File size: 3,438 Bytes
638ea38 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 | import numpy as np
import pandas as pd
from scipy import stats
from common import RES, OX, OS, hom, axes, centre, effects
Z = hom()
ag, ox, os_, ref = axes(Z.columns)
a, b = effects(centre(Z), ox, os_)
base = (a - b).abs()
rows = []
for how in ("mean", "trim"):
a2, b2 = effects(centre(Z, how), ox, os_)
m = base.notna() & (a2 - b2).notna()
rows.append(("centring", how, stats.spearmanr(base[m], (a2 - b2).abs()[m]).statistic, int(m.sum())))
Zc = centre(Z)
for drop in sorted(OX | OS):
keep = ag != drop
a2, b2 = effects(Zc, ox & keep, os_ & keep)
cd2 = (a2 - b2).abs()
m = base.notna() & cd2.notna()
rows.append(("leave-one-agent-out", drop, stats.spearmanr(base[m], cd2[m]).statistic, int(m.sum())))
naive_a, naive_b = Z.loc[:, ox].mean(axis=1), Z.loc[:, os_].mean(axis=1)
bline = Z.median(axis=1)
naive_cd = naive_a - naive_b
signed = a - b
m = base.notna()
rows.append(("identity N_ox = E_ox + b", "max |N_ox - (E_ox + b)|", float((naive_a - (a + bline)).abs().max()), int(m.sum())))
rows.append(("identity N_osm = E_osm + b", "max |N_osm - (E_osm + b)|", float((naive_b - (b + bline)).abs().max()), int(m.sum())))
rows.append(("slow-growth invariance", "max |naive - deconfounded|",
float((naive_cd[m] - signed[m]).abs().max()), int(m.sum())))
rows.append(("slow-growth invariance", "pearson r",
stats.pearsonr(naive_cd[m], signed[m]).statistic, int(m.sum())))
rref = Zc.loc[:, ref].mean(axis=1)
rows.append(("reference-contrast variant", "max |(E_ox - r) - (E_osm - r) - (E_ox - E_osm)|",
float(((a - rref) - (b - rref) - signed)[m].abs().max()), int(m.sum())))
rows.append(("baseline b vs mean of reference conditions", "spearman",
stats.spearmanr(bline, Z.loc[:, ref].mean(axis=1), nan_policy="omit").statistic, int(m.sum())))
rows.append(("per-stress effect: baseline vs reference-mean contrast", "spearman (oxidative)",
stats.spearmanr(a, a - rref, nan_policy="omit").statistic, int(m.sum())))
cond = pd.read_csv(RES / "conditions.csv")
batch = (cond.pool + ", " + cond.scanner).values
X = Z.values
B = np.full_like(X, np.nan)
for bt in np.unique(batch):
cols = np.where((batch == bt) & ref)[0]
B[:, batch == bt] = np.nanmedian(X[:, cols], axis=1)[:, None]
rows.append(("batch-matched baseline", f"non-stress conditions in batch {bt}", len(cols),
int(((batch == bt) & (ox | os_)).sum())))
ab, bb = effects(pd.DataFrame(X - B, index=Z.index, columns=Z.columns), ox, os_)
for name, x, y in (("spearman E_ox", a, ab), ("spearman E_osm", b, bb), ("spearman D", signed, ab - bb),
("spearman |D|", base, (ab - bb).abs())):
rows.append(("batch-matched baseline", name, stats.spearmanr(x, y, nan_policy="omit").statistic, int(m.sum())))
sev = cond.severity_frac_z_gt2.values
keep = ref & (sev < np.quantile(sev[ref], 0.9))
b_sev = np.nanmedian(X[:, keep], axis=1)
as_, bs_ = Z.loc[:, ox].mean(axis=1) - b_sev, Z.loc[:, os_].mean(axis=1) - b_sev
for name, x, y in (("spearman E_ox", a, as_), ("spearman E_osm", b, bs_)):
rows.append(("baseline excluding the 10% most severe non-stress conditions", name,
stats.spearmanr(x, y, nan_policy="omit").statistic, int(keep.sum())))
out = pd.DataFrame(rows, columns=["analysis", "variant", "value", "n"])
out.to_csv(RES / "sensitivity.csv", index=False)
|