Commit Graph
216 Commits
Author SHA1 Message Date
sgeboers 2891e9ee70 feat: add StemAtlas Streamlit app, explorer, Docker deployment, blog charts 2026-03-22 22:38:17 +01:00
sgeboers daa22c5e2b feat: complete parliamentary embedding pipeline with full historical coverage
- Add fused (SVD + text) embedding pipeline for annual windows 2016-2026
- Fix store_fused_embedding duplicate bug: DELETE before INSERT (idempotent)
- Add --text-batch-size CLI flag to run_pipeline.py (default 200)
- Add explicit --start-date/--end-date to download_past_year.py
- Backfill mp_votes for all motions (party-level votes, 111k new rows)
- Add similarity cache recompute: 212k rows across 9 annual windows
- Improve ai_provider retry logic, text_pipeline batching
- Improve analysis/political_axis PCA handling and visualizations
- Add diagnostic/utility scripts: compare_svd, generate_compass, inspect_axis, etc.
- Untrack data/motions.db (3.6GB binary), add to .gitignore with outputs/
- Update continuity ledger with full session state
2026-03-22 16:08:06 +01:00
sgeboers a78bee9b0a feat(similarity): add precomputed similarity cache, fix fusion N+1, add 429 retry
- Add similarity/ package (compute.py, lookup.py) with numpy-based
  pairwise cosine similarity and cached lookup
- database.py: create embeddings + similarity_cache tables in _init_database(),
  add store_similarity_batch/get_cached_similarities/clear_similarity_cache helpers
- pipeline/fusion.py: replace N+1 per-motion embedding SELECT with single
  bulk JOIN using DuckDB QUALIFY window function
- ai_provider.py: retry HTTP 429 with Retry-After header support
- migrations/2026-03-22-add-similarity-cache.sql: make executable
- Add tests for similarity compute, db helpers, and 429 retry (34 pass, 2 skip)
2026-03-22 03:02:25 +01:00
sgeboers a248807e03 Add design: embedding-based motion similarity cache
Precomputed top-K similarity cache replacing the naive Python-scan
search_similar(). Also covers fixes for: embeddings table missing from
_init_database, fusion N+1 query, and ai_provider 429 retry.
2026-03-22 02:39:20 +01:00
sgeboers bf68e48460 fix(analysis): improve PCA handling when PC1 dominates, add pca_residual option and plot autoscaling/variance annotation 2026-03-22 00:13:56 +01:00
sgeboers 23a1234314 feat(analysis): add 2D political compass (PCA/anchor) and 2D trajectories + visualizations
- compute_2d_axes (pca, anchor) with optional L2-normalisation pre-projection
- compute_2d_trajectories: per-MP coords, step vectors, magnitudes, totals
- plot_political_compass and plot_2d_trajectories (Plotly HTML)
- tests/test_political_compass.py (synthetic unit test)
2026-03-21 23:58:38 +01:00
sgeboers 3551a82f83 feat(analysis): add 2D political compass and 2D trajectories
- compute_2d_axes (PCA + anchor)
- compute_2d_trajectories
- plot_political_compass, plot_2d_trajectories
- unit test: tests/test_political_compass.py
2026-03-21 23:53:55 +01:00
sgeboers f7d806dc3a fix(analysis): add Procrustes alignment and normalize vectors for drift computation
SVD sign/rotation is arbitrary per window. Without alignment, drift was
dominated by basis flips (~1.9/step max=2.0) rather than real political movement.

- _procrustes_align_windows(): aligns each window to the previous using
  orthogonal Procrustes on common entities (scipy, falls back gracefully)
- compute_trajectories(): builds aligned window dict before per-MP drift calc,
  adds normalize=True (L2-normalise) to remove cross-window magnitude differences
  caused by varying numbers of motions per quarter
- Results now in sensible range: NSC=2.28, DENK=1.90, ... PVV=0.82, FVD=0.70
- NSC large late jump (1.39 in Q4→Q1 2026) matches its parliamentary fracture
- Add outputs/trajectories_party_aligned.html with cleaned-up drift chart
2026-03-21 23:37:20 +01:00
sgeboers aa2f66ac9f feat(analysis): fetch real MP metadata, fix anchor axis for party-level actors
- fetch_mp_metadata: use real OData URL with pagination (1200 records, 5 pages)
  uses Fractie.Afkorting not NaamNL for abbreviation matching
  skips Verwijderd=true records
- upsert_mp_metadata: keep most recent membership (prefer active over ended,
  then higher Van date) so current party affiliations are not overwritten by historical
- compute_anchor_axis: anchor directly on party-level SVD entities (GroenLinks-PvdA etc)
  before falling back to mp_metadata individual MP lookup
- test_fetch_mp_metadata: fix mock for timeout kwarg + pagination + Afkorting field
- Generated anchor axis HTML for 2025-Q2 through 2026-Q1 in outputs/
2026-03-21 23:33:47 +01:00
sgeboers 5ad83ef1be fix(tests): update test_extract_mp_votes for party-level actor inclusion
New extract_mp_votes behavior inserts all actors (party + individual MPs),
not only comma-name MPs. Test now validates both types and their party column.
Also adds generated HTML visualizations (political axis x5 windows + trajectories).
2026-03-21 23:25:27 +01:00
sgeboers 847b783877 fix(pipeline): fix API pagination, add skip_details fast path, bulk mp_votes insert
- _get_voting_records returns (records, besluit_meta) tuple; paginate via Besluit?expand=Stemming (469/mo vs 8400)
- get_motions(skip_details=True) bypasses per-motion detail chain (3 HTTP calls/motion)
- extract_mp_votes rewritten: bulk DataFrame insert (80k rows in 1.9s), includes party-level actors
- run_pipeline.py fixed: pass db_path not db, handle dict/int return types
- download_past_year.py: skip_details=True default, limit-per-chunk default 50000
2026-03-21 23:24:06 +01:00
sgeboers f2a831dfcf feat(pipeline): add orchestrator CLI, analysis modules, and ActorFractie ingestion
- pipeline/run_pipeline.py: CLI orchestrator for all 5 pipeline phases with
  --dry-run, --skip-*, --window-size, --svd-k, --start/end-date flags
- analysis/{political_axis,trajectory,clustering,visualize}.py: PCA/anchor
  ideological axis, MP drift trajectories, UMAP + KMeans clustering, Plotly HTML output
- api_client.py: capture ActorFractie per individual MP vote (comma in ActorNaam)
  into mp_vote_parties dict on each motion
- database.insert_motion: auto-insert mp_votes rows with party affiliation for
  newly ingested motions when mp_vote_parties is present
- Add scikit-learn to pyproject.toml for KMeans clustering
- tests/test_run_pipeline.py: window generation, dry-run, skip-all paths
- tests/test_analysis.py: PCA axis, anchor axis, trajectory drift, KMeans

Ref: thoughts/shared/plans/2026-03-21-parliamentary-embedding-pipeline-plan.md
2026-03-21 22:40:28 +01:00
sgeboers a36e6cba4e feat(pipeline): implement parliamentary embedding pipeline MVP
- Add 4 migration files: mp_votes, mp_metadata, svd_vectors, fused_embeddings
- Extend database.py with 5 new helper methods and table init
- Add pipeline/ package: extract_mp_votes, fetch_mp_metadata, text_pipeline,
  svd_pipeline (with Procrustes alignment), fusion
- Add full test suite (17 tests) covering all pipeline modules and migrations
- Fix Procrustes alignment bug: scipy scale is a norm value, not a multiplier
- Fix DuckDB date type handling in test assertions (datetime.date vs string)
- Remove duckdb.py shim; tests now run against real duckdb + scipy via uv

Ref: thoughts/shared/plans/2026-03-21-parliamentary-embedding-pipeline-plan.md
2026-03-21 22:31:22 +01:00
sgeboers c498c3467e update plan: replace spike with confirmed FractieZetelPersoon fetch task 2026-03-21 22:05:56 +01:00
sgeboers 0bbda408fb plan: parliamentary embedding pipeline MVP implementation plan 2026-03-21 21:57:39 +01:00
sgeboers fd73da3752 design: parliamentary embedding pipeline (late fusion SVD + text) 2026-03-21 21:54:32 +01:00