wassname/moral-maps
1
0
Fork 0
mirror of https://github.com/wassname/moral-maps.git synced 2026-09-09 11:27:22 +08:00
Code Issues Packages Projects Releases Wiki Activity

plots: drop redundant foundation_dcontrast dumbbell

Browse Source 
The range plot already shows the per-foundation steer (human cloud + AI base
dot + +C/-C arrows) with the human anchor the dumbbell lacks; dcontrast plotted
the same delta on the C-nats scale and failed the eraser test. C stays the
measurement in the CSV/foundations table, just not a separate panel.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
This commit is contained in:
wassname
2026-06-26 04:13:12 +08:00
co-authored by Claudypoo
parent 8ab02adf63
commit 3fcdda4516
1 changed files with 0 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/plot_steer_showcase.py
-41
View File
@@ -106,43 +106,6 @@ def read_profiles(run_dir: Path, name: str, dims: list[str], value_col: str = "m
return {c: np.array([d[f] for f in dims]) for c, d in by_c.items()}, pmass
def plot_ordinal_steer(run_dir: Path, out: Path, name: str, vec_label: str, C: float) -> Path:
"""The steer-effect plot: per-factor change in the logit contrast C (steered minus base), +C (red)
vs -C (blue). This is the ordinal twin of the MFV dlogit dumbbell, and the figure that actually
shows what we steered for -- the E range saturates and hides it, the contrast does not. C_sd in the
CSV gives a per-factor SE over the 18 items for the error bars (uncertainty, not just the mean)."""
instr = get_instrument(name)
dims = instr.dimensions
cprof, pmass = read_profiles(run_dir, name, dims, value_col="C")
sdprof, _ = read_profiles(run_dir, name, dims, value_col="C_sd")
cs = sorted(cprof)
base = cprof[0.0]
pos_c = 1.0 if 1.0 in cprof else max(cs)
neg_c = -1.0 if -1.0 in cprof else min(cs)
n_items = 18 # mfq2/factor; SE = sd / sqrt(n). (big5/16pf/humor differ but this is a rough band.)
dpos, dneg = cprof[pos_c] - base, cprof[neg_c] - base
se = sdprof[0.0] / n_items ** 0.5
y = np.arange(len(dims))[::-1]
fig, ax = plt.subplots(figsize=(6.4, 4.2))
ax.axvline(0, color="0.6", lw=0.8, zorder=1)
POS, NEG = T.maps.POS_COL, T.maps.NEG_COL
for fi, yi in zip(range(len(dims)), y):
ax.plot([dneg[fi], dpos[fi]], [yi, yi], color="0.8", lw=1.0, zorder=2)
ax.errorbar(dpos[fi], yi, xerr=1.96 * se[fi], fmt="o", color=POS, ms=5, capsize=2, zorder=3)
ax.errorbar(dneg[fi], yi, xerr=1.96 * se[fi], fmt="o", color=NEG, ms=5, capsize=2, zorder=3)
ax.set_yticks(y); ax.set_yticklabels([d.capitalize() for d in dims])
ax.set_xlabel("Delta contrast C vs base (nats; agree-minus-disagree, rank-weighted)")
ax.set_title(f"Steered {instr.display}: {vec_label}", fontsize=11)
ax.scatter([], [], color=POS, label=f"+C={C:+.2f}")
ax.scatter([], [], color=NEG, label=f"-C={-C:+.2f}")
ax.legend(fontsize=8, loc="best")
ax.spines[["top", "right"]].set_visible(False)
fig.tight_layout()
path = T.maps.save_both(fig, out / name, "foundation_dcontrast")
plt.close(fig)
return path
def plot_ordinal(run_dir: Path, out: Path, name: str, vec_label: str, C: float) -> list[Path]:
instr = get_instrument(name)
dims = instr.dimensions
@@ -202,10 +165,6 @@ def plot_ordinal(run_dir: Path, out: Path, name: str, vec_label: str, C: float)
figz = T.maps.plot_range_zoom(instr, dims, coh_cs, prof_coh, humans, vec_label)
paths.append(T.maps.save_both(figz, out / name, "range_zoom"))
plt.close(figz)
# steer-effect plot in the sensitive contrast readout (the E map/range above are for human
# comparison; this is "did the steer move it"). Parallels the MFV dlogit dumbbell.
paths.append(plot_ordinal_steer(run_dir, out, name, vec_label, C))
return paths