kmeans_mahalanobis.py
5.91 KB
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
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
from sklearn.cluster import KMeans
import pickle
import numpy as np
import matplotlib.pyplot as plt
from sklearn.manifold import TSNE
from abstract_clustering import AbstractClustering
class kmeansMahalanobis():
def __init__(self):
"""
"""
self.C = None
self.L = None
self.K = None
def predict(self, features):
"""
@param features:
@return:
"""
N = features.shape[0]
distances = np.zeros((N, self.K))
for n in range(N):
for k in range(self.K):
distances[n][k] = self._dist(features[n], self.C[k], self.L[k])
closest_cluster = np.argmin(distances, axis=1)
return closest_cluster
def load(self, model_path):
"""
@param model_path:
@return:
"""
data = None
with open(model_path, "rb") as f:
data = pickle.load(f)
if data is None:
raise Exception("Le modèle n'a pas pu être chargé")
else:
self.C = data["C"]
self.L = data["L"]
self.K = data["K"]
def save(self, modelpath: str):
"""
@param modelpath:
@return:
"""
data = {
"C": self.C,
"L": self.L,
"K": self.K
}
with open(modelpath, "wb") as f:
pickle.dump(data, f)
def fit(self, features, k: int, tol: float, ninit: int, maxiter: int=300, debug: bool=False):
results = []
for i in range(ninit):
results.append(self._train(features, k, tol, maxiter, debug))
losses = [v["loss"] for v in results]
best = results[losses.index(min(losses))]
if debug:
print(f"best: {best['loss']} loss")
self.C = best["C"]
self.L = best["L"]
self.K = best["K"]
def _initialize_model(self, X, number_clusters):
d = X.shape[1]
C = X[np.random.choice(X.shape[0], number_clusters)]
L = np.zeros((number_clusters, d, d))
for k in range(number_clusters):
L[k] = np.identity(d)
return C, L
def _dist(self, a, b, l):
'''
Distance euclidienne
'''
a = np.reshape(a, (-1, 1))
b = np.reshape(b, (-1, 1))
result = np.transpose(a - b).dot(l).dot(a-b)[0][0]
return result
def _plot_iteration(self, iteration, points, clusters, centers):
fig = plt.figure()
ax = fig.add_subplot(111)
scatter = ax.scatter(points[:, 0], points[:, 1], c=clusters, s=50)
#for center in centers:
# ax.scatter(center[0], center[1], s=50, c='red', marker='+')
ax.scatter(centers[:, 0], centers[:, 1], s=50, c='red', marker='+')
ax.set_xlabel('x')
ax.set_ylabel('y')
plt.colorbar(scatter)
#plt.ylim(0, 1)
#plt.xlim(0, 1)
plt.savefig("test_" + str(iteration) + ".pdf")
def _train(self, features, K: int, tol: float, maxiter: int, debug: bool=False):
X = features
N = X.shape[0]
d = X.shape[1]
C, L = self._initialize_model(X, K)
self.C = C
self.L = L
self.K = K
end_algo = False
i = 0
while not end_algo:
if debug:
print("Iteration: ", i)
# Calcul matrix distance
distances = np.zeros((N, self.K))
for n in range(N):
for k in range(self.K):
distances[n][k] = self._dist(X[n], self.C[k], self.L[k])
closest_cluster = np.argmin(distances, axis=1)
loss = np.sum(distances[np.arange(len(distances)), closest_cluster])
if debug:
print(f"loss {loss}")
# -- Debug tool ----------------------
if debug and i % 10 == 0:
# TSNE if needed
X_embedded = np.concatenate((X, self.C), axis=0)
if d > 2:
X_embedded = TSNE(n_components=2).fit_transform(np.concatenate((X, C), axis=0))
# Then plot
self._plot_iteration(
i,
X_embedded[:X.shape[0]],
closest_cluster,
X_embedded[X.shape[0]:]
)
# ------------------------------------
old_c = self.C.copy()
for k in range(K):
# Find subset of X with values closed to the centroid c_k.
X_sub = np.where(closest_cluster == k)
X_sub = np.take(X, X_sub[0], axis=0)
if X_sub.shape[0] == 0:
continue
np.mean(X_sub, axis=0)
C_new = np.mean(X_sub, axis=0)
# -- COMPUTE NEW LAMBDA (here named K) --
K_new = np.zeros((L.shape[1], L.shape[2]))
for x in X_sub:
x = np.reshape(x, (-1, 1))
c_tmp = np.reshape(C_new, (-1, 1))
K_new = K_new + (x - c_tmp).dot((x - c_tmp).transpose())
K_new = K_new / X_sub.shape[0]
K_new = np.linalg.pinv(K_new)
#if end_algo and (not (self.C[k] == C_new).all()): # If the same stop
# end_algo = False
self.C[k] = C_new
self.L[k] = K_new
diff = np.sum(np.absolute((self.C - old_c) / old_c * 100))
if diff > tol:
end_algo = False
if debug:
print(f"{diff}")
else:
if debug:
print(f"Tolerance threshold {tol} reached with diff {diff}")
end_algo = True
i = i + 1
if i > maxiter:
end_algo = True
if debug:
print(f"Iteration {maxiter} reached")
return {
"loss": loss,
"C": self.C,
"K": self.K,
"L": self.L
}