File size: 4,565 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
103
104
105
106
107
108
#!/usr/bin/env python3
"""3-D PCA of the ACT encoder space, before vs after contact.

Each phase gets its own PCA -- reaching and manipulation are dominated by
different directions, so a shared basis would misrepresent both. Contact onset is
detected per episode from finger torque, so "pre" is genuinely before the
GelSight touches anything.

For each phase: the default PC1-PC3 triple and the best-separating triple found
by scoring every 3-combination of the first 8 components on held-out episodes.
Showing both keeps the best-triple choice honest.
"""
import argparse, itertools
import numpy as np
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation, PillowWriter
from mpl_toolkits.mplot3d import Axes3D  # noqa: F401
from sklearn.decomposition import PCA
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler

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"}
TITLE = {"pre": "pre-contact  (GelSight untouched)", "post": "post-contact  (cable in hand)"}


def acc(Z, y, tr, cols):
    c = make_pipeline(StandardScaler(), LogisticRegression(max_iter=3000))
    c.fit(Z[tr][:, list(cols)], y[tr])
    return c.score(Z[~tr][:, list(cols)], y[~tr])


def draw(ax, Z, y, te, cols, title):
    for t in sorted(set(y), key=lambda s: DIA[s]):
        m = te & (y == t)
        ax.scatter(Z[m, cols[0]], Z[m, cols[1]], Z[m, cols[2]], s=11, alpha=0.55,
                   edgecolors="none", c=COLOR[DIA[t]], label=NAME[DIA[t]])
    ax.set_xlabel(f"PC{cols[0]+1}"); ax.set_ylabel(f"PC{cols[1]+1}"); ax.set_zlabel(f"PC{cols[2]+1}")
    ax.set_title(title, fontsize=11.5)
    ax.view_init(elev=18, azim=-60)


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--cache", default="outputs/pca/cache/cache_cable_contact.npz")
    ap.add_argument("--outdir", default="outputs/pca/contact")
    args = ap.parse_args()

    d = np.load(args.cache, allow_pickle=True)
    X, y, ep, ph = d["feats"], d["y"], d["eps"], d["phase"]
    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))

    plt.rcParams.update({"font.family": ["DejaVu Sans"], "font.size": 11})
    fig = plt.figure(figsize=(14, 11.5))
    store = {}

    for row, pname in enumerate(["pre", "post"]):
        w = ph == pname
        Xp, yp, trp = X[w], y[w], tr[w]
        Xc = Xp - Xp.mean(0)
        p = PCA(n_components=10).fit(Xc)
        Z = p.transform(Xc)
        tep = ~trp
        best = max(itertools.combinations(range(8), 3), key=lambda c: acc(Z, yp, trp, c))
        a_def, a_best = acc(Z, yp, trp, (0, 1, 2)), acc(Z, yp, trp, best)
        store[pname] = (Z, yp, tep, best, a_best)

        ax = fig.add_subplot(2, 2, row * 2 + 1, projection="3d")
        draw(ax, Z, yp, tep, (0, 1, 2), f"{TITLE[pname]}\nPC1-PC3 (default)  —  {a_def:.1%}")
        if row == 0:
            ax.legend(frameon=False, title="cable diameter", loc="upper left", fontsize=9.5)
        ax = fig.add_subplot(2, 2, row * 2 + 2, projection="3d")
        draw(ax, Z, yp, tep,  best,
             f"best triple  PC{best[0]+1}, PC{best[1]+1}, PC{best[2]+1}  —  {a_best:.1%}")

        print(f"  {pname:4s}  PC1-PC3 {a_def:.1%}   best PC{best[0]+1},PC{best[1]+1},PC{best[2]+1} {a_best:.1%}",
              flush=True)

    fig.tight_layout()
    out = f"{args.outdir}/cable_contact_pca3d.png"
    fig.savefig(out, dpi=150, bbox_inches="tight")
    print(f"\nwrote {out}")

    # Rotating views -- depth structure is invisible in a still 3-D scatter.
    for pname in ["pre", "post"]:
        Z, yp, tep, best, a = store[pname]
        f2 = plt.figure(figsize=(7, 6.4))
        ax = f2.add_subplot(111, projection="3d")
        draw(ax, Z, yp, tep, best,
             f"{TITLE[pname]}\nPC{best[0]+1}, PC{best[1]+1}, PC{best[2]+1}  —  {a:.1%}")
        ax.legend(frameon=False, title="cable diameter", loc="upper left", fontsize=9.5)
        FuncAnimation(f2, lambda i: (ax.view_init(elev=18, azim=i * 4), [])[1],
                      frames=90, interval=80).save(
            f"{args.outdir}/cable_contact_pca3d_{pname}.gif", writer=PillowWriter(fps=12))
        print(f"wrote {args.outdir}/cable_contact_pca3d_{pname}.gif")


if __name__ == "__main__":
    main()