Add Krea2 tuning coverage report
This commit is contained in:
@@ -0,0 +1,22 @@
|
||||
#!/usr/bin/env python3
|
||||
from __future__ import annotations
|
||||
|
||||
import sys
|
||||
from pathlib import Path
|
||||
|
||||
|
||||
ROOT = Path(__file__).resolve().parents[1]
|
||||
if str(ROOT) in sys.path:
|
||||
sys.path.remove(str(ROOT))
|
||||
sys.path.insert(0, str(ROOT))
|
||||
|
||||
import krea2_tuning_report # noqa: E402
|
||||
|
||||
|
||||
def main() -> int:
|
||||
print(krea2_tuning_report.markdown_report())
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
raise SystemExit(main())
|
||||
+28
-2
@@ -21,8 +21,9 @@ from typing import Any, Callable
|
||||
|
||||
|
||||
ROOT = Path(__file__).resolve().parents[1]
|
||||
if str(ROOT) not in sys.path:
|
||||
sys.path.insert(0, str(ROOT))
|
||||
if str(ROOT) in sys.path:
|
||||
sys.path.remove(str(ROOT))
|
||||
sys.path.insert(0, str(ROOT))
|
||||
|
||||
import caption_naturalizer # noqa: E402
|
||||
import caption_format_route # noqa: E402
|
||||
@@ -63,6 +64,7 @@ import krea_normal_formatter # noqa: E402
|
||||
import krea_pair_formatter # noqa: E402
|
||||
import krea2_eval_log # noqa: E402
|
||||
import krea2_pose_variant_catalog # noqa: E402
|
||||
import krea2_tuning_report # noqa: E402
|
||||
import krea_row_fields # noqa: E402
|
||||
import location_config # noqa: E402
|
||||
import loop_nodes # noqa: E402
|
||||
@@ -6843,6 +6845,29 @@ def smoke_krea2_eval_log_policy() -> None:
|
||||
_expect(clean.get("observation") != "mutation should not leak", "Krea2 eval log leaked caller mutation")
|
||||
|
||||
|
||||
def smoke_krea2_tuning_report_policy() -> None:
|
||||
rows = krea2_tuning_report.coverage_rows()
|
||||
catalog_keys = krea2_pose_variant_catalog.variant_keys()
|
||||
_expect([row.get("key") for row in rows] == catalog_keys, "Krea2 tuning report row order should follow catalog order")
|
||||
by_key = {row.get("key"): row for row in rows}
|
||||
boobjob = by_key.get("pov_boobjob_upright_cleavage") or {}
|
||||
_expect(boobjob.get("coverage_state") == "proven_with_evidence", "Boobjob report should be proven with evidence")
|
||||
_expect(boobjob.get("accepted_evidence_count", 0) >= 1, "Boobjob report lost accepted evidence count")
|
||||
ballsucking = by_key.get("pov_ballsucking_low_head") or {}
|
||||
_expect(ballsucking.get("coverage_state") == "needs_fixed_seed_tests", "Ballsucking report should need fixed-seed tests")
|
||||
_expect(ballsucking.get("accepted_evidence_count") == 0, "Ballsucking report should not have accepted evidence yet")
|
||||
summary = krea2_tuning_report.coverage_summary()
|
||||
_expect(summary.get("status_counts", {}).get("proven") == 3, "Krea2 tuning report proven count changed")
|
||||
_expect(summary.get("status_counts", {}).get("candidate") == 1, "Krea2 tuning report candidate count changed")
|
||||
_expect(
|
||||
summary.get("variants_without_accepted_evidence") == ["pov_ballsucking_low_head"],
|
||||
f"Krea2 tuning report missing-evidence set changed: {summary.get('variants_without_accepted_evidence')}",
|
||||
)
|
||||
markdown = krea2_tuning_report.markdown_report()
|
||||
_expect("pov_ballsucking_low_head" in markdown, "Krea2 tuning report markdown lost candidate variant")
|
||||
_expect("needs_fixed_seed_tests" in markdown, "Krea2 tuning report markdown lost coverage state")
|
||||
|
||||
|
||||
def smoke_krea_pov_penetration_route() -> None:
|
||||
pair = pb.build_insta_of_pair(
|
||||
row_number=1,
|
||||
@@ -9849,6 +9874,7 @@ SMOKE_CASES: list[tuple[str, Callable[[], None]]] = [
|
||||
("krea2_pov_pose_variant_catalog", smoke_krea2_pov_pose_variant_catalog),
|
||||
("krea2_pose_variant_catalog_policy", smoke_krea2_pose_variant_catalog_policy),
|
||||
("krea2_eval_log_policy", smoke_krea2_eval_log_policy),
|
||||
("krea2_tuning_report_policy", smoke_krea2_tuning_report_policy),
|
||||
("krea_pov_penetration_route", smoke_krea_pov_penetration_route),
|
||||
("pov_outercourse_position_routes", smoke_pov_outercourse_position_routes),
|
||||
("pov_oral_position_routes", smoke_pov_oral_position_routes),
|
||||
|
||||
Reference in New Issue
Block a user