`, `evaluate`, and `predict`. This should be a single `Tensor` or
`dict` of same (for multi-head models). If mode is `ModeKeys.PREDICT`,
`labels=None` will be passed. If the `model_fn`'s signature does not
accept `mode`, the `model_fn` must still be able to handle `labels=None`.
mode: Optional. Specifies if this training, evaluation or prediction. See
`ModeKeys`.
params: An HParams instance as returned by get_hyper_parameters().
config: Optional configuration object. Will receive what is passed to
Estimator in `config` parameter, or the default `config`. Allows updating
things in your model_fn based on configuration such as `num_ps_replicas`,
or `model_dir`. Unused currently.
Returns:
A `tf.estimator.EstimatorSpec` for the base feed-forward model. This does
not include graph-based regularization.
"""
is_training = mode == tf.estimator.ModeKeys.TRAIN
# Build the computation graph.
probabilities, _ = feed_forward_model(features, is_training)
predictions = tf.round(probabilities)
if mode == tf.estimator.ModeKeys.PREDICT:
# labels will be None, and no loss to compute.
cross_entropy_loss = None
eval_metric_ops = None
else:
# Loss is required in train and eval modes.
# Flatten 'probabilities' to 1-D.
probabilities = tf.reshape(probabilities, shape=[-1])
cross_entropy_loss = tf.compat.v1.keras.losses.binary_crossentropy(
labels, probabilities)
eval_metric_ops = {
'accuracy': tf.compat.v1.metrics.accuracy(labels, predictions)
}
if is_training:
global_step = tf.compat.v1.train.get_or_create_global_step()
train_op = build_train_op(cross_entropy_loss, global_step)
else:
train_op = None
return tf.estimator.EstimatorSpec(
mode=mode,
predictions={
'probabilities': probabilities,
'predictions': predictions
},
loss=cross_entropy_loss,
train_op=train_op,
eval_metric_ops=eval_metric_ops)
# Tf.Transform considers these features as "raw"
def _get_raw_feature_spec(schema):
return schema_utils.schema_as_feature_spec(schema).feature_spec
def _gzip_reader_fn(filenames):
"""Small utility returning a record reader that can read gzip'ed files."""
return tf.data.TFRecordDataset(
filenames,
compression_type='GZIP')
def _example_serving_receiver_fn(tf_transform_output, schema):
"""Build the serving in inputs.
Args:
tf_transform_output: A TFTransformOutput.
schema: the schema of the input data.
Returns:
Tensorflow graph which parses examples, applying tf-transform to them.
"""
raw_feature_spec = _get_raw_feature_spec(schema)
raw_feature_spec.pop(LABEL_KEY)
# We don't need the ID feature for serving.
raw_feature_spec.pop(ID_FEATURE_KEY)
raw_input_fn = tf.estimator.export.build_parsing_serving_input_receiver_fn(
raw_feature_spec, default_batch_size=None)
serving_input_receiver = raw_input_fn()
transformed_features = tf_transform_output.transform_raw_features(
serving_input_receiver.features)
# Even though, LABEL_KEY was removed from 'raw_feature_spec', the transform
# operation would have injected the transformed LABEL_KEY feature with a
# default value.
transformed_features.pop(_transformed_name(LABEL_KEY))
return tf.estimator.export.ServingInputReceiver(
transformed_features, serving_input_receiver.receiver_tensors)
def _eval_input_receiver_fn(tf_transform_output, schema):
"""Build everything needed for the tf-model-analysis to run the model.
Args:
tf_transform_output: A TFTransformOutput.
schema: the schema of the input data.
Returns:
EvalInputReceiver function, which contains:
- Tensorflow graph which parses raw untransformed features, applies the
tf-transform preprocessing operators.
- Set of raw, untransformed features.
- Label against which predictions will be compared.
"""
# Notice that the inputs are raw features, not transformed features here.
raw_feature_spec = _get_raw_feature_spec(schema)
# We don't need the ID feature for TFMA.
raw_feature_spec.pop(ID_FEATURE_KEY)
raw_input_fn = tf.estimator.export.build_parsing_serving_input_receiver_fn(
raw_feature_spec, default_batch_size=None)
serving_input_receiver = raw_input_fn()
transformed_features = tf_transform_output.transform_raw_features(
serving_input_receiver.features)
labels = transformed_features.pop(_transformed_name(LABEL_KEY))
return tfma.export.EvalInputReceiver(
features=transformed_features,
receiver_tensors=serving_input_receiver.receiver_tensors,
labels=labels)
def _augment_feature_spec(feature_spec, num_neighbors):
"""Augments `feature_spec` to include neighbor features.
Args:
feature_spec: Dictionary of feature keys mapping to TF feature types.
num_neighbors: Number of neighbors to use for feature key augmentation.
Returns:
An augmented `feature_spec` that includes neighbor feature keys.
"""
for i in range(num_neighbors):
feature_spec['{}{}_{}'.format(NBR_FEATURE_PREFIX, i, 'id')] = \
tf.io.VarLenFeature(dtype=tf.string)
# We don't care about the neighbor features corresponding to
# _transformed_name(LABEL_KEY) because the LABEL_KEY feature will be
# removed from the feature spec during training/evaluation.
feature_spec['{}{}_{}'.format(NBR_FEATURE_PREFIX, i, 'text_xf')] = \
tf.io.FixedLenFeature(shape=[HPARAMS.max_seq_length], dtype=tf.int64,
default_value=tf.constant(0, dtype=tf.int64,
shape=[HPARAMS.max_seq_length]))
# The 'NL_num_nbrs' features is currently not used.
# Set the neighbor weight feature keys.
for i in range(num_neighbors):
feature_spec['{}{}{}'.format(NBR_FEATURE_PREFIX, i, NBR_WEIGHT_SUFFIX)] = \
tf.io.FixedLenFeature(shape=[1], dtype=tf.float32, default_value=[0.0])
return feature_spec
def _input_fn(filenames, tf_transform_output, is_training, batch_size=200):
"""Generates features and labels for training or evaluation.
Args:
filenames: [str] list of CSV files to read data from.
tf_transform_output: A TFTransformOutput.
is_training: Boolean indicating if we are in training mode.
batch_size: int First dimension size of the Tensors returned by input_fn
Returns:
A (features, indices) tuple where features is a dictionary of
Tensors, and indices is a single Tensor of label indices.
"""
transformed_feature_spec = (
tf_transform_output.transformed_feature_spec().copy())
# During training, NSL uses augmented training data (which includes features
# from graph neighbors). So, update the feature spec accordingly. This needs
# to be done because we are using different schemas for NSL training and eval,
# but the Trainer Component only accepts a single schema.
if is_training:
transformed_feature_spec =_augment_feature_spec(transformed_feature_spec,
HPARAMS.num_neighbors)
dataset = tf.data.experimental.make_batched_features_dataset(
filenames, batch_size, transformed_feature_spec, reader=_gzip_reader_fn)
transformed_features = tf.compat.v1.data.make_one_shot_iterator(
dataset).get_next()
# We pop the label because we do not want to use it as a feature while we're
# training.
return transformed_features, transformed_features.pop(
_transformed_name(LABEL_KEY))
# TFX will call this function
def trainer_fn(hparams, schema):
"""Build the estimator using the high level API.
Args:
hparams: Holds hyperparameters used to train the model as name/value pairs.
schema: Holds the schema of the training examples.
Returns:
A dict of the following:
- estimator: The estimator that will be used for training and eval.
- train_spec: Spec for training.
- eval_spec: Spec for eval.
- eval_input_receiver_fn: Input function for eval.
"""
train_batch_size = 40
eval_batch_size = 40
tf_transform_output = tft.TFTransformOutput(hparams.transform_output)
train_input_fn = lambda: _input_fn(
hparams.train_files,
tf_transform_output,
is_training=True,
batch_size=train_batch_size)
eval_input_fn = lambda: _input_fn(
hparams.eval_files,
tf_transform_output,
is_training=False,
batch_size=eval_batch_size)
train_spec = tf.estimator.TrainSpec(
train_input_fn,
max_steps=hparams.train_steps)
serving_receiver_fn = lambda: _example_serving_receiver_fn(
tf_transform_output, schema)
exporter = tf.estimator.FinalExporter('imdb', serving_receiver_fn)
eval_spec = tf.estimator.EvalSpec(
eval_input_fn,
steps=hparams.eval_steps,
exporters=[exporter],
name='imdb-eval')
run_config = tf.estimator.RunConfig(
save_checkpoints_steps=999, keep_checkpoint_max=1)
run_config = run_config.replace(model_dir=hparams.serving_model_dir)
estimator = tf.estimator.Estimator(
model_fn=feed_forward_model_fn, config=run_config, params=HPARAMS)
# Create a graph regularization config.
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)
# Invoke the Graph Regularization Estimator wrapper to incorporate
# graph-based regularization for training.
graph_nsl_estimator = nsl.estimator.add_graph_regularization(
estimator,
embedding_fn,
optimizer_fn=optimizer_fn,
graph_reg_config=graph_reg_config)
# Create an input receiver for TFMA processing
receiver_fn = lambda: _eval_input_receiver_fn(
tf_transform_output, schema)
return {
'estimator': graph_nsl_estimator,
'train_spec': train_spec,
'eval_spec': eval_spec,
'eval_input_receiver_fn': receiver_fn
}
Writing imdb_trainer.py
Create and run the Trainer component, passing it the file that we created above.
# Uses user-provided Python function that implements a model using TensorFlow's
# Estimators API.
trainer = Trainer(
module_file=_trainer_module_file,
custom_executor_spec=executor_spec.ExecutorClassSpec(
trainer_executor.Executor),
transformed_examples=graph_augmentation.outputs['augmented_examples'],
schema=schema_gen.outputs['schema'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
context.run(trainer)
WARNING:absl:`custom_executor_spec` is deprecated. Please customize component directly.
WARNING:absl:`transformed_examples` is deprecated. Please use `examples` instead.
running bdist_wheel
running build
running build_py
creating build
creating build/lib
copying imdb_transform.py -> build/lib
copying imdb_trainer.py -> build/lib
installing to /tmp/tmpajjjkpyv
running install
running install_lib
copying build/lib/imdb_trainer.py -> /tmp/tmpajjjkpyv
copying build/lib/imdb_transform.py -> /tmp/tmpajjjkpyv
running install_egg_info
running egg_info
creating tfx_user_code_Trainer.egg-info
writing tfx_user_code_Trainer.egg-info/PKG-INFO
writing dependency_links to tfx_user_code_Trainer.egg-info/dependency_links.txt
/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/setuptools/command/install.py:37: SetuptoolsDeprecationWarning: setup.py install is deprecated. Use build and pip and other standards-based tools.
setuptools.SetuptoolsDeprecationWarning,
I0329 09:16:46.778267 8995 rdbms_metadata_access_object.cc:686] No property is defined for the Type
I0329 09:16:46.781581 8995 rdbms_metadata_access_object.cc:686] No property is defined for the Type
WARNING:absl:Examples artifact does not have payload_format custom property. Falling back to FORMAT_TF_EXAMPLE
WARNING:absl:Examples artifact does not have payload_format custom property. Falling back to FORMAT_TF_EXAMPLE
WARNING:absl:Examples artifact does not have payload_format custom property. Falling back to FORMAT_TF_EXAMPLE
writing top-level names to tfx_user_code_Trainer.egg-info/top_level.txt
writing manifest file 'tfx_user_code_Trainer.egg-info/SOURCES.txt'
reading manifest file 'tfx_user_code_Trainer.egg-info/SOURCES.txt'
writing manifest file 'tfx_user_code_Trainer.egg-info/SOURCES.txt'
Copying tfx_user_code_Trainer.egg-info to /tmp/tmpajjjkpyv/tfx_user_code_Trainer-0.0+b990a2c6a4f23081880867efa3bd3c38db9d7bd0a87a0c9b277ae63714defc8d-py3.7.egg-info
running install_scripts
creating /tmp/tmpajjjkpyv/tfx_user_code_Trainer-0.0+b990a2c6a4f23081880867efa3bd3c38db9d7bd0a87a0c9b277ae63714defc8d.dist-info/WHEEL
creating '/tmp/tmp1454qxd8/tfx_user_code_Trainer-0.0+b990a2c6a4f23081880867efa3bd3c38db9d7bd0a87a0c9b277ae63714defc8d-py3-none-any.whl' and adding '/tmp/tmpajjjkpyv' to it
adding 'imdb_trainer.py'
adding 'imdb_transform.py'
adding 'tfx_user_code_Trainer-0.0+b990a2c6a4f23081880867efa3bd3c38db9d7bd0a87a0c9b277ae63714defc8d.dist-info/METADATA'
adding 'tfx_user_code_Trainer-0.0+b990a2c6a4f23081880867efa3bd3c38db9d7bd0a87a0c9b277ae63714defc8d.dist-info/WHEEL'
adding 'tfx_user_code_Trainer-0.0+b990a2c6a4f23081880867efa3bd3c38db9d7bd0a87a0c9b277ae63714defc8d.dist-info/top_level.txt'
adding 'tfx_user_code_Trainer-0.0+b990a2c6a4f23081880867efa3bd3c38db9d7bd0a87a0c9b277ae63714defc8d.dist-info/RECORD'
removing /tmp/tmpajjjkpyv
Processing /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/_wheels/tfx_user_code_Trainer-0.0+b990a2c6a4f23081880867efa3bd3c38db9d7bd0a87a0c9b277ae63714defc8d-py3-none-any.whl
Installing collected packages: tfx-user-code-Trainer
Successfully installed tfx-user-code-Trainer-0.0+b990a2c6a4f23081880867efa3bd3c38db9d7bd0a87a0c9b277ae63714defc8d
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': 999, '_save_checkpoints_secs': None, '_session_config': allow_soft_placement: true
graph_options {
rewrite_options {
meta_optimizer_iterations: ONE
}
}
, '_keep_checkpoint_max': 1, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_experimental_max_worker_delay_secs': None, '_session_creation_timeout_secs': 7200, '_checkpoint_save_graph_def': True, '_service': None, '_cluster_spec': ClusterSpec({}), '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': 999, '_save_checkpoints_secs': None, '_session_config': allow_soft_placement: true
graph_options {
rewrite_options {
meta_optimizer_iterations: ONE
}
}
, '_keep_checkpoint_max': 1, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_experimental_max_worker_delay_secs': None, '_session_creation_timeout_secs': 7200, '_checkpoint_save_graph_def': True, '_service': None, '_cluster_spec': ClusterSpec({}), '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}
INFO:tensorflow:Not using Distribute Coordinator.
INFO:tensorflow:Not using Distribute Coordinator.
INFO:tensorflow:Running training and evaluation locally (non-distributed).
INFO:tensorflow:Running training and evaluation locally (non-distributed).
INFO:tensorflow:Start train and evaluate loop. The evaluate will happen after every checkpoint. Checkpoint frequency is determined based on RunConfig arguments: save_checkpoints_steps 999 or save_checkpoints_secs None.
INFO:tensorflow:Start train and evaluate loop. The evaluate will happen after every checkpoint. Checkpoint frequency is determined based on RunConfig arguments: save_checkpoints_steps 999 or save_checkpoints_secs None.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/training_util.py:236: Variable.initialized_value (from tensorflow.python.ops.variables) is deprecated and will be removed in a future version.
Instructions for updating:
Use Variable.read_value. Variables in 2.X are initialized automatically both in eager and graph (inside tf.defun) contexts.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/training_util.py:236: Variable.initialized_value (from tensorflow.python.ops.variables) is deprecated and will be removed in a future version.
Instructions for updating:
Use Variable.read_value. Variables in 2.X are initialized automatically both in eager and graph (inside tf.defun) contexts.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/rmsprop.py:123: calling Ones.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.
Instructions for updating:
Call initializer instance with the dtype argument instead of passing it to the constructor
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/rmsprop.py:123: calling Ones.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.
Instructions for updating:
Call initializer instance with the dtype argument instead of passing it to the constructor
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/ops/array_ops.py:5049: calling gather (from tensorflow.python.ops.array_ops) with validate_indices is deprecated and will be removed in a future version.
Instructions for updating:
The `validate_indices` argument has no effect. Indices are always validated on CPU and never validated on GPU.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/ops/array_ops.py:5049: calling gather (from tensorflow.python.ops.array_ops) with validate_indices is deprecated and will be removed in a future version.
Instructions for updating:
The `validate_indices` argument has no effect. Indices are always validated on CPU and never validated on GPU.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 0...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 0...
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 0...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 0...
INFO:tensorflow:loss = 0.692926, step = 0
INFO:tensorflow:loss = 0.692926, step = 0
INFO:tensorflow:global_step/sec: 242.299
INFO:tensorflow:global_step/sec: 242.299
INFO:tensorflow:loss = 0.6923196, step = 100 (0.414 sec)
INFO:tensorflow:loss = 0.6923196, step = 100 (0.414 sec)
INFO:tensorflow:global_step/sec: 313.601
INFO:tensorflow:global_step/sec: 313.601
INFO:tensorflow:loss = 0.69262165, step = 200 (0.319 sec)
INFO:tensorflow:loss = 0.69262165, step = 200 (0.319 sec)
INFO:tensorflow:global_step/sec: 312.868
INFO:tensorflow:global_step/sec: 312.868
INFO:tensorflow:loss = 0.6906095, step = 300 (0.319 sec)
INFO:tensorflow:loss = 0.6906095, step = 300 (0.319 sec)
INFO:tensorflow:global_step/sec: 323.854
INFO:tensorflow:global_step/sec: 323.854
INFO:tensorflow:loss = 0.6918747, step = 400 (0.309 sec)
INFO:tensorflow:loss = 0.6918747, step = 400 (0.309 sec)
INFO:tensorflow:global_step/sec: 322.192
INFO:tensorflow:global_step/sec: 322.192
INFO:tensorflow:loss = 0.6913161, step = 500 (0.310 sec)
INFO:tensorflow:loss = 0.6913161, step = 500 (0.310 sec)
INFO:tensorflow:global_step/sec: 324.039
INFO:tensorflow:global_step/sec: 324.039
INFO:tensorflow:loss = 0.6899584, step = 600 (0.309 sec)
INFO:tensorflow:loss = 0.6899584, step = 600 (0.309 sec)
INFO:tensorflow:global_step/sec: 317.332
INFO:tensorflow:global_step/sec: 317.332
INFO:tensorflow:loss = 0.6885657, step = 700 (0.315 sec)
INFO:tensorflow:loss = 0.6885657, step = 700 (0.315 sec)
INFO:tensorflow:global_step/sec: 320.686
INFO:tensorflow:global_step/sec: 320.686
INFO:tensorflow:loss = 0.68853056, step = 800 (0.312 sec)
INFO:tensorflow:loss = 0.68853056, step = 800 (0.312 sec)
INFO:tensorflow:global_step/sec: 318.538
INFO:tensorflow:global_step/sec: 318.538
INFO:tensorflow:loss = 0.689386, step = 900 (0.314 sec)
INFO:tensorflow:loss = 0.689386, step = 900 (0.314 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 999...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 999...
INFO:tensorflow:Saving checkpoints for 999 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 999 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/saver.py:971: remove_checkpoint (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.
Instructions for updating:
Use standard file APIs to delete files with this prefix.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/training/saver.py:971: remove_checkpoint (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.
Instructions for updating:
Use standard file APIs to delete files with this prefix.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 999...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 999...
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-03-29T09:16:55
INFO:tensorflow:Starting evaluation at 2022-03-29T09:16:55
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-999
INFO:tensorflow:Restoring parameters from /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-999
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [500/5000]
INFO:tensorflow:Evaluation [500/5000]
INFO:tensorflow:Evaluation [1000/5000]
INFO:tensorflow:Evaluation [1000/5000]
INFO:tensorflow:Evaluation [1500/5000]
INFO:tensorflow:Evaluation [1500/5000]
INFO:tensorflow:Evaluation [2000/5000]
INFO:tensorflow:Evaluation [2000/5000]
INFO:tensorflow:Evaluation [2500/5000]
INFO:tensorflow:Evaluation [2500/5000]
INFO:tensorflow:Evaluation [3000/5000]
INFO:tensorflow:Evaluation [3000/5000]
INFO:tensorflow:Evaluation [3500/5000]
INFO:tensorflow:Evaluation [3500/5000]
INFO:tensorflow:Evaluation [4000/5000]
INFO:tensorflow:Evaluation [4000/5000]
INFO:tensorflow:Evaluation [4500/5000]
INFO:tensorflow:Evaluation [4500/5000]
INFO:tensorflow:Evaluation [5000/5000]
INFO:tensorflow:Evaluation [5000/5000]
INFO:tensorflow:Inference Time : 4.95307s
INFO:tensorflow:Inference Time : 4.95307s
INFO:tensorflow:Finished evaluation at 2022-03-29-09:17:00
INFO:tensorflow:Finished evaluation at 2022-03-29-09:17:00
INFO:tensorflow:Saving dict for global step 999: accuracy = 0.6719, global_step = 999, loss = 0.68655944
INFO:tensorflow:Saving dict for global step 999: accuracy = 0.6719, global_step = 999, loss = 0.68655944
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 999: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-999
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 999: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-999
INFO:tensorflow:global_step/sec: 18.2471
INFO:tensorflow:global_step/sec: 18.2471
INFO:tensorflow:loss = 0.68463147, step = 1000 (5.480 sec)
INFO:tensorflow:loss = 0.68463147, step = 1000 (5.480 sec)
INFO:tensorflow:global_step/sec: 318.613
INFO:tensorflow:global_step/sec: 318.613
INFO:tensorflow:loss = 0.6890757, step = 1100 (0.314 sec)
INFO:tensorflow:loss = 0.6890757, step = 1100 (0.314 sec)
INFO:tensorflow:global_step/sec: 322.339
INFO:tensorflow:global_step/sec: 322.339
INFO:tensorflow:loss = 0.68656504, step = 1200 (0.310 sec)
INFO:tensorflow:loss = 0.68656504, step = 1200 (0.310 sec)
INFO:tensorflow:global_step/sec: 323.843
INFO:tensorflow:global_step/sec: 323.843
INFO:tensorflow:loss = 0.676161, step = 1300 (0.309 sec)
INFO:tensorflow:loss = 0.676161, step = 1300 (0.309 sec)
INFO:tensorflow:global_step/sec: 325.691
INFO:tensorflow:global_step/sec: 325.691
INFO:tensorflow:loss = 0.6784999, step = 1400 (0.307 sec)
INFO:tensorflow:loss = 0.6784999, step = 1400 (0.307 sec)
INFO:tensorflow:global_step/sec: 316.657
INFO:tensorflow:global_step/sec: 316.657
INFO:tensorflow:loss = 0.67178994, step = 1500 (0.316 sec)
INFO:tensorflow:loss = 0.67178994, step = 1500 (0.316 sec)
INFO:tensorflow:global_step/sec: 327.406
INFO:tensorflow:global_step/sec: 327.406
INFO:tensorflow:loss = 0.6722702, step = 1600 (0.305 sec)
INFO:tensorflow:loss = 0.6722702, step = 1600 (0.305 sec)
INFO:tensorflow:global_step/sec: 316.891
INFO:tensorflow:global_step/sec: 316.891
INFO:tensorflow:loss = 0.6667036, step = 1700 (0.315 sec)
INFO:tensorflow:loss = 0.6667036, step = 1700 (0.315 sec)
INFO:tensorflow:global_step/sec: 314.255
INFO:tensorflow:global_step/sec: 314.255
INFO:tensorflow:loss = 0.6751104, step = 1800 (0.318 sec)
INFO:tensorflow:loss = 0.6751104, step = 1800 (0.318 sec)
INFO:tensorflow:global_step/sec: 327.995
INFO:tensorflow:global_step/sec: 327.995
INFO:tensorflow:loss = 0.67359537, step = 1900 (0.305 sec)
INFO:tensorflow:loss = 0.67359537, step = 1900 (0.305 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 1998...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 1998...
INFO:tensorflow:Saving checkpoints for 1998 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 1998 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 1998...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 1998...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:global_step/sec: 271.709
INFO:tensorflow:global_step/sec: 271.709
INFO:tensorflow:loss = 0.6581868, step = 2000 (0.368 sec)
INFO:tensorflow:loss = 0.6581868, step = 2000 (0.368 sec)
INFO:tensorflow:global_step/sec: 318.494
INFO:tensorflow:global_step/sec: 318.494
INFO:tensorflow:loss = 0.6538587, step = 2100 (0.314 sec)
INFO:tensorflow:loss = 0.6538587, step = 2100 (0.314 sec)
INFO:tensorflow:global_step/sec: 327.953
INFO:tensorflow:global_step/sec: 327.953
INFO:tensorflow:loss = 0.66053313, step = 2200 (0.305 sec)
INFO:tensorflow:loss = 0.66053313, step = 2200 (0.305 sec)
INFO:tensorflow:global_step/sec: 316.355
INFO:tensorflow:global_step/sec: 316.355
INFO:tensorflow:loss = 0.66581553, step = 2300 (0.316 sec)
INFO:tensorflow:loss = 0.66581553, step = 2300 (0.316 sec)
INFO:tensorflow:global_step/sec: 323.629
INFO:tensorflow:global_step/sec: 323.629
INFO:tensorflow:loss = 0.6656166, step = 2400 (0.309 sec)
INFO:tensorflow:loss = 0.6656166, step = 2400 (0.309 sec)
INFO:tensorflow:global_step/sec: 326.794
INFO:tensorflow:global_step/sec: 326.794
INFO:tensorflow:loss = 0.6519306, step = 2500 (0.306 sec)
INFO:tensorflow:loss = 0.6519306, step = 2500 (0.306 sec)
INFO:tensorflow:global_step/sec: 326.264
INFO:tensorflow:global_step/sec: 326.264
INFO:tensorflow:loss = 0.6646646, step = 2600 (0.307 sec)
INFO:tensorflow:loss = 0.6646646, step = 2600 (0.307 sec)
INFO:tensorflow:global_step/sec: 322.66
INFO:tensorflow:global_step/sec: 322.66
INFO:tensorflow:loss = 0.6505408, step = 2700 (0.310 sec)
INFO:tensorflow:loss = 0.6505408, step = 2700 (0.310 sec)
INFO:tensorflow:global_step/sec: 310.665
INFO:tensorflow:global_step/sec: 310.665
INFO:tensorflow:loss = 0.657897, step = 2800 (0.322 sec)
INFO:tensorflow:loss = 0.657897, step = 2800 (0.322 sec)
INFO:tensorflow:global_step/sec: 322.767
INFO:tensorflow:global_step/sec: 322.767
INFO:tensorflow:loss = 0.64274955, step = 2900 (0.310 sec)
INFO:tensorflow:loss = 0.64274955, step = 2900 (0.310 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 2997...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 2997...
INFO:tensorflow:Saving checkpoints for 2997 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 2997 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 2997...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 2997...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:global_step/sec: 275.463
INFO:tensorflow:global_step/sec: 275.463
INFO:tensorflow:loss = 0.6232832, step = 3000 (0.363 sec)
INFO:tensorflow:loss = 0.6232832, step = 3000 (0.363 sec)
INFO:tensorflow:global_step/sec: 322.71
INFO:tensorflow:global_step/sec: 322.71
INFO:tensorflow:loss = 0.64490813, step = 3100 (0.310 sec)
INFO:tensorflow:loss = 0.64490813, step = 3100 (0.310 sec)
INFO:tensorflow:global_step/sec: 323.342
INFO:tensorflow:global_step/sec: 323.342
INFO:tensorflow:loss = 0.6196912, step = 3200 (0.309 sec)
INFO:tensorflow:loss = 0.6196912, step = 3200 (0.309 sec)
INFO:tensorflow:global_step/sec: 308.915
INFO:tensorflow:global_step/sec: 308.915
INFO:tensorflow:loss = 0.5888808, step = 3300 (0.324 sec)
INFO:tensorflow:loss = 0.5888808, step = 3300 (0.324 sec)
INFO:tensorflow:global_step/sec: 323.694
INFO:tensorflow:global_step/sec: 323.694
INFO:tensorflow:loss = 0.6377993, step = 3400 (0.309 sec)
INFO:tensorflow:loss = 0.6377993, step = 3400 (0.309 sec)
INFO:tensorflow:global_step/sec: 323.834
INFO:tensorflow:global_step/sec: 323.834
INFO:tensorflow:loss = 0.5785633, step = 3500 (0.309 sec)
INFO:tensorflow:loss = 0.5785633, step = 3500 (0.309 sec)
INFO:tensorflow:global_step/sec: 318.644
INFO:tensorflow:global_step/sec: 318.644
INFO:tensorflow:loss = 0.6006882, step = 3600 (0.314 sec)
INFO:tensorflow:loss = 0.6006882, step = 3600 (0.314 sec)
INFO:tensorflow:global_step/sec: 322.295
INFO:tensorflow:global_step/sec: 322.295
INFO:tensorflow:loss = 0.6115326, step = 3700 (0.310 sec)
INFO:tensorflow:loss = 0.6115326, step = 3700 (0.310 sec)
INFO:tensorflow:global_step/sec: 320.336
INFO:tensorflow:global_step/sec: 320.336
INFO:tensorflow:loss = 0.6047396, step = 3800 (0.312 sec)
INFO:tensorflow:loss = 0.6047396, step = 3800 (0.312 sec)
INFO:tensorflow:global_step/sec: 324.363
INFO:tensorflow:global_step/sec: 324.363
INFO:tensorflow:loss = 0.5679431, step = 3900 (0.308 sec)
INFO:tensorflow:loss = 0.5679431, step = 3900 (0.308 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 3996...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 3996...
INFO:tensorflow:Saving checkpoints for 3996 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 3996 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 3996...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 3996...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:global_step/sec: 272.702
INFO:tensorflow:global_step/sec: 272.702
INFO:tensorflow:loss = 0.6215052, step = 4000 (0.367 sec)
INFO:tensorflow:loss = 0.6215052, step = 4000 (0.367 sec)
INFO:tensorflow:global_step/sec: 302.827
INFO:tensorflow:global_step/sec: 302.827
INFO:tensorflow:loss = 0.6152085, step = 4100 (0.330 sec)
INFO:tensorflow:loss = 0.6152085, step = 4100 (0.330 sec)
INFO:tensorflow:global_step/sec: 322.167
INFO:tensorflow:global_step/sec: 322.167
INFO:tensorflow:loss = 0.5864973, step = 4200 (0.310 sec)
INFO:tensorflow:loss = 0.5864973, step = 4200 (0.310 sec)
INFO:tensorflow:global_step/sec: 312.863
INFO:tensorflow:global_step/sec: 312.863
INFO:tensorflow:loss = 0.57598597, step = 4300 (0.320 sec)
INFO:tensorflow:loss = 0.57598597, step = 4300 (0.320 sec)
INFO:tensorflow:global_step/sec: 324.834
INFO:tensorflow:global_step/sec: 324.834
INFO:tensorflow:loss = 0.55068254, step = 4400 (0.308 sec)
INFO:tensorflow:loss = 0.55068254, step = 4400 (0.308 sec)
INFO:tensorflow:global_step/sec: 319.41
INFO:tensorflow:global_step/sec: 319.41
INFO:tensorflow:loss = 0.581579, step = 4500 (0.313 sec)
INFO:tensorflow:loss = 0.581579, step = 4500 (0.313 sec)
INFO:tensorflow:global_step/sec: 323.985
INFO:tensorflow:global_step/sec: 323.985
INFO:tensorflow:loss = 0.5446909, step = 4600 (0.309 sec)
INFO:tensorflow:loss = 0.5446909, step = 4600 (0.309 sec)
INFO:tensorflow:global_step/sec: 324.026
INFO:tensorflow:global_step/sec: 324.026
INFO:tensorflow:loss = 0.58940125, step = 4700 (0.309 sec)
INFO:tensorflow:loss = 0.58940125, step = 4700 (0.309 sec)
INFO:tensorflow:global_step/sec: 312.769
INFO:tensorflow:global_step/sec: 312.769
INFO:tensorflow:loss = 0.57637256, step = 4800 (0.320 sec)
INFO:tensorflow:loss = 0.57637256, step = 4800 (0.320 sec)
INFO:tensorflow:global_step/sec: 319.88
INFO:tensorflow:global_step/sec: 319.88
INFO:tensorflow:loss = 0.5708374, step = 4900 (0.313 sec)
INFO:tensorflow:loss = 0.5708374, step = 4900 (0.313 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 4995...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 4995...
INFO:tensorflow:Saving checkpoints for 4995 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 4995 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 4995...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 4995...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:global_step/sec: 273.217
INFO:tensorflow:global_step/sec: 273.217
INFO:tensorflow:loss = 0.5555646, step = 5000 (0.366 sec)
INFO:tensorflow:loss = 0.5555646, step = 5000 (0.366 sec)
INFO:tensorflow:global_step/sec: 323.747
INFO:tensorflow:global_step/sec: 323.747
INFO:tensorflow:loss = 0.51535964, step = 5100 (0.309 sec)
INFO:tensorflow:loss = 0.51535964, step = 5100 (0.309 sec)
INFO:tensorflow:global_step/sec: 322.782
INFO:tensorflow:global_step/sec: 322.782
INFO:tensorflow:loss = 0.5341888, step = 5200 (0.310 sec)
INFO:tensorflow:loss = 0.5341888, step = 5200 (0.310 sec)
INFO:tensorflow:global_step/sec: 324.091
INFO:tensorflow:global_step/sec: 324.091
INFO:tensorflow:loss = 0.57570624, step = 5300 (0.309 sec)
INFO:tensorflow:loss = 0.57570624, step = 5300 (0.309 sec)
INFO:tensorflow:global_step/sec: 323.875
INFO:tensorflow:global_step/sec: 323.875
INFO:tensorflow:loss = 0.4591224, step = 5400 (0.309 sec)
INFO:tensorflow:loss = 0.4591224, step = 5400 (0.309 sec)
INFO:tensorflow:global_step/sec: 322.337
INFO:tensorflow:global_step/sec: 322.337
INFO:tensorflow:loss = 0.48855156, step = 5500 (0.310 sec)
INFO:tensorflow:loss = 0.48855156, step = 5500 (0.310 sec)
INFO:tensorflow:global_step/sec: 322.381
INFO:tensorflow:global_step/sec: 322.381
INFO:tensorflow:loss = 0.55077356, step = 5600 (0.310 sec)
INFO:tensorflow:loss = 0.55077356, step = 5600 (0.310 sec)
INFO:tensorflow:global_step/sec: 323.824
INFO:tensorflow:global_step/sec: 323.824
INFO:tensorflow:loss = 0.47171304, step = 5700 (0.309 sec)
INFO:tensorflow:loss = 0.47171304, step = 5700 (0.309 sec)
INFO:tensorflow:global_step/sec: 325.481
INFO:tensorflow:global_step/sec: 325.481
INFO:tensorflow:loss = 0.5261248, step = 5800 (0.307 sec)
INFO:tensorflow:loss = 0.5261248, step = 5800 (0.307 sec)
INFO:tensorflow:global_step/sec: 322.044
INFO:tensorflow:global_step/sec: 322.044
INFO:tensorflow:loss = 0.4494802, step = 5900 (0.311 sec)
INFO:tensorflow:loss = 0.4494802, step = 5900 (0.311 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 5994...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 5994...
INFO:tensorflow:Saving checkpoints for 5994 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 5994 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 5994...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 5994...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:global_step/sec: 274.823
INFO:tensorflow:global_step/sec: 274.823
INFO:tensorflow:loss = 0.47958514, step = 6000 (0.364 sec)
INFO:tensorflow:loss = 0.47958514, step = 6000 (0.364 sec)
INFO:tensorflow:global_step/sec: 316.528
INFO:tensorflow:global_step/sec: 316.528
INFO:tensorflow:loss = 0.57186216, step = 6100 (0.316 sec)
INFO:tensorflow:loss = 0.57186216, step = 6100 (0.316 sec)
INFO:tensorflow:global_step/sec: 314.289
INFO:tensorflow:global_step/sec: 314.289
INFO:tensorflow:loss = 0.48464423, step = 6200 (0.318 sec)
INFO:tensorflow:loss = 0.48464423, step = 6200 (0.318 sec)
INFO:tensorflow:global_step/sec: 325.432
INFO:tensorflow:global_step/sec: 325.432
INFO:tensorflow:loss = 0.44150937, step = 6300 (0.307 sec)
INFO:tensorflow:loss = 0.44150937, step = 6300 (0.307 sec)
INFO:tensorflow:global_step/sec: 325.726
INFO:tensorflow:global_step/sec: 325.726
INFO:tensorflow:loss = 0.5231165, step = 6400 (0.307 sec)
INFO:tensorflow:loss = 0.5231165, step = 6400 (0.307 sec)
INFO:tensorflow:global_step/sec: 323.035
INFO:tensorflow:global_step/sec: 323.035
INFO:tensorflow:loss = 0.5252115, step = 6500 (0.310 sec)
INFO:tensorflow:loss = 0.5252115, step = 6500 (0.310 sec)
INFO:tensorflow:global_step/sec: 318.26
INFO:tensorflow:global_step/sec: 318.26
INFO:tensorflow:loss = 0.4332117, step = 6600 (0.314 sec)
INFO:tensorflow:loss = 0.4332117, step = 6600 (0.314 sec)
INFO:tensorflow:global_step/sec: 328.609
INFO:tensorflow:global_step/sec: 328.609
INFO:tensorflow:loss = 0.45119706, step = 6700 (0.304 sec)
INFO:tensorflow:loss = 0.45119706, step = 6700 (0.304 sec)
INFO:tensorflow:global_step/sec: 309.263
INFO:tensorflow:global_step/sec: 309.263
INFO:tensorflow:loss = 0.5138131, step = 6800 (0.323 sec)
INFO:tensorflow:loss = 0.5138131, step = 6800 (0.323 sec)
INFO:tensorflow:global_step/sec: 320.628
INFO:tensorflow:global_step/sec: 320.628
INFO:tensorflow:loss = 0.4461824, step = 6900 (0.312 sec)
INFO:tensorflow:loss = 0.4461824, step = 6900 (0.312 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 6993...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 6993...
INFO:tensorflow:Saving checkpoints for 6993 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 6993 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 6993...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 6993...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:global_step/sec: 274.917
INFO:tensorflow:global_step/sec: 274.917
INFO:tensorflow:loss = 0.52072084, step = 7000 (0.364 sec)
INFO:tensorflow:loss = 0.52072084, step = 7000 (0.364 sec)
INFO:tensorflow:global_step/sec: 317.325
INFO:tensorflow:global_step/sec: 317.325
INFO:tensorflow:loss = 0.42709062, step = 7100 (0.315 sec)
INFO:tensorflow:loss = 0.42709062, step = 7100 (0.315 sec)
INFO:tensorflow:global_step/sec: 324.052
INFO:tensorflow:global_step/sec: 324.052
INFO:tensorflow:loss = 0.43844005, step = 7200 (0.308 sec)
INFO:tensorflow:loss = 0.43844005, step = 7200 (0.308 sec)
INFO:tensorflow:global_step/sec: 325.318
INFO:tensorflow:global_step/sec: 325.318
INFO:tensorflow:loss = 0.45420846, step = 7300 (0.307 sec)
INFO:tensorflow:loss = 0.45420846, step = 7300 (0.307 sec)
INFO:tensorflow:global_step/sec: 314.86
INFO:tensorflow:global_step/sec: 314.86
INFO:tensorflow:loss = 0.5319026, step = 7400 (0.318 sec)
INFO:tensorflow:loss = 0.5319026, step = 7400 (0.318 sec)
INFO:tensorflow:global_step/sec: 319.616
INFO:tensorflow:global_step/sec: 319.616
INFO:tensorflow:loss = 0.5418476, step = 7500 (0.313 sec)
INFO:tensorflow:loss = 0.5418476, step = 7500 (0.313 sec)
INFO:tensorflow:global_step/sec: 328.495
INFO:tensorflow:global_step/sec: 328.495
INFO:tensorflow:loss = 0.39630955, step = 7600 (0.304 sec)
INFO:tensorflow:loss = 0.39630955, step = 7600 (0.304 sec)
INFO:tensorflow:global_step/sec: 327.694
INFO:tensorflow:global_step/sec: 327.694
INFO:tensorflow:loss = 0.5574594, step = 7700 (0.305 sec)
INFO:tensorflow:loss = 0.5574594, step = 7700 (0.305 sec)
INFO:tensorflow:global_step/sec: 325.202
INFO:tensorflow:global_step/sec: 325.202
INFO:tensorflow:loss = 0.34146416, step = 7800 (0.308 sec)
INFO:tensorflow:loss = 0.34146416, step = 7800 (0.308 sec)
INFO:tensorflow:global_step/sec: 322.032
INFO:tensorflow:global_step/sec: 322.032
INFO:tensorflow:loss = 0.46923068, step = 7900 (0.311 sec)
INFO:tensorflow:loss = 0.46923068, step = 7900 (0.311 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 7992...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 7992...
INFO:tensorflow:Saving checkpoints for 7992 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 7992 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 7992...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 7992...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:global_step/sec: 271.09
INFO:tensorflow:global_step/sec: 271.09
INFO:tensorflow:loss = 0.42599657, step = 8000 (0.369 sec)
INFO:tensorflow:loss = 0.42599657, step = 8000 (0.369 sec)
INFO:tensorflow:global_step/sec: 322.534
INFO:tensorflow:global_step/sec: 322.534
INFO:tensorflow:loss = 0.41598743, step = 8100 (0.310 sec)
INFO:tensorflow:loss = 0.41598743, step = 8100 (0.310 sec)
INFO:tensorflow:global_step/sec: 322.699
INFO:tensorflow:global_step/sec: 322.699
INFO:tensorflow:loss = 0.35861862, step = 8200 (0.310 sec)
INFO:tensorflow:loss = 0.35861862, step = 8200 (0.310 sec)
INFO:tensorflow:global_step/sec: 312.485
INFO:tensorflow:global_step/sec: 312.485
INFO:tensorflow:loss = 0.40294614, step = 8300 (0.320 sec)
INFO:tensorflow:loss = 0.40294614, step = 8300 (0.320 sec)
INFO:tensorflow:global_step/sec: 314.361
INFO:tensorflow:global_step/sec: 314.361
INFO:tensorflow:loss = 0.4703371, step = 8400 (0.318 sec)
INFO:tensorflow:loss = 0.4703371, step = 8400 (0.318 sec)
INFO:tensorflow:global_step/sec: 322.03
INFO:tensorflow:global_step/sec: 322.03
INFO:tensorflow:loss = 0.31964254, step = 8500 (0.311 sec)
INFO:tensorflow:loss = 0.31964254, step = 8500 (0.311 sec)
INFO:tensorflow:global_step/sec: 320.946
INFO:tensorflow:global_step/sec: 320.946
INFO:tensorflow:loss = 0.41026768, step = 8600 (0.311 sec)
INFO:tensorflow:loss = 0.41026768, step = 8600 (0.311 sec)
INFO:tensorflow:global_step/sec: 326.351
INFO:tensorflow:global_step/sec: 326.351
INFO:tensorflow:loss = 0.47896525, step = 8700 (0.306 sec)
INFO:tensorflow:loss = 0.47896525, step = 8700 (0.306 sec)
INFO:tensorflow:global_step/sec: 328.483
INFO:tensorflow:global_step/sec: 328.483
INFO:tensorflow:loss = 0.41821286, step = 8800 (0.304 sec)
INFO:tensorflow:loss = 0.41821286, step = 8800 (0.304 sec)
INFO:tensorflow:global_step/sec: 325.848
INFO:tensorflow:global_step/sec: 325.848
INFO:tensorflow:loss = 0.35640258, step = 8900 (0.307 sec)
INFO:tensorflow:loss = 0.35640258, step = 8900 (0.307 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 8991...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 8991...
INFO:tensorflow:Saving checkpoints for 8991 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 8991 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 8991...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 8991...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:global_step/sec: 273.453
INFO:tensorflow:global_step/sec: 273.453
INFO:tensorflow:loss = 0.39678425, step = 9000 (0.365 sec)
INFO:tensorflow:loss = 0.39678425, step = 9000 (0.365 sec)
INFO:tensorflow:global_step/sec: 314.627
INFO:tensorflow:global_step/sec: 314.627
INFO:tensorflow:loss = 0.5041427, step = 9100 (0.318 sec)
INFO:tensorflow:loss = 0.5041427, step = 9100 (0.318 sec)
INFO:tensorflow:global_step/sec: 322.689
INFO:tensorflow:global_step/sec: 322.689
INFO:tensorflow:loss = 0.3935093, step = 9200 (0.310 sec)
INFO:tensorflow:loss = 0.3935093, step = 9200 (0.310 sec)
INFO:tensorflow:global_step/sec: 322.043
INFO:tensorflow:global_step/sec: 322.043
INFO:tensorflow:loss = 0.4669891, step = 9300 (0.311 sec)
INFO:tensorflow:loss = 0.4669891, step = 9300 (0.311 sec)
INFO:tensorflow:global_step/sec: 323.42
INFO:tensorflow:global_step/sec: 323.42
INFO:tensorflow:loss = 0.4558824, step = 9400 (0.309 sec)
INFO:tensorflow:loss = 0.4558824, step = 9400 (0.309 sec)
INFO:tensorflow:global_step/sec: 322.661
INFO:tensorflow:global_step/sec: 322.661
INFO:tensorflow:loss = 0.41133726, step = 9500 (0.310 sec)
INFO:tensorflow:loss = 0.41133726, step = 9500 (0.310 sec)
INFO:tensorflow:global_step/sec: 319.962
INFO:tensorflow:global_step/sec: 319.962
INFO:tensorflow:loss = 0.36401567, step = 9600 (0.313 sec)
INFO:tensorflow:loss = 0.36401567, step = 9600 (0.313 sec)
INFO:tensorflow:global_step/sec: 321.068
INFO:tensorflow:global_step/sec: 321.068
INFO:tensorflow:loss = 0.5478149, step = 9700 (0.312 sec)
INFO:tensorflow:loss = 0.5478149, step = 9700 (0.312 sec)
INFO:tensorflow:global_step/sec: 329.709
INFO:tensorflow:global_step/sec: 329.709
INFO:tensorflow:loss = 0.47268537, step = 9800 (0.303 sec)
INFO:tensorflow:loss = 0.47268537, step = 9800 (0.303 sec)
INFO:tensorflow:global_step/sec: 321.553
INFO:tensorflow:global_step/sec: 321.553
INFO:tensorflow:loss = 0.42842835, step = 9900 (0.311 sec)
INFO:tensorflow:loss = 0.42842835, step = 9900 (0.311 sec)
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 9990...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 9990...
INFO:tensorflow:Saving checkpoints for 9990 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 9990 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 9990...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 9990...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 10000...
INFO:tensorflow:Calling checkpoint listeners before saving checkpoint 10000...
INFO:tensorflow:Saving checkpoints for 10000 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Saving checkpoints for 10000 into /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt.
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 10000...
INFO:tensorflow:Calling checkpoint listeners after saving checkpoint 10000...
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Skip the current checkpoint eval due to throttle secs (600 secs).
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2022-03-29T09:17:29
INFO:tensorflow:Starting evaluation at 2022-03-29T09:17:29
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-10000
INFO:tensorflow:Restoring parameters from /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-10000
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Evaluation [500/5000]
INFO:tensorflow:Evaluation [500/5000]
INFO:tensorflow:Evaluation [1000/5000]
INFO:tensorflow:Evaluation [1000/5000]
INFO:tensorflow:Evaluation [1500/5000]
INFO:tensorflow:Evaluation [1500/5000]
INFO:tensorflow:Evaluation [2000/5000]
INFO:tensorflow:Evaluation [2000/5000]
INFO:tensorflow:Evaluation [2500/5000]
INFO:tensorflow:Evaluation [2500/5000]
INFO:tensorflow:Evaluation [3000/5000]
INFO:tensorflow:Evaluation [3000/5000]
INFO:tensorflow:Evaluation [3500/5000]
INFO:tensorflow:Evaluation [3500/5000]
INFO:tensorflow:Evaluation [4000/5000]
INFO:tensorflow:Evaluation [4000/5000]
INFO:tensorflow:Evaluation [4500/5000]
INFO:tensorflow:Evaluation [4500/5000]
INFO:tensorflow:Evaluation [5000/5000]
INFO:tensorflow:Evaluation [5000/5000]
INFO:tensorflow:Inference Time : 4.98955s
INFO:tensorflow:Inference Time : 4.98955s
INFO:tensorflow:Finished evaluation at 2022-03-29-09:17:34
INFO:tensorflow:Finished evaluation at 2022-03-29-09:17:34
INFO:tensorflow:Saving dict for global step 10000: accuracy = 0.8024, global_step = 10000, loss = 0.4427534
INFO:tensorflow:Saving dict for global step 10000: accuracy = 0.8024, global_step = 10000, loss = 0.4427534
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 10000: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-10000
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 10000: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-10000
INFO:tensorflow:Performing the final export in the end of training.
INFO:tensorflow:Performing the final export in the end of training.
WARNING:tensorflow:Loading a TF2 SavedModel but eager mode seems disabled.
WARNING:tensorflow:Loading a TF2 SavedModel but eager mode seems disabled.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/saved_model/signature_def_utils_impl.py:201: build_tensor_info (from tensorflow.python.saved_model.utils_impl) is deprecated and will be removed in a future version.
Instructions for updating:
This function will only be available through the v1 compatibility library as tf.compat.v1.saved_model.utils.build_tensor_info or tf.compat.v1.saved_model.build_tensor_info.
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tensorflow/python/saved_model/signature_def_utils_impl.py:201: build_tensor_info (from tensorflow.python.saved_model.utils_impl) is deprecated and will be removed in a future version.
Instructions for updating:
This function will only be available through the v1 compatibility library as tf.compat.v1.saved_model.utils.build_tensor_info or tf.compat.v1.saved_model.build_tensor_info.
INFO:tensorflow:Signatures INCLUDED in export for Classify: None
INFO:tensorflow:Signatures INCLUDED in export for Classify: None
INFO:tensorflow:Signatures INCLUDED in export for Regress: None
INFO:tensorflow:Signatures INCLUDED in export for Regress: None
INFO:tensorflow:Signatures INCLUDED in export for Predict: ['serving_default']
INFO:tensorflow:Signatures INCLUDED in export for Predict: ['serving_default']
INFO:tensorflow:Signatures INCLUDED in export for Train: None
INFO:tensorflow:Signatures INCLUDED in export for Train: None
INFO:tensorflow:Signatures INCLUDED in export for Eval: None
INFO:tensorflow:Signatures INCLUDED in export for Eval: None
INFO:tensorflow:Restoring parameters from /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-10000
INFO:tensorflow:Restoring parameters from /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-10000
INFO:tensorflow:Assets added to graph.
INFO:tensorflow:Assets added to graph.
INFO:tensorflow:Assets written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/export/imdb/temp-1648545454/assets
INFO:tensorflow:Assets written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/export/imdb/temp-1648545454/assets
INFO:tensorflow:SavedModel written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/export/imdb/temp-1648545454/saved_model.pb
INFO:tensorflow:SavedModel written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/export/imdb/temp-1648545454/saved_model.pb
INFO:tensorflow:Loss for final step: 0.39912862.
INFO:tensorflow:Loss for final step: 0.39912862.
WARNING:tensorflow:Loading a TF2 SavedModel but eager mode seems disabled.
WARNING:tensorflow:Loading a TF2 SavedModel but eager mode seems disabled.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Signatures INCLUDED in export for Classify: None
INFO:tensorflow:Signatures INCLUDED in export for Classify: None
INFO:tensorflow:Signatures INCLUDED in export for Regress: None
INFO:tensorflow:Signatures INCLUDED in export for Regress: None
INFO:tensorflow:Signatures INCLUDED in export for Predict: None
INFO:tensorflow:Signatures INCLUDED in export for Predict: None
INFO:tensorflow:Signatures INCLUDED in export for Train: None
INFO:tensorflow:Signatures INCLUDED in export for Train: None
INFO:tensorflow:Signatures INCLUDED in export for Eval: ['eval']
INFO:tensorflow:Signatures INCLUDED in export for Eval: ['eval']
WARNING:tensorflow:Export includes no default signature!
WARNING:tensorflow:Export includes no default signature!
INFO:tensorflow:Restoring parameters from /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-10000
INFO:tensorflow:Restoring parameters from /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-Serving/model.ckpt-10000
INFO:tensorflow:Assets added to graph.
INFO:tensorflow:Assets added to graph.
INFO:tensorflow:Assets written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-TFMA/temp-1648545455/assets
INFO:tensorflow:Assets written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-TFMA/temp-1648545455/assets
INFO:tensorflow:SavedModel written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-TFMA/temp-1648545455/saved_model.pb
INFO:tensorflow:SavedModel written to: /tmp/tfx-interactive-2022-03-29T09_14_16.111420-x6kmk49t/Trainer/model_run/9/Format-TFMA/temp-1648545455/saved_model.pb
WARNING:absl:Support for estimator-based executor and model export will be deprecated soon. Please use export structure <ModelExportPath>/serving_model_dir/saved_model.pb"
WARNING:absl:Support for estimator-based executor and model export will be deprecated soon. Please use export structure <ModelExportPath>/eval_model_dir/saved_model.pb"
Take a peek at the trained model which was exported from Trainer.
train_uri = trainer.outputs['model'].get()[0].uri
serving_model_path = os.path.join(train_uri, 'Format-Serving')
exported_model = tf.saved_model.load(serving_model_path)
exported_model.graph.get_operations()[:10] + ["..."]
[<tf.Operation 'global_step/Initializer/zeros' type=Const>, <tf.Operation 'global_step' type=VarHandleOp>, <tf.Operation 'global_step/IsInitialized/VarIsInitializedOp' type=VarIsInitializedOp>, <tf.Operation 'global_step/Assign' type=AssignVariableOp>, <tf.Operation 'global_step/Read/ReadVariableOp' type=ReadVariableOp>, <tf.Operation 'input_example_tensor' type=Placeholder>, <tf.Operation 'ParseExample/ParseExampleV2/names' type=Const>, <tf.Operation 'ParseExample/ParseExampleV2/sparse_keys' type=Const>, <tf.Operation 'ParseExample/ParseExampleV2/dense_keys' type=Const>, <tf.Operation 'ParseExample/ParseExampleV2/ragged_keys' type=Const>, '...']
Let's visualize the model's metrics using Tensorboard.
#docs_infra: no_execute
# Get the URI of the output artifact representing the training logs,
# which is a directory
model_run_dir = trainer.outputs['model_run'].get()[0].uri
%load_ext tensorboard
%tensorboard --logdir {model_run_dir}
Model Serving
Graph regularization only affects the training workflow by adding a regularization term to the loss function. As a result, the model evaluation and serving workflows remain unchanged. It is for the same reason that we've also omitted downstream TFX components that typically come after the Trainer component like the Evaluator, Pusher, etc.
Conclusion
We have demonstrated the use of graph regularization using the Neural Structured Learning (NSL) framework in a TFX pipeline even when the input does not contain an explicit graph. We considered the task of sentiment classification of IMDB movie reviews for which we synthesized a similarity graph based on review embeddings. We encourage users to experiment further by using different embeddings for graph construction, varying hyperparameters, changing the amount of supervision, and by defining different model architectures.