Repository navigation
measurement: schema version 2 — task and family, detection parity, vendor figures - #11
Closed
nathanhubens wants to merge 2 commits into
Closed
nathanhubens wants to merge 2 commits into
nathanhubens wants to merge 2 commits into
Conversation
…ndor figures - task (classification | detection, required) and family on both kinds; one column after join. - quality: accuracy_source (measured | vendor_published); argmax_parity becomes parity + parity_metric (argmax | box_iou90); intervals and parity rule required when measured, absent when a vendor published the figure. - latency: latency_source may be vendor_published; power_w; the known recipe keys (RECIPE_KEYS) checked for type. - Every row is read against the fields of its own schema_version (version-1 files stay valid as written); upgrade(row, task, family) gives the version-2 row; join refuses mixed versions; to_parquet writes one version per file. - A vendor figure is well formed and never VALID as a measurement.
nathanhubens
force-pushed
the
feat/measurement-schema-v2
branch
from
October 14, 2026 12:38
fc199f7 to
581ca22
Compare
Contributor
Author
|
Closing: the measurement-row schema is no longer maintained in this package, so this change is superseded. fastermodels keeps loading, evaluation, cards and the publication gate. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Schema version 2 for
fastermodels.measurement. Version-1 rows stay valid as written.What changes
task(classification/detection, required) andfamily; one column afterjoin, both sides must agreeaccuracy_source(measured/vendor_published);argmax_parity→parity+parity_metric(argmax/box_iou90: same class, IoU ≥ 0.9);ci_*,parity_metric,parity_floorrequired when measured, absent for a vendor figurelatency_sourcemay bevendor_published;power_w; knownrecipekeys checked for type (RECIPE_KEYS), other keys freeCompatibility
schema_version(FIELDS_BY_VERSION);FIELDSis the current version.upgrade(row, task, family=None)returns the version-2 row a version-1 row is (argmax_parity→parity,parity_metric='argmax',accuracy_source='measured').joinrefuses to pair two versions;to_parquetwrites one version per file, with that version's columns.schema_versionreturns that single problem.Why
Detection rows need a box parity, not an argmax one; comparisons against vendor reference models need a figure that is recorded and labelled, never a measurement (well formed,
INVALIDfor anything that needs a measurement).Tests
Existing tests pinned to version 1; a new
v2cell covers upgrade, version separation, detection rows, vendor rows, recipe keys, power, join and Parquet. Full suite passes.