Optimization2025/plot_step_aligned.py at main · dfy37/Optimization2025 · GitHub

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
import torch
import matplotlib.pyplot as plt
import matplotlib as mpl

mpl.rcParams.update({
    "font.size": 16,          # 默认文字
    "axes.titlesize": 16,     # 子图标题
    "axes.labelsize": 16,     # x/y 轴标题
    "xtick.labelsize": 16,     # x 轴刻度
    "ytick.labelsize": 16,     # y 轴刻度
    "legend.fontsize": 16,
})

h_sgd   = torch.load("hist_sgd.pt")
h_cover = torch.load("hist_cover.pt")
h_civr  = torch.load("hist_civr.pt")

fig, axes = plt.subplots(2, 1, figsize=(6, 6), sharex=True)

# =====================================================
# (1) Test Accuracy
# =====================================================
axes[0].plot(h_sgd["step"],   h_sgd["test_acc"],   label="ERM-SGD", linewidth=2)
axes[0].plot(h_cover["step"], h_cover["test_acc"], label="COVER",   linewidth=2)
axes[0].plot(h_civr["step"],  h_civr["test_acc"],  label="CIVR",    linewidth=2)

axes[0].set_ylabel("Test Accuracy (%)")
axes[0].legend()
axes[0].grid(alpha=0.3)

# =====================================================
# (2) Train CE Loss
# =====================================================
axes[1].plot(h_sgd["step"],   h_sgd["train_ce"], label="ERM-SGD", linewidth=2)
axes[1].plot(h_cover["step"], h_cover["train_ce"], label="COVER", linewidth=2)
axes[1].plot(h_civr["step"],  h_civr["train_ce"], label="CIVR", linewidth=2)

axes[1].set_xlabel("Oracle calls (samples)")
axes[1].set_ylabel("Train CE Loss")
axes[1].grid(alpha=0.3)

plt.tight_layout()
plt.savefig("step_aligned_acc_ce.png", dpi=200)
plt.show()