wassname/moral-maps
1
0
Fork 0
mirror of https://github.com/wassname/moral-maps.git synced 2026-09-11 12:20:38 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add Inglehart-Welzel zone hulls + outlier labels to ipsative maps

Browse Source 
Echoes the Economist WVS 'Godless hippies' chart: shaded convex-hull blobs per
IW cultural zone (inline 2D hull, no scipy dep so the maps extra stays
matplotlib-only) and bold-first labels for named outliers. Caller owns the zone
taxonomy + name/ISO2 normalizer, fails loud on unmapped countries; the corrupt
'(nu' big5 row is explicitly excluded with a warning.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
This commit is contained in:
wassname
2026-07-04 19:16:41 +08:00
co-authored by Claudypoo
parent 3ae1fd48c9
commit 0e7178dfbd
2 changed files with 163 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/plot_steer_showcase.py
+87 -2
View File
@@ -35,11 +35,94 @@ matplotlib.use("Agg")
import matplotlib.pyplot as plt
import numpy as np
from loguru import logger
import tinymfv as T
from tinymfv import get_instrument
ORDINAL = ["mfq2", "big5", "16pf", "humor_styles"]
# --- Inglehart-Welzel cultural zones (for the map's zone hulls) ---------------------------------
# WVS Wave 7 nine-cluster taxonomy. Membership is VALUE-based not geographic, so a handful are
# judgment calls: Ireland->English-Speaking, Switzerland->Protestant Europe, Philippines->Latin
# America, South Africa/Turkey->African-Islamic, India/Pakistan/Thailand->South Asia. Source: WVS
# Findings + en.wikipedia.org/wiki/Inglehart-Welzel_cultural_map_of_the_world. -- added by Claude
IW_ZONE = {
# English-Speaking
"United States": "English-Speaking", "Great Britain": "English-Speaking",
"Australia": "English-Speaking", "Canada": "English-Speaking",
"New Zealand": "English-Speaking", "Ireland": "English-Speaking",
# Protestant Europe
"Germany": "Protestant Europe", "Sweden": "Protestant Europe", "Norway": "Protestant Europe",
"Denmark": "Protestant Europe", "Netherlands": "Protestant Europe",
"Finland": "Protestant Europe", "Switzerland": "Protestant Europe",
# Catholic Europe
"France": "Catholic Europe", "Belgium": "Catholic Europe", "Italy": "Catholic Europe",
"Spain": "Catholic Europe", "Poland": "Catholic Europe", "Portugal": "Catholic Europe",
"Croatia": "Catholic Europe",
# Orthodox / Ex-Communist
"Russia": "Orthodox", "Ukraine": "Orthodox", "Bulgaria": "Orthodox", "Serbia": "Orthodox",
"Greece": "Orthodox", "Romania": "Orthodox", "Bosnia & Herzegovina": "Orthodox",
"Hungary": "Orthodox",
# Baltic
"Latvia": "Baltic", "Estonia": "Baltic",
# Confucian
"Japan": "Confucian", "China": "Confucian", "South Korea": "Confucian",
"Hong Kong": "Confucian", "Vietnam": "Confucian", "Singapore": "Confucian",
# Latin America
"Argentina": "Latin America", "Chile": "Latin America", "Colombia": "Latin America",
"Mexico": "Latin America", "Peru": "Latin America", "Brazil": "Latin America",
"Ecuador": "Latin America", "Philippines": "Latin America",
# African-Islamic
"Egypt": "African-Islamic", "Kenya": "African-Islamic", "Morocco": "African-Islamic",
"Nigeria": "African-Islamic", "Saudi Arabia": "African-Islamic",
"United Arab Emirates": "African-Islamic", "Turkey": "African-Islamic",
"Iran": "African-Islamic", "Indonesia": "African-Islamic", "Malaysia": "African-Islamic",
"South Africa": "African-Islamic",
# South Asia
"India": "South Asia", "Pakistan": "South Asia", "Thailand": "South Asia",
}
# raw country string (as it appears in the human CSVs) -> canonical IW_ZONE key. Our CSVs mix full
# names (mfv/mfq2/humor, with a "Columbia" typo) and ISO2 codes (big5/16pf). A `None` value marks a
# row we KNOW is corrupt and deliberately exclude from hulls (surfaced by a loud warning, not a
# silent drop); an unrecognised country that is NOT here falls through to IW_ZONE and KeyErrors.
_COUNTRY_CANON = {
"AE": "United Arab Emirates", "AU": "Australia", "BR": "Brazil", "CA": "Canada",
"CN": "China", "DE": "Germany", "DK": "Denmark", "EC": "Ecuador", "ES": "Spain",
"FI": "Finland", "FR": "France", "GB": "Great Britain", "GR": "Greece", "HK": "Hong Kong",
"HR": "Croatia", "ID": "Indonesia", "IE": "Ireland", "IN": "India", "IT": "Italy",
"MX": "Mexico", "MY": "Malaysia", "NL": "Netherlands", "NO": "Norway", "NZ": "New Zealand",
"PH": "Philippines", "PK": "Pakistan", "PL": "Poland", "RO": "Romania", "SE": "Sweden",
"SG": "Singapore", "TH": "Thailand", "TR": "Turkey", "US": "United States", "ZA": "South Africa",
"Columbia": "Colombia", "UAE": "United Arab Emirates",
"(nu": None, # corrupt big5 row (n=369); country unidentifiable from the aggregate CSV
}
# The named outliers on the Economist chart, bolded on our maps where present. -- added by Claude
ECONOMIST_OUTLIERS = {"China", "South Korea", "United States", "Great Britain", "Japan",
"Nigeria", "Pakistan", "Sweden"}
def zones_for(countries: list[str]) -> tuple[dict[str, list[str]], set[str]]:
"""Group verbatim country strings by IW zone + the subset to emphasize. Fails loud (KeyError)
on a country absent from the taxonomy so a name-normalization bug can't silently drop a dot from
its hull; a `None` canon (known-corrupt row) is excluded with a warning instead."""
groups: dict[str, list[str]] = {}
dropped: list[str] = []
emph: set[str] = set()
for c in countries:
canon = _COUNTRY_CANON.get(c, c)
if canon is None:
dropped.append(c)
continue
groups.setdefault(IW_ZONE[canon], []).append(c)
if canon in ECONOMIST_OUTLIERS:
emph.add(c)
if dropped:
logger.warning(f"excluded known-unmapped countries from zone hulls: {dropped}")
return groups, emph
def _frac(x, scale_max: int) -> np.ndarray:
return (np.asarray(x, float) - 1) / (scale_max - 1)
@@ -191,10 +274,11 @@ def plot_ordinal(run_dir: Path, out: Path, name: str, vec_label: str, C: float,
else:
respondents, haze = None, human_haze(instr)
traj = {c: _frac(prof_c[c], instr.scale_max) for c in coh_cs}
zones, emph = zones_for(countries)
figm = T.maps.plot_ipsative_pca(instr, dims, countries, Mfrac,
_frac(base, instr.scale_max), _frac(pos, instr.scale_max),
_frac(neg, instr.scale_max), respondents=respondents, haze=haze,
traj=traj, labels=labels)
traj=traj, zones=zones, emphasize=emph, labels=labels)
figm.axes[0].set_title(f"{instr.display}: humans vs LLMs steered for {vec_label}", fontsize=10)
paths = [T.maps.save_both(figm, out / name, "map_pca_ipsative")]
plt.close(figm)
@@ -268,8 +352,9 @@ def plot_mfv_map(run_dir: Path, out: Path, vec_label: str, C: float, coh_cs: lis
neg_c = min(c for c in coh_cs if c < 0.0)
labels = ("base (c=0)", f"c={pos_c:+g}", f"c={neg_c:+g}")
traj = {c: prof[c] for c in coh_cs}
zones, emph = zones_for(countries)
fig = T.maps.plot_ipsative_pca(_MFV_INSTR, founds, countries, M, prof[0.0], prof[pos_c], prof[neg_c],
traj=traj, labels=labels)
traj=traj, zones=zones, emphasize=emph, labels=labels)
fig.axes[0].set_title(f"MFV vignettes: humans vs LLMs steered for {vec_label}", fontsize=10)
path = T.maps.save_both(fig, out / "mfv", "map_pca_ipsative")
plt.close(fig)
src/tinymfv/maps.py
+76 -8
View File
@@ -74,6 +74,41 @@ GROUP_PITCH = 1.55 # x-distance between factors; > pair width so each (soc
# base is neutral, +c is red, -c is blue, human societies are grey.
C_BASE, C_HON, C_DIS, C_HUM = "#111111", POS_COL, NEG_COL, "#888888"
# Stable per-zone fill colors so the SAME Inglehart-Welzel zone reads the same across every
# instrument's map (research consistency). Keyed by the zone names the caller passes; an unlisted
# zone falls back to grey. -- added by Claude
ZONE_COLORS = {
"English-Speaking": "#4e79a7",
"Protestant Europe": "#59a14f",
"Catholic Europe": "#8cd17d",
"Orthodox": "#b6992d",
"Baltic": "#499894",
"Confucian": "#e15759",
"Latin America": "#f28e2b",
"African-Islamic": "#9c755f",
"South Asia": "#b07aa1",
}
def convex_hull(pts: np.ndarray) -> np.ndarray:
"""2D convex-hull vertices (CCW) via Andrew's monotone chain. Inline instead of scipy so the
`maps` install extra stays matplotlib-only (scipy is dev-only). pts (n,2) -> polygon (m,2)."""
P = sorted(map(tuple, pts.tolist()))
if len(P) <= 2:
return np.array(P, dtype=float)
cross = lambda o, a, b: (a[0] - o[0]) * (b[1] - o[1]) - (a[1] - o[1]) * (b[0] - o[0])
lower: list = []
for p in P:
while len(lower) >= 2 and cross(lower[-2], lower[-1], p) <= 0:
lower.pop()
lower.append(p)
upper: list = []
for p in reversed(P):
while len(upper) >= 2 and cross(upper[-2], upper[-1], p) <= 0:
upper.pop()
upper.append(p)
return np.array(lower[:-1] + upper[:-1], dtype=float)
def save_both(fig, fig_dir: Path, stem: str, dpi: int = 200) -> Path:
fig_dir.mkdir(parents=True, exist_ok=True)
@@ -180,6 +215,7 @@ def plot_ipsative_pca(instr: Instrument, dims: list[str], countries: list[str],
*, respondents: np.ndarray | None = None, haze: np.ndarray | None = None,
traj: dict[float, np.ndarray] | None = None, traj_incoherent: set | None = None,
boots: dict | None = None,
zones: dict[str, list[str]] | None = None, emphasize: set[str] | None = None,
labels: tuple[str, str, str] = ("baseline (c=0)", "honest (c=+2)", "dishonest (c=-2)")):
"""Ipsative culture map. M is societies x K (0-1 fraction); base / pos / neg are the length-K
fraction vectors for the base model and its two steer poles (or None). `labels` is the legend
@@ -194,8 +230,12 @@ def plot_ipsative_pca(instr: Instrument, dims: list[str], countries: list[str],
path through PC space. Public README plots pass only the coherent prefix; incoherent c values
are omitted.
`traj_incoherent` is the subset of those c whose admin pmass fell below the coherence floor --
drawn hollow. `boots` optionally maps 'base'/'honest'/'dis' -> (n x K) bootstrap matrices. Returns
the Figure."""
drawn hollow. `boots` optionally maps 'base'/'honest'/'dis' -> (n x K) bootstrap matrices.
`zones` maps an Inglehart-Welzel zone name to the subset of `countries` (verbatim strings) in it;
each zone with >=3 members gets a shaded convex hull (echoes the Economist WVS map's zone blobs),
testing whether moral-foundation space recovers the WVS clusters. `emphasize` is a subset of
`countries` labelled bold-first so named outliers (China, US, Sweden...) always survive the
label-collision drop. Returns the Figure."""
try:
import textalloc as ta
except ImportError:
@@ -216,16 +256,44 @@ def plot_ipsative_pca(instr: Instrument, dims: list[str], countries: list[str],
Pi = (cloud @ Pc - mu) @ Vt[:2].T
ax.scatter(Pi[:, 0], Pi[:, 1], s=4, c="#8f8a7e", alpha=0.14, edgecolors="none",
zorder=1, rasterized=True)
# Inglehart-Welzel zone hulls: a shaded convex blob per zone with >=3 member societies, drawn
# UNDER the society dots (zorder<3). The zone name sits at the hull centroid in grey, echoing the
# Economist WVS map. A 2-member zone has no polygon, so it's shown as its connecting segment.
if zones:
cidx = {c: i for i, c in enumerate(countries)}
for zname, members in zones.items():
mi = [cidx[c] for c in members if c in cidx]
if len(mi) < 2:
continue
zpts = P[mi]
zcol = ZONE_COLORS.get(zname, "#888888")
if len(mi) >= 3:
hull = convex_hull(zpts)
ax.add_patch(plt.Polygon(hull, closed=True, facecolor=zcol, edgecolor=zcol,
alpha=0.13, lw=1.0, zorder=1.6))
ax.plot(*np.vstack([hull, hull[:1]]).T, color=zcol, lw=1.0, alpha=0.45, zorder=1.7)
else:
ax.plot(zpts[:, 0], zpts[:, 1], color=zcol, lw=1.2, alpha=0.5, zorder=1.7)
cx, cy = zpts.mean(0)
ax.text(cx, cy, zname, fontsize=8.5, color="#6b6b6b", ha="center", va="center",
style="italic", zorder=2, alpha=0.85)
ax.scatter(P[:, 0], P[:, 1], s=26, c=C_HUM, alpha=0.7, edgecolors="white", linewidths=0.5, zorder=3)
# Society labels: each 2-letter ISO code is pinned RIGHT NEXT to its dot (small fixed offset, no
# leader line). A label is dropped entirely if its box would collide with an already-placed one --
# better an omitted code than one flung far from its point. No relocation, no arrows.
# Society labels: each name/ISO code is pinned RIGHT NEXT to its dot (small fixed offset, no
# leader line). A label is dropped if its box would collide with an already-placed one -- better an
# omitted code than one flung far from its point. `emphasize` countries are placed FIRST (so they
# win contested space) and drawn bold+dark, so the named outliers always survive the drop.
fig.canvas.draw()
renderer = fig.canvas.get_renderer()
placed_boxes = []
for i in np.argsort(P[:, 0]): # left-to-right; leftmost wins contested space
t = ax.annotate(countries[i], (P[i, 0], P[i, 1]), fontsize=7, color="#555555",
xytext=(3, 2), textcoords="offset points", zorder=6)
emph = emphasize or set()
order_lr = list(np.argsort(P[:, 0])) # left-to-right; leftmost wins contested space
order = [i for i in order_lr if countries[i] in emph] + [i for i in order_lr if countries[i] not in emph]
for i in order:
is_e = countries[i] in emph
t = ax.annotate(countries[i], (P[i, 0], P[i, 1]),
fontsize=8.5 if is_e else 7, color="#111111" if is_e else "#555555",
fontweight="bold" if is_e else "normal",
xytext=(3, 2), textcoords="offset points", zorder=7 if is_e else 6)
bb = t.get_window_extent(renderer)
if any(bb.overlaps(b) for b in placed_boxes):
t.remove()