Migrate downstream consumers to multi-metric replay API (#5139)#5139
Open
ltiao wants to merge 3 commits intofacebook:mainfrom
Open
Migrate downstream consumers to multi-metric replay API (#5139)#5139ltiao wants to merge 3 commits intofacebook:mainfrom
ltiao wants to merge 3 commits intofacebook:mainfrom
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #5139 +/- ##
==========================================
+ Coverage 96.40% 96.45% +0.05%
==========================================
Files 613 613
Lines 68191 68458 +267
==========================================
+ Hits 65740 66032 +292
+ Misses 2451 2426 -25 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Summary: The experiment replay system (`MapDataReplayMetric`, `MapDataReplayRunner`, `replay_experiment`) is hardcoded for single-objective optimization, blocking multi-objective early stopping. `MapDataReplayMetric` conflates data serving with progression state, so multiple metrics cannot share a coherent timeline. This diff series extracts shared state into a `MapDataReplayState` coordinator. This diff (1/3) adds the `MapDataReplayState` class -- a shared state coordinator that manages normalized cursor-based progression across multiple metrics and trials. Uses a global min/max MAP_KEY to preserve cross-metric timing alignment. Serves original MAP_KEY values; downstream ESS normalizes independently via `_maybe_normalize_map_key`. Existing `MapDataReplayMetric` and helpers are unchanged in this diff. Differential Revision: D98741817
…layState (facebook#5138) Summary: The experiment replay system (`MapDataReplayMetric`, `MapDataReplayRunner`, `replay_experiment`) is hardcoded for single-objective optimization, blocking multi-objective early stopping. `MapDataReplayMetric` conflates data serving with progression state, so multiple metrics cannot share a coherent timeline. This diff series extracts shared state into a `MapDataReplayState` coordinator. This diff (2/3) rewrites `MapDataReplayMetric` and `MapDataReplayRunner` to delegate to the shared `MapDataReplayState` introduced in D98741817. Also updates `replay_experiment` and `estimate_hypothetical_early_stopping_savings` for multi-metric support: accepts `metrics: list[Metric]`, builds single-objective or multi-objective `OptimizationConfig`, extracts objective thresholds from the historical config, re-indexes non-contiguous trial indices, and deprecates `num_samples_per_curve`. Backward-compat `metric=` kwarg preserved for downstream callers (removed in D98741814). Differential Revision: D98741816
Summary: The experiment replay system (`MapDataReplayMetric`, `MapDataReplayRunner`, `replay_experiment`) is hardcoded for single-objective optimization, blocking multi-objective early stopping. This diff series extracts shared state into a `MapDataReplayState` coordinator. This diff (3/3) migrates all downstream consumers of `replay_experiment` and `estimate_hypothetical_early_stopping_savings` from the deprecated `metric=` (singular) kwarg to the new `metrics=` (list) API introduced in D98741816, and removes the backward-compat shim from `experiment_replay.py`. Consumers updated: - `ax/analysis/healthcheck/early_stopping_healthcheck.py` Differential Revision: D98741814
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
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.
Summary:
The experiment replay system (
MapDataReplayMetric,MapDataReplayRunner,replay_experiment) is hardcoded for single-objective optimization, blocking multi-objective early stopping. This diff series extracts shared state into aMapDataReplayStatecoordinator.This diff (3/3) migrates all downstream consumers of
replay_experimentandestimate_hypothetical_early_stopping_savingsfrom the deprecatedmetric=(singular) kwarg to the newmetrics=(list) API introduced in D98741816, and removes the backward-compat shim fromexperiment_replay.py.Consumers updated:
ax/analysis/healthcheck/early_stopping_healthcheck.pyDifferential Revision: D98741814