4 Commits

Author	SHA1	Message	Date
bvandeusen	6e3c5f697f	feat(ml): tag-eval backend — head-vs-centroid learning-curve eval (persisted) ... Slice 1 of milestone #114 (tagging v2). Proves the frozen-embedding + trained- head spine on the operator's own data, reusing the SigLIP embeddings already stored on image_record — no re-embedding, no GPU. Per concept: train a logistic-regression HEAD (positives + negatives = explicit rejections + sampled unlabeled) vs the old single-CENTROID baseline; report cross-validated precision/recall/AP for both, a LEARNING CURVE (AP/F1 as tagged positives grow 10→30→100→300), and example image ids (head-would-suggest / head-doubts-positive) to eyeball. Persisted so the report SURVIVES navigation (operator-flagged): the run + full report live in a new tag_eval_run row (mirrors library_audit_run); the admin card will rehydrate from GET on mount, not transient state. - models.TagEvalRun + migration 0056; runs on the ml queue (only worker with numpy/sklearn) — numpy/sklearn lazy-imported so the API can still enqueue. - services/ml/tag_eval (compute + start helper, one-running guard), tasks.ml .tag_eval_run, api/tag-eval (POST create, GET history light / detail w/ report). - recover_stalled_tag_eval_runs sweep + retention (keep last 20) + 5-min beat (rule 89). scikit-learn added to requirements-ml. - tests: param normalization + the rehydrate read-path + create/conflict. Frontend admin card (trigger + render persisted report) follows next. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-06-27 22:49:10 -04:00
bvandeusen	681d7777f3	fix(build): install CPU-only torch in ml image (drops ~5.6GB CUDA layer)	2026-05-15 21:04:26 -04:00
bvandeusen	0f47c7485b	chore(fc2a): catch up on stale dep pins — current versions across the board ... Audit against PyPI revealed almost every pin in requirements.txt and requirements-ml.txt was significantly behind. Most bumps are minor-major catch-up; a handful are major-version jumps. requirements.txt: - quart 0.19 -> 0.20 - hypercorn 0.16 -> 0.18 (now declares 3.14 support) - asyncpg 0.30 -> 0.31 - psycopg 3.2 -> 3.3 - alembic 1.13 -> 1.18 (5 minors stale) - pgvector 0.2 -> 0.4 (2 minors stale) - celery 5.4 -> 5.6 - redis 5.0 -> 7.4 (major jump) - cryptography 44 -> 48 - pillow 11.1 -> 12 (major jump; 12.x has 3.14 wheels) - gallery-dl 1.27 -> 1.32 (5 minors stale) - python-dotenv 1.0 -> 1.2 - structlog 24.1 -> 25.5 (major jump) (sqlalchemy 2.0 line is current. imagehash 4.3.2 was already in range.) requirements-ml.txt (not exercised by CI yet; FC-2b territory): - torch 2.2 -> 2.12 (10 minors stale) - torchvision 0.17 -> 0.27 (CAVEAT: excludes Python 3.14.1 specifically — inline comment added) - transformers 4.40 -> 5.8 (major) - onnxruntime 1.17 -> 1.26 - huggingface-hub 0.22 -> 1.14 (major) - opencv-python-headless 4.9 -> 4.13 The torchvision 3.14.1 exclusion is a real footgun — the python-ci runner pulls python:3.14-bookworm (latest patch); if that ever resolves to 3.14.1, the ml-worker image build will fail. FC-2b will exercise this path for the first time, so the constraint is documented inline rather than worked around now. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-14 13:26:37 -04:00
bvandeusen	13eaa35f1c	feat: scaffold backend Python project (Quart + SQLAlchemy + Celery deps) ... Pins runtime and ML deps separately so the regular web image stays lean. Configures ruff for py312 with bugbear, async, and pyupgrade lints enabled. psycopg sync driver included up-front for alembic. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-14 07:31:39 -04:00