feat: add seq2seq forecasting training job

[TheMichaelHu]

Thank you for opening a Pull Request! Before submitting your PR, there are a few things you can do to make sure it goes smoothly:

• Make sure to open an issue as a bug/issue before writing your code! That way we can discuss the change, evaluate designs, and agree on the general idea
• Ensure the tests and linter pass
• Code coverage does not decrease (if any source code was changed)
• Appropriate docs were updated (if necessary)

Fixes b/229909845 🦕

---

Adds a SequenceToSequencePlusForecastingTrainingJob to training jobs. This job has the exact same signature as AutoMLForecastingTrainingJob, but we are creating a separate job in case the two models diverge in the future.

The logic for AutoMLForecastingTrainingJob has been moved to a new abstract base class _ForecastingTrainingJob. The only things that differ between the seq2seq and automl training jobs that extend it are the model_type and training_task_definition.

[geraint0923]

Overall LG, just one minor comment.

[geraint0923]

LGTM, thanks!

[ivanmkc]

Can you just remove the run override and defer to super?

[TheMichaelHu]

If we remove this, Sphinx will not generate documentation for the run() function. It can pull the docstring from the parent class (tested with nox -s docs), but I couldn't find a way to make it document the function signature other than by writing it out manually.

[ivanmkc]

Hm, that's quite annoying. Let me ask the SDK team for more solutions.

[ivanmkc]

It looks like you can simplify both AutoMLForecastingTrainingJob and SequenceToSequencePlusForecastingTrainingJob by removing all methods, since they are the exact same as the super's implementation.

We also need to add system tests for these. See tests/system/aiplatform/test_e2e_tabular.py for an example.

[ivanmkc]

Regarding my last comment, you can simplify to:

class AutoMLForecastingTrainingJob(_ForecastingTrainingJob):
    _model_type = "AutoML"
    _training_task_definition = schema.training_job.definition.automl_forecasting
    _supported_training_schemas = (schema.training_job.definition.automl_forecasting,)

class SequenceToSequencePlusForecastingTrainingJob(_ForecastingTrainingJob):
    _model_type = "Seq2Seq"
    _training_task_definition = schema.training_job.definition.seq2seq_plus_forecasting
    _supported_training_schemas = (
        schema.training_job.definition.seq2seq_plus_forecasting,
    )

I ran the unit tests and it works fine. Otherwise, everything else looks pretty good.

[TheMichaelHu]

Thanks for the review @ivanmkc! qq: how will simplifying the way you suggested impact the documentation? Is there a way to check/ensure that there will be no impact? Like if the code used to generate the docs has an option to document functions from super()?

[ivanmkc]

LGTM if system tests pass.