Replace single_site_matching with single_site#552
Merged
Conversation
bruno-f-cruz
changed the title
bruno-f-cruz
requested changes
May 30, 2026
Member
bruno-f-cruz
left a comment
bruno-f-cruz
left a comment
There was a problem hiding this comment.
Small style things just so make sure we can read this easily in the future.
Member
There was a problem hiding this comment.
This has to be renamed "readme.md" to be picked up by the doc building process. Also I am not sure this is up-to-date. I will leave it up to you to include it or not, but if you don't feel like maintaining it just drop it.
Comment on lines
+107
to
+109
| inter_patch.truncation_parameters.max = ip_max | ||
| gen.inter_site.length_distribution.distribution_parameters.value = is_len | ||
| gen.reward_site.length_distribution.distribution_parameters.value = rs_len |
Member
There was a problem hiding this comment.
Guard against other distributions just so we fail loudly in case it happens:
else:
raise ValueError(f"Unexpected inter-patch distribution type: {type(inter_patch)}")| reward-site lengths from the compressed start toward full spacing, scaled by | ||
| prior locomotion (n_patches_seen / GEOMETRY_EASE_PATCHES, clamped to 1). On the | ||
| first session (no prior metrics) the geometry stays compressed.""" | ||
| frac = _clamp(metrics.n_patches_seen / GEOMETRY_EASE_PATCHES, 0.0, 1.0) |
Member
There was a problem hiding this comment.
Can you use slightly longer names? I kinda know what these are, but being explicit with names here will help quite alot in the long run.
ip = inter_patch
is = inter_site
gen = patch_generator
inter_patch = inter_patch_generator
|
|
||
|
|
||
| class TestLearnToChooseToProbabilityGridShort: | ||
| def _ok(self) -> SingleSiteMetrics: |
Member
There was a problem hiding this comment.
Consider renaming to something like: _make_passing_metrics()
Port the single_site_no_matching curriculum from
Aind.Behavior.VrForaging.Curricula#40 into the monorepo as a new
independent curriculum, renamed to `single_site` per Bruno's request.
Delete the old `single_site_matching` curriculum entirely.
Changes vs. the original branch in .Curricula:
- Rename single_site_no_matching -> single_site (package + class +
CURRICULUM_NAME + schema filename).
- Drop cross-curriculum dependency on
single_site_matching.helpers.clamp; inline as a local _clamp helper
in policies.py.
- Adapt to monorepo task_logic API drift:
- aind_behavior_services.task_logic.distributions ->
aind_behavior_services.task.distributions.
- vr_task_logic.OdorSpecification(index=i, concentration=1) ->
one-hot list[int] (OdorMixture).
- Regenerate schema/single_site.json against monorepo pydantic models.
Tests: 31 single_site tests + 7 cli tests + 13 depletion tests pass.
…, stochastic delay ramp)
- Restrict graduated grid to summed reward probability {0.8, 1.0, 1.2} (13
blocks) and rename graduated_* -> probability_grid_short/long_delay.
- learn_to_stop: fixed 1.0 s stop (drop stop-duration updater), fast velocity
ramp (GAIN on_success 0.93), compressed->full geometry easing via
p_learn_to_run, and a cross-session reward-probability water gate
(p_reward_water_gate: 1.0 while struggling, 0.8 once earning water).
- Remove learn_to_stop_low_p; struggling animals stay in learn_to_stop.
- Fold three_contrast into probability_grid_short_delay: move the
REWARD_DELAY_OFFSET ramp (0 -> 1.5) into the grid stage; drop the coupled
stop-duration shrink ramp (stop is now fixed).
- Keep reward delay stochastic in both grid stages (Exp base + within-session
additive ramp in short, stationary in long) with a 0.2 s floor.
- Rewrite tests for the 4-stage chain and regenerate schema/single_site.json.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Inline the _learn_to_stop_saturation_met alias into its only caller. - p_learn_to_run: explicit variable names + raise on an unexpected inter-patch distribution type. - Rename test helper _ok -> _make_passing_metrics. - Replace the stale SUMMARY.md with a concise, current README.md (so the doc builder picks it up). - Speed up the probability_grid_short patience ramp (REWARD_DELAY_OFFSET on_success 0.005 -> 0.01) so it graduates in ~1 session. - Lower the learn_to_stop graduation gate n_patches_seen 300 -> 250. - Regenerate schema. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
main's #547 generalized environment sampling: Block.environment_statistics -> Block.environment, and EnvironmentStatistics -> MarkovEnvironment. Update make_block, the geometry/reward policies, and the tests accordingly, and regenerate schema/single_site.json against task-logic 1.0.0rc6. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
No behavior change — schema/single_site.json is byte-identical. - Extract task-logic builders into helpers.py (make_patch, make_block, make_default_operation_control, make_reward_delay, clamp, lerp), matching the per-curriculum helpers pattern used by depletion/template. - Reuse ..utils.trainer_state_from_file / metrics_from_dataset_path instead of redefining them locally; drop the now-dead imports. - Use the `task_logic` import alias (was vr_task_logic); normalize the transition predicates to `return (...)`. - Rename GRADUATED_* -> PROBABILITY_GRID_* to match the stage names, and de-duplicate the 13-block grid via _POST_STOP_PATCH_KWARGS + _probability_grid_blocks. - Remove the dead last_stop_duration_offset_updater metric (left from three_contrast); drop stale S1/S2/S3 references; name the cross-session seeding factors. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
galenlynch
force-pushed
the
feat-single-site-curriculum
branch
from
b55c64f to
f72a9e5
Compare
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
2 participants
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.
Port the single_site_no_matching curriculum from
Aind.Behavior.VrForaging.Curricula#40 into the monorepo as a new independent curriculum, renamed to
single_siteper Bruno's request. Delete the oldsingle_site_matchingcurriculum entirely.Changes vs. the original branch in .Curricula:
Tests: 31 single_site tests + 7 cli tests + 13 depletion tests pass.