AnandInguva commented on code in PR #26795: URL: https://github.com/apache/beam/pull/26795#discussion_r1247176420
########## sdks/python/apache_beam/ml/transforms/base.py: ########## @@ -0,0 +1,165 @@ +# +# Licensed to the Apache Software Foundation (ASF) under one or more +# contributor license agreements. See the NOTICE file distributed with +# this work for additional information regarding copyright ownership. +# The ASF licenses this file to You under the Apache License, Version 2.0 +# (the "License"); you may not use this file except in compliance with +# the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from typing import Generic +from typing import List +from typing import Optional +from typing import TypeVar + +import apache_beam as beam + +# TODO: Abstract methods are not getting pickled with dill. +# https://github.com/uqfoundation/dill/issues/332 +# import abc + +__all__ = ['MLTransform'] + +TransformedDatasetT = TypeVar('TransformedDatasetT') +TransformedMetadataT = TypeVar('TransformedMetadataT') + +# Input/Output types to the MLTransform. +ExampleT = TypeVar('ExampleT') +MLTransformOutputT = TypeVar('MLTransformOutputT') + +# Input to the process data. This could be same or different from ExampleT. +ProcessInputT = TypeVar('ProcessInputT') +# Output of the process data. This could be same or different +# from MLTransformOutputT +ProcessOutputT = TypeVar('ProcessOutputT') + +# Input to the apply() method of BaseOperation. +OperationInputT = TypeVar('OperationInputT') +# Output of the apply() method of BaseOperation. +OperationOutputT = TypeVar('OperationOutputT') + + +class ArtifactMode(object): + PRODUCE = 'produce' + CONSUME = 'consume' + + +class BaseOperation(Generic[OperationInputT, OperationOutputT]): + def apply( + self, inputs: OperationInputT, column_name: str, *args, + **kwargs) -> OperationOutputT: + """ + Define any processing logic in the apply() method. + processing logics are applied on inputs and returns a transformed + output. + Args: + inputs: input data. + """ + raise NotImplementedError + + +class _ProcessHandler(Generic[ProcessInputT, ProcessOutputT]): + """ + Only for internal use. No backwards compatibility guarantees. + """ + def process_data( + self, pcoll: beam.PCollection[ProcessInputT] + ) -> beam.PCollection[ProcessOutputT]: + """ + Logic to process the data. This will be the entrypoint in + beam.MLTransform to process incoming data. + """ + raise NotImplementedError + + def append_transform(self, transform: BaseOperation): + raise NotImplementedError + + +class MLTransform(beam.PTransform[beam.PCollection[ExampleT], + beam.PCollection[MLTransformOutputT]], + Generic[ExampleT, MLTransformOutputT]): + def __init__( + self, + *, + artifact_location: str, + artifact_mode: str = ArtifactMode.PRODUCE, + transforms: Optional[List[BaseOperation]] = None, + is_input_record_batches: bool = False, + output_record_batches: bool = False, Review Comment: Right now, the columns are specified at operation level instead of transform level. The entry point for column x could be at the beginning but the entry point at column y could be in the middle of the list. User might pass `input_record_batch=True` to the y column if they doesn't understand exactly what we were instructing. If we ask the user to provide like this, I feel like it could get a little complicated ``` transforms = [ op1(columns=['x']), op2(columns=['y', input_record_batch=True], op3(columns=['z'] ] ``` We can also iterate on this in the v2 since I guess this needs another discussion and remove the option for Record batches for now. We would support Dict[str, Any] in V1. what do you think? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
