Allow training on a broader variable pool than inference uses#1241
Closed
mcgibbon wants to merge 1 commit into
Closed
Allow training on a broader variable pool than inference uses#1241mcgibbon wants to merge 1 commit into
mcgibbon wants to merge 1 commit into
Conversation
Add an optional `all_training_names` property to `StepConfigABC` that declares the variable pool the data loader should fetch during training, distinct from the (typically narrower) `input_names + output_names` set that inference walks. When `None` (the default for existing steps), the old behavior is preserved. Add `InfillPredictionStepConfig` / `InfillPredictionStep` as the first consumer: training pulls all `all_names` variables, missing ones are zero-filled, and per-channel mask indicators are appended so the network can distinguish real inputs from fill values. At inference, the step routes via `inference_scheme` and behaves like `SingleModuleStep`. Existing subclasses (`SingleModuleStepConfig`, `MultiCallStepConfig`, `SecondaryModuleStepConfig`, `SeparateRadiationStepConfig`, `FCN3StepConfig`, `MockStepConfig`) get a `return None` override to satisfy the abstract-or-final rule.
Contributor
Author
|
Tried splitting this from #1224 but it doesn't make much sense on its own. Closing. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Extracts the "training supports a broader variable pool than inference" piece from
feature/combine-infill-predictionso it can land independently of the multi-task training work, which will follow in a separate PR.Changes:
fme.core.step.step.StepConfigABC.all_training_names: new abstract property; returningNone(the default added on each existing subclass) preserves the oldinput_names + output_namesbehavior, while returning an explicit list tells the data loader to fetch the broader training set.fme.ace.stepper.single_module.StepperConfig.all_names: returnsstep.all_training_nameswhen set, falls back to the previous union otherwise.fme.core.step.infill_prediction.InfillPredictionStepConfig/InfillPredictionStep/InferenceSchemeConfig: first consumer. Training pulls allall_namesvariables, missing ones are zero-filled, and per-channel mask indicators are appended so the network knows which inputs are real. At inference the step routes viainference_schemeand behaves likeSingleModuleStep. Registered as"infill_prediction"and re-exported fromfme.ace.Tests added (
fme/core/step/test_infill_prediction.py)If dependencies changed, "deps only" image rebuilt and "latest_deps_only_image.txt" file updated