Skip to content

validation.md

Latest commit

 

History

History
94 lines (63 loc) · 2.49 KB

validation.md

File metadata and controls

94 lines (63 loc) · 2.49 KB
Raw

Validation

Regression and lab-check artifacts live under validation/. Each intentional run should include manifest.json, metrics CSV, and usually report.md (plus optional mesh/preview).

Quick health checks

morphostack doctor
pytest -q
cd apps/web && npm run build

Optional heavy tests that load large real files:

pytest -m slow

Synthetic geometry

python scripts/validate_synthetic.py
python scripts/validate_synthetic_ellipsoid.py
python scripts/validate_synthetic_touching.py

Outputs under:

  • validation/runs/synthetic-sphere/
  • validation/runs/synthetic-ellipsoid/
  • validation/runs/synthetic-touching-failure/ (negative / hard case)

Sphere and ellipsoid regressions should PASS within their configured tolerances.

Crowded real stacks

python scripts/validate_crowded.py

Creates multi-object seed runs such as:

  • validation/runs/czi-1650-crowded-two-objects/
  • validation/runs/rbc-image46-crowded-two-objects/

Each object folder records threshold, profile, seed, Z range, voxel source, tracking diagnostics, and mesh path in manifest.json.

DOPC

# PowerShell helper
./scripts/validate_dopc.ps1
# or seeded variant
python scripts/validate_dopc_seeded.py

See validation/data-notes/dopc-movie1.md and citations.md for dataset attribution.

Aggregate report

python scripts/validate_all.py

Writes validation/reports/validation-all.md (and JSON companion).

Active surfaces vs threshold

python scripts/compare_active_surfaces_threshold.py

Reports land in validation/reports/active-surfaces-vs-threshold-*.md.

Compare two CSV exports

morphostack validate reference_metrics.csv new_metrics.csv --tolerance 0.000001
morphostack validate ref_batch.csv new_batch.csv --key-column source_path --all-columns

CSV validation proves numeric regression stability, not biological truth, unless the reference is a curated lab standard.

What to inspect in a run

  1. voxel_source — never treat default as calibrated biology.
  2. object_seed / tracking — same object for preview, analyze, mesh.
  3. warnings — default_voxel_size, tracking_lost_*, roi_boundary_touch, likely_neighbor_merge.
  4. Crowded pairs — distinct mean area_um2 for object A vs B when both are real separate objects.
  5. Preview PNGs — human visual QC.

Related

troubleshooting.md · limitations.md · methods.md