60 lines
1.9 KiB
Plaintext
60 lines
1.9 KiB
Plaintext
import json
|
|
import numpy as np
|
|
import matplotlib.pyplot as plt
|
|
from matplotlib import cm
|
|
import copy
|
|
|
|
|
|
|
|
def main():
|
|
data = {}
|
|
with open("results/cifar10_fgsm.json", "r") as infile:
|
|
data = json.load(infile)
|
|
|
|
attack = data["attack"]
|
|
epsilons = data["epsilons"]
|
|
filters = data["filters"]
|
|
dataset = data["dataset"]
|
|
strength_count = len(filters[list(filters.keys())[0]][0])
|
|
|
|
strengths = np.arange(strength_count)
|
|
epsilons = np.array(epsilons)
|
|
|
|
# Assume constant step of strength and epsilon
|
|
dstrength = strengths[1] - strengths[0]
|
|
depsilon = epsilons[1] - epsilons[0]
|
|
|
|
# Make a grid from strengths and epsilons
|
|
strengths, epsilons = np.meshgrid(strengths,epsilons)
|
|
#strengths, epsilons = strengths.ravel(), epsilons.ravel()
|
|
|
|
colors = ('blue', 'orange', 'red', 'purple', 'green', 'yellow', 'brown')
|
|
|
|
z = np.zeros_like(strengths)
|
|
best_performance = np.zeros_like(strengths)
|
|
bottoms = np.zeros_like(strengths)
|
|
|
|
for i, filter in enumerate(filters):
|
|
performance = np.array(filters[filter])
|
|
best_performance = np.where(performance > best_performance, i, best_performance)
|
|
z = np.where(performance > z, performance, z)
|
|
print(best_performance)
|
|
|
|
for i, filter in enumerate(filters):
|
|
fig, ax = plt.subplots(subplot_kw={"projection": "3d"})
|
|
tops = np.where(best_performance == i, best_performance, 0)
|
|
print(strengths.shape, epsilons.shape, bottoms.shape, dstrength.shape, depsilon.shape, tops.shape)
|
|
ax.bar3d(strengths, epsilons, bottoms, dstrength, depsilon, tops, color=colors[i])
|
|
plt.show()
|
|
|
|
ax.view_init(90, -90, 0)
|
|
|
|
plt.legend()
|
|
plt.title(f"{filter} Performance")
|
|
plt.xlabel(f"{attack} Attack Strength ($\\epsilon$)")
|
|
plt.ylabel(f"{dataset} Classification Accuracy")
|
|
plt.show()
|
|
|
|
if __name__ == "__main__":
|
|
main()
|