Repository navigation
v0.5.46 — Active-learning recovery benchmark (Loop 25) - #144
Merged
Merged
Conversation
Add benchmark validating that the batch-2 selector can recover known active candidates faster than random baseline: - active_learning/benchmark.py: run_active_learning_benchmark() simulates multi-round active learning, hides N active candidates, tracks rounds-to- first-recovery and cumulative recall. Compares against random baseline (20-trial average). - generate_benchmark_pool() creates synthetic candidates with realistic ensemble/disagreement signal separation. - Pre-registered thresholds: PREREGISTERED_MAX_ROUNDS_TO_FIRST_RECOVERY=3, PREREGISTERED_MIN_RECALL=0.33. The passed field encodes both. - CLI : generates synthetic pool inline if no CSV. - 8 new tests, 1832 total passing.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Loop 25 — Active-learning benchmark on synthetic data
Bottleneck
The batch-2 selector could pick candidates, but there was no benchmark validating it recovers known active sequences faster than random baseline.
Changes
active_learning/benchmark.py:run_active_learning_benchmark()simulates multi-round recovery, hides N active candidates from selection, tracks rounds-to-first-recovery and recall, compares to 20-trial random baselinegenerate_benchmark_pool()creates synthetic pool where actives have higher ensemble/lower disagreementbench active-learning: generates synthetic pool inline if no--pool-csvVerification
Next
Loop 26 — Integration between active-learning selector and calibration pipeline