-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathlocalsurrogate.py
executable file
·167 lines (138 loc) · 6.02 KB
/
localsurrogate.py
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
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
import numpy as np
import pandas as pd
import copy
from sklearn.linear_model import Ridge, RidgeClassifier
import matplotlib.pyplot as plt
from matplotlib import style
style.use("ggplot")
class localSurrogate():
def __init__(self, dataset, instance, attr, clf):
self.instance = instance
self.attr = attr
self.clf = clf
self.dataset = np.array(dataset)
self.mean = np.sum(dataset, axis=0)/len(dataset)
self.sigma = np.sqrt(np.sum((dataset-self.mean)*(dataset-self.mean), axis=0)/len(dataset))
self.instance_trafo = self.transform([instance[attr[0]], instance[attr[1]]])
def get_explainer(self):
return self.explainer
def get_name(self):
return 'local surrogate'
def rand_sphere(self, a0, a1):
dim = 2
x = np.random.normal(0,1,dim)
u = np.power(np.random.uniform(np.power(a0,dim),np.power(a1,dim)), (1/dim))
x = u * x / np.sqrt(np.sum(x*x))
return x
def sampling(self, nsample, a0, a1):
Z = []
for i in range(0,nsample):
z = self.rand_sphere(a0, a1)
Z.append([])
Z[i].append(i)
Z[i].append(z[0])
Z[i].append(z[1])
inst_trafo = self.transform([self.instance[self.attr[0]], self.instance[self.attr[1]]])
for i in range(0, nsample):
coord1 = Z[i][1] + inst_trafo[0]
coord2 = Z[i][2] + inst_trafo[1]
z = self.retransform([coord1, coord2])
Z[i][1] = z[0]
Z[i][2] = z[1]
dummy = copy.copy(self.instance)
dummy[self.attr[0]] = z[0]
dummy[self.attr[1]] = z[1]
coord3 = np.array(self.clf.predict_proba(np.array(dummy).reshape(1, -1))[0])[1]
coord3 = int(coord3 >= 0.5)
Z[i].append(coord3)
Z = np.array(Z)
# plt.scatter(Z[:,1], Z[:,2], c=['limegreen'*int(Z[i,3])+'tomato'*(1-int(Z[i,3])) for i in range(0,len(Z[:,3]))], cmap = plt.cm.Paired, marker='.', s=25)
# plt.show()
return Z
def transform(self, z):
return [(z[0] - self.mean[self.attr[0]])/self.sigma[self.attr[0]], (z[1] - self.mean[self.attr[1]])/self.sigma[self.attr[1]]]
def retransform(self, z):
return [z[0] * self.sigma[self.attr[0]] + self.mean[self.attr[0]], z[1] * self.sigma[self.attr[1]] + self.mean[self.attr[1]]]
def growingSpheres(self, nsample=60, eta=1.0):
check = 1
while check > 0:
Z = np.array(self.sampling(nsample=nsample, a0=0, a1=eta))
check = np.sum(Z[:,3])
if check > 0:
eta = eta/2
a0 = eta
a1 = 2*eta
check = 0
while check < 1:
Z = np.array(self.sampling(nsample=nsample, a0=a0, a1=a1))
check = np.sum(Z[:,3])
if check < 1:
a0 = a1
a1 = a1 + eta
#choice = int(np.random.choice(np.array(Z[Z[:,3]==1])[:,0]))
cands = np.array(Z[Z[:,3]==1])[:,0]
inst_trafo = self.transform([self.instance[self.attr[0]], self.instance[self.attr[1]]])
min_dist = 1000
min_cand = 0
for i in list(cands):
z = [Z[int(i),1], Z[int(i),2]]
z_trafo = self.transform(z)
new_dist = np.sqrt((inst_trafo[0] - z_trafo[0])*(inst_trafo[0] - z_trafo[0]) + (inst_trafo[1] - z_trafo[1])*(inst_trafo[1] - z_trafo[1]))
if min_dist > new_dist:
min_dist = new_dist
min_cand = int(i)
choice = min_cand
self.mittelpunkt = self.transform([Z[choice][1], Z[choice][2]])
self.radius = a1 - a0
def explain_ls(self, nsample=200):
self.Z = []
self.Z0 = []
for i in range(0, nsample):
rand = self.rand_sphere(0,self.radius/2)
z = [self.mittelpunkt[0] + rand[0], self.mittelpunkt[1] + rand[1]]
self.Z.append([])
self.Z[i].append(z[0])
self.Z[i].append(z[1])
z0 = self.retransform(z)
dummy = copy.copy(self.instance)
dummy[self.attr[0]] = z0[0]
dummy[self.attr[1]] = z0[1]
self.Z0.append([])
self.Z0[i].append(z0[0])
self.Z0[i].append(z0[1])
coord3 = np.array(self.clf.predict_proba(np.array(dummy).reshape(1, -1))[0])[1]
self.Z[i].append(int(coord3 >= 0.5))
clf = RidgeClassifier(alpha=1.0)
X=np.array(self.Z)[:,0:2]
y=np.array(self.Z)[:,2]
clf.fit(X, y)
self.explainer = clf
# self.ls_m = -clf.coef_[0] / clf.coef_[1]
# self.ls_c = (0.5 - clf.intercept_) / clf.coef_[1]
def drawLS(self, attr1, attr2):
#--- Bestimme nun Punkte auf der LIME-Geraden im skalierten Raum
x_line = np.linspace(self.mittelpunkt[0] - self.radius, self.mittelpunkt[0] + self.radius, 100)
y_line = self.ls_m * x_line + self.ls_c
#--- Punkte reskalieren
x_line0 = []
y_line0 = []
for i in range(0, len(x_line)):
z = self.retransform([x_line[i], y_line[i]])
x_line0.append(z[0])
y_line0.append(z[1])
#--- Zeichne LIME-Gerade
#------
#--- Visualize Attributspace
sample = np.array(pd.read_csv("attribut_space.csv"))
yy = np.array(sample)[:,3]
for i in range(0,len(yy)):
yy[i] = int(yy[i] >= 0.5)
yy = np.array(yy)
plt.plot(x_line0, y_line0, 'm-', lw=1)
plt.scatter(np.array(self.Z0)[:, 0], np.array(self.Z0)[:, 1], c=['limegreen'*int(np.array(self.Z)[i,2])+'tomato'*(1-int(np.array(self.Z)[i,2])) for i in range(0,len(np.array(self.Z)[:,2]))], cmap = plt.cm.Paired, marker='.', s=25)
plt.scatter([self.instance[attr1]], [self.instance[attr2]], s=100, c='blue', marker='X')
plt.scatter(self.retransform(self.mittelpunkt)[0], self.retransform(self.mittelpunkt)[1], s=100, c='blue', marker='+')
plt.xlabel("Attribut " + str(attr1))
plt.ylabel("Attribut " + str(attr2))
plt.title("Attributraum und dessen Klassifizierung")
plt.show()