motief/tests/test_political_compass.py

import numpy as np
import types
import sys

import pytest


# ---------------------------------------------------------------------------
# Helpers shared by orientation tests
# ---------------------------------------------------------------------------


def _make_fake_traj(aligned):
    fake = types.SimpleNamespace()
    fake._load_window_ids = lambda db: list(aligned.keys())
    fake._load_mp_vectors_for_window = lambda db, w: aligned.get(w, {})
    fake._procrustes_align_windows = lambda x: aligned
    return fake


def test_compute_2d_axes_pca_synthetic(monkeypatch):
    """Synthetic test for compute_2d_axes using patched alignment helper."""

    # Create a fake trajectory module with required helpers
    fake_traj = types.SimpleNamespace()

    # _load_window_ids should return ordered windows
    fake_traj._load_window_ids = lambda db: ["w1", "w2"]

    # Provide aligned vectors directly
    aligned = {
        "w1": {"Alice": np.array([1.0, 0.0, 0.0]), "Bob": np.array([0.0, 1.0, 0.0])},
        "w2": {"Alice": np.array([0.8, 0.2, 0.0]), "Bob": np.array([0.1, 0.9, 0.0])},
    }

    # _load_mp_vectors_for_window returns the pre-aligned vectors (needed for padding step)
    fake_traj._load_mp_vectors_for_window = lambda db, w: aligned.get(w, {})

    fake_traj._procrustes_align_windows = lambda x: aligned

    # Insert fake module into sys.modules for import by analysis.political_axis
    monkeypatch.setitem(sys.modules, "analysis.trajectory", fake_traj)

    # Now import the function under test
    from analysis.political_axis import compute_2d_axes

    positions_by_window, axis_def = compute_2d_axes(
        db_path="dummy", window_ids=["w1", "w2"], method="pca"
    )

    assert "w1" in positions_by_window and "w2" in positions_by_window
    for wid in ("w1", "w2"):
        for name, coord in positions_by_window[wid].items():
            assert len(coord) == 2
            assert np.isfinite(coord[0]) and np.isfinite(coord[1])

    assert axis_def.get("method") == "pca"


def test_pca_axis_orientation(monkeypatch):
    """PCA axes must be oriented so right parties score higher on X and
    progressive parties score higher on Y than their respective opposites.

    We construct a minimal vote-matrix world where:
    - Right MPs (PVV, VVD members) cluster in one direction on dim-0.
    - Left MPs (SP, GroenLinks-PvdA members) cluster in the opposite direction.
    - Progressive MPs cluster on dim-1; conservative MPs on the opposite side.

    The orientation logic in compute_2d_axes should flip axis signs so that
    right_x > left_x and prog_y > cons_y regardless of the raw SVD sign.
    """
    # Build vectors so that right parties are at +1 on dim-0 and
    # progressive parties are at +1 on dim-1.
    # We deliberately negate them to test that auto-orient flips them back.
    # Right/left use magnitude 3, prog/cons use magnitude 1 so that dim-0
    # dominates PCA variance — ensuring PC1 = left-right axis, PC2 = prog-cons.
    right_vec = np.array([-3.0, 0.0, 0.0])  # intentionally negative on dim-0
    left_vec = np.array([3.0, 0.0, 0.0])  # intentionally positive on dim-0
    prog_vec = np.array([0.0, -1.0, 0.0])  # intentionally negative on dim-1
    cons_vec = np.array([0.0, 1.0, 0.0])  # intentionally positive on dim-1

    aligned = {
        "w1": {
            # Right-leaning MPs
            "Wilders, G.": right_vec,
            "Rutte, M.": right_vec + np.array([0.0, 0.0, 0.05]),
            # Left-leaning MPs
            "Marijnissen, L.": left_vec,
            "Klever, A.": left_vec + np.array([0.0, 0.0, 0.05]),
            # Progressive MPs
            "Bromet, L.": prog_vec,
            "Nijboer, H.": prog_vec + np.array([0.0, 0.0, -0.05]),
            # Conservative MPs
            "Segers, G.": cons_vec,
            "Omtzigt, P.": cons_vec + np.array([0.0, 0.0, -0.05]),
        }
    }

    # mp_metadata rows used by the orientation code (party affiliation)
    mp_metadata = [
        ("Wilders, G.", "PVV"),
        ("Rutte, M.", "VVD"),
        ("Marijnissen, L.", "SP"),
        ("Klever, A.", "GroenLinks-PvdA"),
        ("Bromet, L.", "GroenLinks-PvdA"),
        ("Nijboer, H.", "SP"),
        ("Segers, G.", "CDA"),
        ("Omtzigt, P.", "Nieuw Sociaal Contract"),
    ]

    fake_traj = _make_fake_traj(aligned)
    monkeypatch.setitem(sys.modules, "analysis.trajectory", fake_traj)

    # Patch duckdb so the orientation helper can fetch mp_metadata
    import types as _types

    fake_conn = _types.SimpleNamespace(
        execute=lambda q: _types.SimpleNamespace(fetchall=lambda: mp_metadata),
        close=lambda: None,
    )
    import duckdb as _duckdb

    monkeypatch.setattr(_duckdb, "connect", lambda db_path, **kw: fake_conn)

    # Need to reload the module so monkeypatched sys.modules takes effect
    import importlib, analysis.political_axis as _ax

    importlib.reload(_ax)
    from analysis.political_axis import compute_2d_axes

    positions_by_window, axis_def = compute_2d_axes(
        db_path="dummy", window_ids=["w1"], method="pca"
    )

    pos = positions_by_window["w1"]

    # X-axis: right parties should score higher than left parties
    right_x = np.mean([pos["Wilders, G."][0], pos["Rutte, M."][0]])
    left_x = np.mean([pos["Marijnissen, L."][0], pos["Klever, A."][0]])
    assert right_x > left_x, (
        f"Expected right parties (x={right_x:.3f}) > left parties (x={left_x:.3f}) on X-axis"
    )

    # Y-axis: progressive parties should score higher than conservative parties
    prog_y = np.mean([pos["Bromet, L."][1], pos["Nijboer, H."][1]])
    cons_y = np.mean([pos["Segers, G."][1], pos["Omtzigt, P."][1]])
    assert prog_y > cons_y, (
        f"Expected progressive parties (y={prog_y:.3f}) > conservative parties (y={cons_y:.3f}) on Y-axis"
    )