Details

    • New Feature
    • Status: Open
    • P2
    • Resolution: Unresolved
    • None
    • None
    • sdk-py-core

    Description

      Users of machine learning frameworks must currently implement their own transforms for running ML inferences. The exception is the TensorFlow RunInference transform. However, this is hosted in its own repo, and has an API that is exclusively geared towards the TensorFlow TFX library. Our goal is to add new implementations of RunInference for the two other popular machine learning frameworks: scikit-learn and Pytorch.

      Please see main design document here.

      Attachments

        1.
        Implement RunInference common interface Sub-task Resolved Andy Ye

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 14h 20m
        2.
        Implement Generic RunInference Base class Sub-task Resolved Ryan Thompson

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 21h 10m
        3.
        Implement RunInference for Scikit-learn Sub-task Resolved Ryan Thompson

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 9h
        4.
        Implement RunInference for PyTorch Sub-task Resolved Andy Ye

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 6h 50m
        5.
        Provide GPU support Sub-task Open Unassigned  
        6.
        Implement Batched DoFn in RunInferenceImpl Sub-task Resolved Brian Hulette  
        7.
        Add Metrics to base class Sub-task Open Unassigned  
        8.
        Hook In Batching DoFn Apis to RunInference Sub-task Open Brian Hulette

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 2h 20m
        9.
        Add documentation page to Beam website Sub-task Open Unassigned  
        10.
        RunInference Benchmarking tests Sub-task Open Anand Inguva

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 10h 20m
        11.
        Implement E2E tests for RunInference classes Sub-task Open Unassigned  
        12.
        Investigate releasing models in inference base class Sub-task Open Unassigned  
        13.
        Add batch_size back off in RunInferenceBase Sub-task Open Unassigned  
        14.
        Figure out how type hints should work Sub-task Open Unassigned  
        15.
        Add resource location hints to base RunInference Implementation Sub-task Open Ryan Thompson

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 2h 20m
        16.
        Correctly comment or remove Metrics Cache Sub-task Open Ryan Thompson

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 1h 50m
        17.
        Drop the clock abastraction and just use time.time for time measurements Sub-task Open Ryan Thompson

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 3h 20m
        18.
        Remove base_run_inference in TFX-BSL and use inference.base in Beam Sub-task Open Unassigned  
        19.
        Add call in RunInference API Sub-task Resolved Andy Ye

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 1h 10m
        20.
        Support **kwargs for PyTorch models. Sub-task In Progress Andy Ye

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 8h 20m
        21.
        Explore versions of pytorch to test in Tox Sub-task Open Andy Ye  
        22.
        Investigate windows Permissions Error on Jenkins Windows Unit tests Sub-task Open Unassigned  
        23.
        Add flag to drop example from PredicitonResult Sub-task Open Unassigned  
        24.
        Investigate load state_dict vs loading whole model Sub-task Open Anand Inguva

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 2h 20m
        25.
        Add support for Pandas Dataframes to sklearn RunInference Implementation Sub-task Open Ryan Thompson

        100%

        Original Estimate - Not Specified Original Estimate - Not Specified
        Time Spent - 1.5h

        Activity

          People

            Unassigned Unassigned
            yeandy Andy Ye
            Votes:
            0 Vote for this issue
            Watchers:
            4 Start watching this issue

            Dates

              Created:
              Updated:

              Time Tracking

                Estimated:
                Original Estimate - Not Specified
                Not Specified
                Remaining:
                Remaining Estimate - 0h
                0h
                Logged:
                Time Spent - 84h 50m
                84h 50m