人臉認證-ROC曲線繪製計算AUC和ACC
阿新 • • 發佈:2019-01-06
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from sklearn.metrics import roc_auc_score
scores = []
with open("result.txt") as f:
lines = f.readlines()
for line in lines:
line = line.strip("/r/n")
line = line.replace("[[","")
line = line .replace("]]","")
score = float(line)
scores.append(score)
y = []
for i in range(6000):
if i<3000: #N
y.append(0)
else:
y.append(1)
roc_x = []
roc_y = []
min_score = min(scores)
max_score = max(scores)
thr = np.linspace(min_score, max_score, 100)
FP = 0
TP = 0
TN = 0
N = 3000
P = 3000
acc_list = []
for(i, T) in enumerate(thr):
for i in range(0, len(scores)):
if(scores[i] > T):
if(y[i] == 1):
TP = TP + 1
if(y[i] == 0):
FP = FP + 1
else:
if(y[i] == 0)
TN = TN + 1
roc_x.append(FP/float(N))
roc_y.append(TP/float(P))
acc= (TP + TN)*1.0 /(N+P)
acc_list.append(acc)
FP = 0
TP =0
TN = 0
plt.plot(roc_x, roc_y, '--*b')
plt.savefig("roc.png")
auc = roc_auc_score(y, scores)
print "AUC:", auc
print "ACC:", max(acc_list)