ExampleGen TFX 파이프라인 구성 요소

from tfx.utils.dsl_utils import csv_input
from tfx.components.example_gen.csv_example_gen.component import CsvExampleGen

examples = csv_input(os.path.join(base_dir, 'data/simple'))
example_gen = CsvExampleGen(input=examples)

from tfx.utils.dsl_utils import tfrecord_input
from tfx.components.example_gen.import_example_gen.component import ImportExampleGen

examples = tfrecord_input(path_to_tfrecord_dir)
example_gen = ImportExampleGen(input=examples)

from  tfx.proto import example_gen_pb2

# Input has a single split 'input_dir/*'.
# Output 2 splits: train:eval=3:1.
output = example_gen_pb2.Output(
             split_config=example_gen_pb2.SplitConfig(splits=[
                 example_gen_pb2.SplitConfig.Split(name='train', hash_buckets=3),
                 example_gen_pb2.SplitConfig.Split(name='eval', hash_buckets=1)
             ]))
examples = csv_input(input_dir)
example_gen = CsvExampleGen(input=examples, output_config=output)

from  tfx.proto import example_gen_pb2

# Input train split is 'input_dir/train/*', eval split is 'input_dir/eval/*'.
# Output splits are generated one-to-one mapping from input splits.
input = example_gen_pb2.Input(splits=[
                example_gen_pb2.Input.Split(name='train', pattern='train/*'),
                example_gen_pb2.Input.Split(name='eval', pattern='eval/*')
            ])
examples = csv_input(input_dir)
example_gen = CsvExampleGen(input=examples, input_config=input)

from  tfx.proto import example_gen_pb2

# Input has a single split 'input_dir/*'.
# Output 2 splits based on 'user_id' features: train:eval=3:1.
output = example_gen_pb2.Output(
             split_config=example_gen_pb2.SplitConfig(splits=[
                 example_gen_pb2.SplitConfig.Split(name='train', hash_buckets=3),
                 example_gen_pb2.SplitConfig.Split(name='eval', hash_buckets=1)
             ],
             partition_feature_name='user_id'))
examples = csv_input(input_dir)
example_gen = CsvExampleGen(input=examples, output_config=output)

splits {
  name: 'train'
  pattern: 'span-{SPAN}/train/*'
}
splits {
  name: 'eval'
  pattern: 'span-{SPAN}/eval/*'
}

from  tfx.proto import example_gen_pb2

input = example_gen_pb2.Input(splits=[
                example_gen_pb2.Input.Split(name='train',
                                            pattern='span-{SPAN}/train/*'),
                example_gen_pb2.Input.Split(name='eval',
                                            pattern='span-{SPAN}/eval/*')
            ])
examples = csv_input('/tmp')
example_gen = CsvExampleGen(input=examples, input_config=input)

splits {
  name: 'train'
  pattern: '{YYYY}-{MM}-{DD}/train/*'
}
splits {
  name: 'eval'
  pattern: '{YYYY}-{MM}-{DD}/eval/*'
}

from  tfx.proto import example_gen_pb2

input = example_gen_pb2.Input(splits=[
                example_gen_pb2.Input.Split(name='train',
                                            pattern='{YYYY}-{MM}-{DD}/train/*'),
                example_gen_pb2.Input.Split(name='eval',
                                            pattern='{YYYY}-{MM}-{DD}/eval/*')
            ])
examples = csv_input('/tmp')
example_gen = CsvExampleGen(input=examples, input_config=input)

splits {
  name: 'train'
  pattern: 'span-{SPAN}/ver-{VERSION}/train/*'
}
splits {
  name: 'eval'
  pattern: 'span-{SPAN}/ver-{VERSION}/eval/*'
}

from  tfx.proto import example_gen_pb2

input = example_gen_pb2.Input(splits=[
                example_gen_pb2.Input.Split(name='train',
                                            pattern='span-{SPAN}/ver-{VERSION}/train/*'),
                example_gen_pb2.Input.Split(name='eval',
                                            pattern='span-{SPAN}/ver-{VERSION}/eval/*')
            ])
examples = csv_input('/tmp')
example_gen = CsvExampleGen(input=examples, input_config=input)

from  tfx.proto import example_gen_pb2
from  tfx.proto import range_config_pb2

# In cases where files have zero-padding, the width modifier in SPAN spec is
# required so TFX can correctly substitute spec with zero-padded span number.
input = example_gen_pb2.Input(splits=[
                example_gen_pb2.Input.Split(name='train',
                                            pattern='span-{SPAN:2}/train/*'),
                example_gen_pb2.Input.Split(name='eval',
                                            pattern='span-{SPAN:2}/eval/*')
            ])
# Specify the span number to be processed here using StaticRange.
range = range_config_pb2.RangeConfig(
                static_range=range_config_pb2.StaticRange(
                        start_span_number=1, end_span_number=1)
            )

# After substitution, the train and eval split patterns will be
# 'input_dir/span-01/train/*' and 'input_dir/span-01/eval/*', respectively.
examples = csv_input(input_dir)
example_gen = CsvExampleGen(input=examples, input_config=input,
                            range_config=range)

from  tfx.components.example_gen import utils
from  tfx.proto import example_gen_pb2
from  tfx.proto import range_config_pb2

input = example_gen_pb2.Input(splits=[
                example_gen_pb2.Input.Split(name='train',
                                            pattern='{YYYY}-{MM}-{DD}/train/*'),
                example_gen_pb2.Input.Split(name='eval',
                                            pattern='{YYYY}-{MM}-{DD}/eval/*')
            ])
# Specify date to be converted to span number to be processed using StaticRange.
span = utils.date_to_span_number(1970, 1, 2)
range = range_config_pb2.RangeConfig(
                static_range=range_config_pb2.StaticRange(
                        start_span_number=span, end_span_number=span)
            )

# After substitution, the train and eval split patterns will be
# 'input_dir/1970-01-02/train/*' and 'input_dir/1970-01-02/eval/*',
# respectively.
examples = csv_input(input_dir)
example_gen = CsvExampleGen(input=examples, input_config=input,
                            range_config=range)

from tfx.components.base import executor_spec
from tfx.components.example_gen.component import FileBasedExampleGen
from tfx.components.example_gen.csv_example_gen import executor
from tfx.utils.dsl_utils import external_input

examples = external_input(os.path.join(base_dir, 'data/simple'))
example_gen = FileBasedExampleGen(
    input=examples,
    custom_executor_spec=executor_spec.ExecutorClassSpec(executor.Executor))

from tfx.examples.custom_components.presto_example_gen.proto import presto_config_pb2
from tfx.examples.custom_components.presto_example_gen.presto_component.component import PrestoExampleGen

presto_config = presto_config_pb2.PrestoConnConfig(host='localhost', port=8080)
example_gen = PrestoExampleGen(presto_config, query='SELECT * FROM chicago_taxi_trips')

from tfx import components

...

# Exclude the 'eval' split.
statistics_gen = components.StatisticsGen(
             examples=example_gen.outputs['examples'],
             exclude_splits=['eval'])

from tfx import components

...

# Exclude the 'eval' split.
schema_gen = components.SchemaGen(
             statistics=statistics_gen.outputs['statistics'],
             exclude_splits=['eval'])

from tfx import components

...

# Exclude the 'eval' split.
example_validator = components.ExampleValidator(
             statistics=statistics_gen.outputs['statistics'],
             schema=schema_gen.outputs['schema'],
             exclude_splits=['eval'])

from tfx import components
from  tfx.proto import transform_pb2

...

# Analyze the 'train' split and transform all splits.
transform = components.Transform(
      examples=example_gen.outputs['examples'],
      schema=schema_gen.outputs['schema'],
      module_file=_taxi_module_file,
      splits_config=transform_pb2.SplitsConfig(analyze=['train'],
                                               transform=['train', 'eval']))

from tfx import components
from  tfx.proto import trainer_pb2

...

# Train on the 'train' split and evaluate on the 'eval' split.
Trainer = components.Trainer(
      module_file=_taxi_module_file,
      examples=transform.outputs['transformed_examples'],
      schema=schema_gen.outputs['schema'],
      transform_graph=transform.outputs['transform_graph'],
      train_args=trainer_pb2.TrainArgs(splits=['train'], num_steps=10000),
      eval_args=trainer_pb2.EvalArgs(splits=['eval'], num_steps=5000))

from tfx import components
from  tfx.proto import evaluator_pb2

...

# Compute metrics on the 'eval1' split and the 'eval2' split.
Trainer = components.Evaluator(
      examples=example_gen.outputs['examples'],
      model=trainer.outputs['model'],
      example_splits=['eval1', 'eval2'])