A Minor Refactor ...

[ben-nowacki]

Description

... just about everything changed

A quick summary:

• All FeatureSet data is now backed by PyArrow tables. All downstream access utilizes no-copy views (e.g., splitting, batching, etc)
• DAG topology restructured to better separate cocerns into distinct hierachical classses
  • ExperimentNode defines identity and unique UUID/str-based references within a given ExperimentContext
  • GraphNode defines topology and connection capabilities of subclasses
  • ComputeNode defines computation/forward interface for ModelNode and MergeNode
• Samplers return BatchViews (no-copy views over FeatureSet). Not materialized into backend-specific tensors until ModelGraph execution
• Orchestration layer implemented.
  • Experiment classes contains ModelGraph and list of ExperimentPhases to be executed. Also tracks results
  • TrainPhase, EvalPhase, and FitPhase defined for different model execution needs
  • Callbacks implemented for full control over injecting custom execution at any transition point within Experiment transitions (eg, on_phase_start/end) and phase transitions (e.g., on_epoch_start/end, on_batch_start/end).
  • Common Callbacks implemented: EarlyStopping, Evaluation, metric monitoring, etc
• Serialization upgrades. All core classes are fully serializable via a structured Serializer class and registered Handlers. ModelGraph, Expeirment, etc can fully serialize state, configuration, and results, providing a common .save/.load interface.
• Experiment evaluation strategies outlined with concrete subclass implemented for CrossValidation.
• Visualizer class updated to reflect changes.
  • FeatureSets can now be visualized to show split nesting, overlap, and references features/targets/tags
  • ModelGraph now uses different color scheme for frozen vs non-frozen nodes
  • ExperimentPhases now implement the .visualize interface. TrainPhases introduce Samplers to the displayed flowchart, making it easy to see all phase configuration details.
• Docs: reorganized into "how_to", "explanation", "tutorials", and "reference" to better align with the Diataxis model
  • Tutorials currently only has placeholder notebooks.

Related Issues

Closes #16

How Has This Been Tested?

• Local pytest run (unit tests are severely lacking at the moment)
• CI passes (all pass, but unit tests are minimal; no integration tests)
• Manual checks (describe if needed)

Checklist

• My code follows project style guidelines: nox -s pre-commit
• I have added tests that prove my fix is effective or my feature works: nox -s unit
• I have updated documentation if needed: nox -s docs
• I have linked related issues

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 61.39279% with 632 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
modularml/core/data/featureset.py	33.49%	280 Missing ⚠️
modularml/callbacks/early_stopping.py	31.09%	82 Missing ⚠️
modularml/callbacks/evaluation.py	38.20%	55 Missing ⚠️
modularml/core/data/sample_collection_mixin.py	71.05%	55 Missing ⚠️
modularml/core/data/sample_collection.py	85.51%	52 Missing ⚠️
modularml/callbacks/eval_loss_metric.py	27.02%	27 Missing ⚠️
modularml/core/data/batch.py	76.31%	27 Missing ⚠️
modularml/core/data/featureset_view.py	73.49%	22 Missing ⚠️
modularml/core/data/batch_view.py	83.17%	18 Missing ⚠️
modularml/callbacks/metric.py	67.64%	11 Missing ⚠️
... and 1 more

📢 Thoughts on this report? Let us know!