File size: 4,104 Bytes
3c41fb5 | 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 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 | #!/usr/bin/env python3
"""Cable clusters in the ACT encoder representation, ordered by cable diameter.
Figure only -- titles and annotations are kept minimal so the plot can sit in a
slide or paper with its own caption. Numbers needed for that caption (held-out
R^2, per-class predicted diameter) are printed to stdout instead of drawn.
left : unsupervised PCA of the 512-d encoder feature, on the two components
that actually carry cable identity (PC1/PC2 are dominated by arm pose)
right : ridge regression from the same features onto physical diameter, fitted
on training episodes -- if thickness is encoded as a quantity, 2.6 mm
lands between 1.7 and 4.5 without the ordering ever being supplied
Episode-level split throughout: frames inside one episode are near-duplicates,
so a frame-level split would leak the answer.
"""
import argparse
import numpy as np
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
from sklearn.decomposition import PCA
from sklearn.linear_model import RidgeCV
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import make_pipeline
DIA = {"cable 1.70 sq": 1.7, "blue cable": 2.6, "cable 3.30 sq": 4.5}
NAME = {1.7: "1.7 mm", 2.6: "2.6 mm", 4.5: "4.5 mm"}
COLOR = {1.7: "#C05621", 2.6: "#2B6CB0", 4.5: "#2F855A"}
def sep_ratio(Z, y, k):
g = [Z[y == t, k] for t in sorted(set(y))]
return np.var([x.mean() for x in g]) / (np.mean([x.var() for x in g]) + 1e-12)
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--cache", default="outputs/pca/cache_cable_feats.npz")
ap.add_argument("--out", default="outputs/pca/cable_diameter_clusters.png")
args = ap.parse_args()
d = np.load(args.cache, allow_pickle=True)
X, y, ep = d["feats"], d["y"], d["eps"]
dia = np.array([DIA[t] for t in y])
rng = np.random.default_rng(0)
tr = np.zeros(len(y), bool)
for t in np.unique(y):
e = np.unique(ep[y == t])
tr |= np.isin(ep, rng.choice(e, size=int(len(e) * 0.7), replace=False))
te = ~tr
plt.rcParams.update({"font.family": ["DejaVu Sans"], "font.size": 12})
fig, axes = plt.subplots(1, 2, figsize=(13, 5.6))
Xc = X - X.mean(0)
p = PCA(n_components=30).fit(Xc)
Z = p.transform(Xc)
ratios = np.array([sep_ratio(Z, y, k) for k in range(30)])
i, j = np.argsort(ratios)[::-1][:2]
for t in sorted(set(y), key=lambda s: DIA[s]):
m = te & (y == t)
axes[0].scatter(Z[m, i], Z[m, j], s=16, alpha=0.6, edgecolors="none",
c=COLOR[DIA[t]], label=NAME[DIA[t]])
axes[0].set_xlabel(f"PC{i+1}")
axes[0].set_ylabel(f"PC{j+1}")
axes[0].legend(frameon=False)
reg = make_pipeline(StandardScaler(), RidgeCV(alphas=np.logspace(-2, 4, 25))).fit(X[tr], dia[tr])
pred = reg.predict(X)
for t in sorted(set(y), key=lambda s: DIA[s]):
m = te & (y == t)
axes[1].scatter(pred[m], Z[m, j], s=16, alpha=0.6, edgecolors="none",
c=COLOR[DIA[t]], label=NAME[DIA[t]])
for v in (1.7, 2.6, 4.5):
axes[1].axvline(v, color=COLOR[v], lw=1.1, ls="--", alpha=0.7)
axes[1].set_xlabel("predicted cable diameter [mm]")
axes[1].set_ylabel(f"PC{j+1}")
axes[1].set_xticks([1.7, 2.6, 4.5])
axes[1].legend(frameon=False)
for ax in axes:
ax.spines[["top", "right"]].set_visible(False)
fig.tight_layout()
fig.savefig(args.out, dpi=160, bbox_inches="tight")
# Caption material, printed rather than drawn.
print(f"wrote {args.out}\n")
print(f" left panel : PC{i+1} vs PC{j+1} "
f"({p.explained_variance_ratio_[i]:.1%} and {p.explained_variance_ratio_[j]:.1%} of variance)")
print(f" right panel : held-out R^2 = {reg.score(X[te], dia[te]):.2f}")
for t in sorted(set(y), key=lambda s: DIA[s]):
m = te & (y == t)
print(f" true {DIA[t]:>4} mm -> predicted {pred[m].mean():.2f} +- {pred[m].std():.2f} mm (n={m.sum()})")
print(f" frames: {te.sum()} held-out of {len(y)}; episode-level split")
if __name__ == "__main__":
main()
|