Quarterly windows (29 of 41 total) diluted PC1 explained variance ratio
from ~20% down to ~14.6%. The fix splits the vector collection loop into:
- pca_vecs: annual windows only (re.match r'^\d{4}$') -> M_pca used for SVD
- all_vecs: every window -> M used for projections onto derived axes
Centering for SVD and global_mean for projection both now use M_pca.mean(axis=0)
so axes are consistent. Falls back to all windows if no annual windows exist.
Replaces static ideology CSV as primary axis classification signal with
per-year motion projection + Dutch keyword classifier. Adds axis-swap
logic so left-right is conventionally on X when present. Adds Option C
UI expander showing top motions per axis pole.
The global PCA X-axis flip uses centroids averaged across all windows,
which can leave individual windows with left/right inverted (e.g. PvdA
appearing right of VVD in 2020). Mirror the existing per-window Y-axis
correction to also check and flip X values per window.
Replace hardcoded 'Links-Rechts' / 'Progressief-Conservatief' axis labels
with values from classify_axes(). Add per-year interpretation caption when
axis quality score is below the 0.65 correlation threshold.
classify_axes() correlates per-party PCA positions against party_ideologies.csv
to assign honest dynamic labels (Links-Rechts, Coalitie-Oppositie, etc.)
instead of always assuming the first PCA axis is left-right.
Add design for honest PCA axis labeling — validates each compass axis
against a party ideology reference CSV and labels dynamically (Links–Rechts,
Coalitie–Oppositie, or fallback) instead of hardcoding Left–Right always.
The global orientation check using party centroids averaged across all
windows was insufficient — individual windows (notably 2023) could still
have conservative parties above progressive ones on the Y-axis.
Added a per-window flip in compute_2d_axes (PCA branch) that checks
prog_avg_y vs cons_avg_y for each window independently and negates all
Y values in that window when cons > prog. Flipped window IDs are stored
in axis_def['y_flipped_windows'] for diagnostics.
Moved the canonical party set definitions outside the orientation try-
block so they are always in scope for the per-window correction.
Added test_per_window_y_orientation to cover the case where one window
is globally fine but locally inverted.
- Rename app to 'Motief: de stematlas' in Home.py
- Remove PCA variance caption from compass tab
- Hardcode db_path and window_size; remove sidebar inputs
- Change trajectories default to [CDA, D66, VVD]
- Move quiz to pages/1_Stemwijzer.py; wrap in st.form
- Remove quiz tab from main explorer
- Add pytest dev dep + fix test fixtures (_load_mp_vectors_for_window)
- Add test_pca_axis_orientation with proper PCA variance dominance
- Replace gtfs/bokeh deploy with motief/streamlit (port 8501)
- Update inventory to motief.sgeboers.nl
- Remove stale .drone.yml
- Add CI guard to forbid .env in repo
- Add env removal report and secrets rotation checklist
The mp_votes table contains both party-aggregate rows (e.g. 'PVV', 'NSC')
and individual MP rows (e.g. 'Aardema, M.'). Running SVD on both together
creates a block-diagonal vote matrix where party codes and individual MPs
occupy disjoint SVD dimensions — causing dim 0 to be zero for all 421 MPs.
Fix: _build_expanded_rows() converts every party-level vote to individual
MP votes using mp_metadata date ranges (active MPs on motion date). Motions
that already have individual MP records are kept as-is. A party name mapping
handles NSC/Nieuw Sociaal Contract and other canonical name variants.
Results for current_parliament: 517 individual MPs, all 8732 motions covered,
dim 0 std=23.1 (was 0.0 for all MPs). PVV/NSC/BBB on positive end, SP/GL/PvdD
on negative end — matches expected left-right political axis.
All 11 annual windows (2016-2026) re-run with the new pipeline.
- load_party_axis_scores now loads individual MP vectors and averages
per party (same data source as the political compass), so SVD axis
rankings are consistent between the two tabs. Previously it used
party-aggregate rows which gave structurally 0 signal on dim 3.
- Re-add component 4 (dim 3) to SVD_THEMES and revert comp_options
filter — with individual MPs averaged, dim 3 now shows real party
separation (the 'publieke voorzieningen vs marktwerking' axis).
- Scree plot y-axis now shows percentage of total variance instead of
raw L2-norms; hover also updated to show '% van totaal'.
sgeboers