Performative-Prediction-Real-Data/plotter.py at main · nadavru/Performative-Prediction-Real-Data · 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
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
109
110
111
112
113
114
115
116
117
118
119
120
import matplotlib.pyplot as plt
import numpy as np
import math
import os

#################################
#optimizers = ["RRM", "RRM + ADAM (0.1)", "RRM + ADAM (0.1) + lookahead", "RRM + ADAM (0.1) + learned lookahead", "RRM + ADAM (0.9)", "RRM + ADAM (0.9) + lookahead", "RRM + ADAM (0.9) + learned lookahead"]
optimizers = ["RRM", "RRM + ADAM (0.1)", "RRM + ADAM (0.1) + lookahead", "RRM + ADAM (0.1) + learned lookahead", "RRM + ADAM (0.9) + lookahead", "RRM + ADAM (0.9) + learned lookahead"]

data_folder = "GiveMeSomeCredit"
tran = "tran1"
meta = {"eps": -1000}
# eps: [0.01, 1, 100, 1000, -0.01, -1, -100, -1000]
seeds = list(range(1234,1244))

smooth = 10

#trans = [["tran1", {"eps": [0.1, 1, 25, 50]}], ["tran2", {}]]

result_folder = "results"
#################################

#all_colors = ["r","g","b","c","m","y","purple"]
all_colors = ["r","g","b","c","y","purple"]

theta_diffs = []
step_losses = []
perf_risks = []
perf_accs = []

meta_name = ""
for key, val in meta.items():
    meta_name += f"{key} {float(val)} , "
meta_name = meta_name[:-3]
for optimizer in optimizers:
    opt_theta_diffs = [] # [seeds,100]
    opt_step_losses = [] # [seeds,100]
    opt_perf_risks = [] # [seeds,101]
    opt_perf_accs = [] # [seeds,101]

    for seed in seeds:
        exp_folder = f"{result_folder}/{optimizer}/{data_folder}/{tran}/{meta_name}/{seed}"

        with open(f"{exp_folder}/theta_diffs.txt") as f:
            opt_theta_diffs.append([float(val) for val in f.read().splitlines()])
        with open(f"{exp_folder}/step_losses.txt") as f:
            opt_step_losses.append([float(val) for val in f.read().splitlines()])
        with open(f"{exp_folder}/perf_accs.txt") as f:
            opt_perf_accs.append([float(val) for val in f.read().splitlines()])
        with open(f"{exp_folder}/perf_risks.txt") as f:
            opt_perf_risks.append([float(val) for val in f.read().splitlines()])

    opt_theta_diffs = np.array(opt_theta_diffs).mean(0)
    opt_step_losses = np.array(opt_step_losses).mean(0)
    opt_perf_risks = np.array(opt_perf_risks).mean(0)
    opt_perf_accs = np.array(opt_perf_accs).mean(0)

    theta_diffs.append(opt_theta_diffs)
    step_losses.append(opt_step_losses)
    perf_risks.append(opt_perf_risks)
    perf_accs.append(opt_perf_accs)

theta_diffs = np.array(theta_diffs) # [optimizers,100]
step_losses = np.array(step_losses) # [optimizers,100]
perf_risks = np.array(perf_risks) # [optimizers,101]
perf_accs = np.array(perf_accs) # [optimizers,101]


kernel = np.full((smooth,), 1/smooth)
perf_accs = np.stack([np.convolve(perf_acc, kernel, "valid") for perf_acc in perf_accs])
perf_risks = np.stack([np.convolve(perf_risk, kernel, "valid") for perf_risk in perf_risks])

##################################################################################
for j, opt in enumerate(optimizers):
    plt.plot(perf_accs[j], color=all_colors[j], label=opt)

#plt.ylim(ymin=0)
plt.xlabel('Round')
plt.ylabel('Accuracy')
plt.title(f"Performative Accuracy")

plt.legend()
#plt.yscale("log")
plt.show()
##################################################################################
for j, opt in enumerate(optimizers):
    plt.plot(perf_risks[j], color=all_colors[j], label=opt)

#plt.ylim(ymin=0)
#plt.ylim(ymax=3)
plt.xlabel('Round')
plt.ylabel('Risk')
plt.title("Performative Risk")

plt.legend()
plt.yscale("log")
plt.show()
##################################################################################
'''for j, opt in enumerate(optimizers):
    plt.plot(theta_diffs[j], color=all_colors[j], label=opt)

plt.ylim(ymin=0)
plt.xlabel('Round')
plt.ylabel('Distance between thetas')
plt.title(f"Theta Differences")

plt.legend()
plt.show()'''
##################################################################################
'''for j, opt in enumerate(optimizers):
    plt.plot(step_losses[j], color=all_colors[j], label=opt)

plt.ylim(ymin=0)
plt.xlabel('Round')
plt.ylabel('Round loss')
plt.title(f"Losses")

plt.legend()
plt.show()'''
##################################################################################