init commit
This commit is contained in:
75
DLSTSVM.py
Executable file
75
DLSTSVM.py
Executable file
@@ -0,0 +1,75 @@
|
||||
"""
|
||||
Article : Deep Least Squares Support Vector Machine
|
||||
Link : New
|
||||
Author : Saeed Khosravi
|
||||
"""
|
||||
import numpy as np
|
||||
import LSTSVM
|
||||
class DLSTSVM:
|
||||
def __init__(self, X, y, C, eps = 1e-4):
|
||||
self.X = X
|
||||
self.y = y
|
||||
self.C = C
|
||||
self.eps = eps
|
||||
|
||||
def fit(self):
|
||||
#LSTSVM 1
|
||||
C1 = self.C[0]
|
||||
C2 = self.C[1]
|
||||
y = self.y
|
||||
lstsvm1 = LSTSVM.LSTSVM(self.X, y, C1, C2)
|
||||
lstsvm1.fit()
|
||||
self.w11, self.b11, self.w12, self.b12 = lstsvm1.get_params()
|
||||
self.f1 = self.f_(self.X, self.w11, self.b11, self.w12, self.b12)
|
||||
|
||||
#LSTSVM 2
|
||||
C1 = self.C[2]
|
||||
C2 = self.C[3]
|
||||
y = self.y
|
||||
lstsvm2 = LSTSVM.LSTSVM(self.X, y, C1, C2)
|
||||
lstsvm2.fit()
|
||||
self.w21, self.b21, self.w22, self.b22 = lstsvm2.get_params()
|
||||
self.f2 = self.f_(self.X, self.w21, self.b21, self.w22, self.b22)
|
||||
|
||||
#LSTSVM Main
|
||||
C1 = self.C[4]
|
||||
C2 = self.C[5]
|
||||
X = self.f(self.f1, self.f2)
|
||||
y = self.y
|
||||
lstsvm_M = LSTSVM.LSTSVM(X, y, C1, C2)
|
||||
lstsvm_M.fit()
|
||||
self.w1, self.b1, self.w2, self.b2 = lstsvm_M.get_params()
|
||||
|
||||
def predict(self, x_test, y_test):
|
||||
f1 = self.f_(x_test, self.w11, self.b11, self.w12, self.b12)
|
||||
f2 = self.f_(x_test, self.w21, self.b21, self.w22, self.b22)
|
||||
f = self.f(f1, f2)
|
||||
distance_1 = np.abs(np.dot(f, self.w1) + self.b1)
|
||||
distance_2 = np.abs(np.dot(f, self.w2) + self.b2)
|
||||
y_pred = np.zeros_like(distance_1)
|
||||
for i in range(y_pred.shape[0]):
|
||||
if (distance_1[i] < distance_2[i]):
|
||||
y_pred[i][0] = 1;
|
||||
else:
|
||||
y_pred[i][0] = -1;
|
||||
self.preds = y_pred
|
||||
|
||||
def f_(self, x, w1, b1, w2, b2):
|
||||
f = np.concatenate((np.dot(x, w1)+b1, np.dot(x, w2)+b2), axis = 1)
|
||||
return f
|
||||
|
||||
def f(self, f1, f2):
|
||||
f = np.concatenate((f1,f2), axis = 1)
|
||||
return f
|
||||
|
||||
|
||||
def get_hidden_params(self):
|
||||
return self.w11, self.b11, self.w12, self.b12, self.w21, self.b21, self.w22, self.b22
|
||||
|
||||
def get_output_params(self):
|
||||
return self.w1, self.b1, self.w2, self.b2
|
||||
|
||||
def get_preds(self):
|
||||
return self.preds
|
||||
|
||||
|
||||
79
DTSVM.py
Executable file
79
DTSVM.py
Executable file
@@ -0,0 +1,79 @@
|
||||
"""
|
||||
Article : Deep Twin Support Vector Machine
|
||||
Link : https://sci-hub.tw/https://ieeexplore.ieee.org/abstract/document/7022580
|
||||
Author : Saeed Khosravi
|
||||
|
||||
"""
|
||||
|
||||
import numpy as np
|
||||
import TSVM
|
||||
|
||||
class DTSVM:
|
||||
|
||||
def __init__(self, X, y, C, eps = 1e-4):
|
||||
|
||||
self.X = X
|
||||
self.y = y
|
||||
self.C = C
|
||||
self.eps = eps
|
||||
|
||||
def fit(self):
|
||||
#LSTSVM 1
|
||||
C1 = self.C[0]
|
||||
C2 = self.C[1]
|
||||
y = self.y
|
||||
tsvm1 = TSVM.TSVM(self.X, y, C1, C2)
|
||||
tsvm1.fit()
|
||||
self.w11, self.b11, self.w12, self.b12 = tsvm1.get_params()
|
||||
self.f1 = self.f_(self.X, self.w11, self.b11, self.w12, self.b12)
|
||||
|
||||
#LSTSVM 2
|
||||
C1 = self.C[2]
|
||||
C2 = self.C[3]
|
||||
y = self.y
|
||||
tsvm2 = TSVM.TSVM(self.X, y, C1, C2)
|
||||
tsvm2.fit()
|
||||
self.w21, self.b21, self.w22, self.b22 = tsvm2.get_params()
|
||||
self.f2 = self.f_(self.X, self.w21, self.b21, self.w22, self.b22)
|
||||
|
||||
#LSTSVM Main
|
||||
C1 = self.C[4]
|
||||
C2 = self.C[5]
|
||||
X = self.f(self.f1, self.f2)
|
||||
y = self.y
|
||||
tsvm_M = TSVM.TSVM(X, y, C1, C2)
|
||||
tsvm_M.fit()
|
||||
self.w1, self.b1, self.w2, self.b2 = tsvm_M.get_params()
|
||||
|
||||
|
||||
def predict(self, x_test, y_test):
|
||||
f1 = self.f_(x_test, self.w11, self.b11, self.w12, self.b12)
|
||||
f2 = self.f_(x_test, self.w21, self.b21, self.w22, self.b22)
|
||||
f = self.f(f1, f2)
|
||||
distance_1 = np.abs(np.dot(f, self.w1) + self.b1)
|
||||
distance_2 = np.abs(np.dot(f, self.w2) + self.b2)
|
||||
y_pred = np.zeros_like(distance_1)
|
||||
for i in range(y_pred.shape[0]):
|
||||
if (distance_1[i] < distance_2[i]):
|
||||
y_pred[i][0] = 1;
|
||||
else:
|
||||
y_pred[i][0] = -1;
|
||||
self.preds = y_pred
|
||||
|
||||
def f_(self, x, w1, b1, w2, b2):
|
||||
f = np.concatenate((np.dot(x, w1)+b1, np.dot(x, w2)+b2), axis = 1)
|
||||
return f
|
||||
|
||||
def f(self, f1, f2):
|
||||
f = np.concatenate((f1,f2), axis = 1)
|
||||
return f
|
||||
|
||||
|
||||
def get_hidden_params(self):
|
||||
return self.w11, self.b11, self.w12, self.b12, self.w21, self.b21, self.w22, self.b22
|
||||
|
||||
def get_output_params(self):
|
||||
return self.w1, self.b1, self.w2, self.b2
|
||||
|
||||
def get_preds(self):
|
||||
return self.preds
|
||||
100
LSTSVM.py
Executable file
100
LSTSVM.py
Executable file
@@ -0,0 +1,100 @@
|
||||
"""
|
||||
Article : Least squares twin support vector machines for pattern classification
|
||||
Link : https://sci-hub.tw/https://www.sciencedirect.com/science/article/abs/pii/S0957417408006854
|
||||
Author : Saeed Khosravi
|
||||
"""
|
||||
import numpy as np
|
||||
class LSTSVM:
|
||||
"""
|
||||
Least Squares Support Vector Machines
|
||||
A = Instances with label +1
|
||||
B = Instances with label -1
|
||||
C1 = hyperparameter for hyperplane 1
|
||||
C2 = hyperparameter for hyperplane 2
|
||||
|
||||
"""
|
||||
def __init__(self, X, y, C1, C2, eps = 1e-4):
|
||||
self.A = X[np.ix_(y[:,0] == 1),:][0,:,:]
|
||||
self.B = X[np.ix_(y[:,0] == -1),:][0,:,:]
|
||||
self.C1 = C1
|
||||
self.C2 = C2
|
||||
self.eps = eps
|
||||
|
||||
def fit(self):
|
||||
A = self.A
|
||||
B = self.B
|
||||
C1 = self.C1
|
||||
C2 = self.C2
|
||||
eps = self.eps
|
||||
m1, n = A.shape
|
||||
m2, n = B.shape
|
||||
e1 = np.ones((m1, 1))
|
||||
e2 = np.ones((m2, 1))
|
||||
X = np.concatenate((A, B), axis=0)
|
||||
G = np.concatenate((A, e1), axis=1)
|
||||
H = np.concatenate((B, e2), axis=1)
|
||||
|
||||
|
||||
if(m1 < m2):
|
||||
Y = self.calc_Y_or_Z(H)
|
||||
|
||||
#w1, b1
|
||||
GYGT = np.dot(np.dot(G, Y), G.T)
|
||||
I = np.eye(GYGT.shape[0], GYGT.shape[1])
|
||||
w1_b1 = - np.dot(Y - np.dot(np.dot(np.dot(Y, G.T), np.linalg.inv(C1*I + GYGT)), np.dot(G, Y)),
|
||||
np.dot(H.T, np.ones((H.T.shape[1], 1))))
|
||||
w1 = w1_b1[:-1, :]
|
||||
b1 = w1_b1[ -1, :]
|
||||
|
||||
#w2, b2
|
||||
w2_b2 = C2 * np.dot(Y - np.dot(np.dot(np.dot(Y, G.T), np.linalg.inv((I/C2)+GYGT)), np.dot(G, Y)),
|
||||
np.dot(G.T, np.ones((G.T.shape[1], 1))))
|
||||
w2 = w2_b2[:-1, :]
|
||||
b2 = w2_b2[ -1, :]
|
||||
|
||||
else:
|
||||
Z = self.calc_Y_or_Z(G)
|
||||
|
||||
#w1, b1
|
||||
HZHT = np.dot(np.dot(H, Z), H.T)
|
||||
I = np.eye(HZHT.shape[0], HZHT.shape[1])
|
||||
w1_b1 = -C1*np.dot(Z - np.dot(np.dot(np.dot(Z, H.T), np.linalg.inv((I/C1) + HZHT)), np.dot(H, Z)),
|
||||
np.dot(H.T, np.ones((H.T.shape[1], 1))))
|
||||
w1 = w1_b1[:-1, :]
|
||||
b1 = w1_b1[ -1, :]
|
||||
|
||||
#w2, b2
|
||||
w2_b2 = np.dot(Z - np.dot(np.dot(np.dot(Z, H.T), np.linalg.inv(C2*I + HZHT)), np.dot(H, Z)),
|
||||
np.dot(G.T, np.ones((G.T.shape[1], 1))))
|
||||
w2 = w2_b2[:-1, :]
|
||||
b2 = w2_b2[ -1, :]
|
||||
|
||||
self.w1 = w1
|
||||
self.w2 = w2
|
||||
self.b1 = b1
|
||||
self.b2 = b2
|
||||
|
||||
def predict(self, x_test, y_test):
|
||||
distance1 = np.abs(np.dot(x_test, self.w1) + self.b1)
|
||||
distance2 = np.abs(np.dot(x_test, self.w2) + self.b2)
|
||||
y_pred = np.zeros_like(y_test)
|
||||
for d in range(y_pred.shape[0]):
|
||||
if (distance1[d] < distance2[d]):
|
||||
y_pred[d][0] = 1;
|
||||
else:
|
||||
y_pred[d][0] = -1;
|
||||
self.preds = y_pred
|
||||
|
||||
def calc_Y_or_Z(self, M):
|
||||
MMT = np.dot(M, M.T)
|
||||
I = np.eye(MMT.shape[0], MMT.shape[1])
|
||||
tmp = np.dot(np.dot(M.T, np.linalg.inv(self.eps*I + MMT)), M)
|
||||
I = np.eye(tmp.shape[0], tmp.shape[1])
|
||||
return (1/self.eps)*(I-tmp)
|
||||
|
||||
def get_params(self):
|
||||
return self.w1, self.b1, self.w2, self.b2
|
||||
|
||||
|
||||
def get_preds(self):
|
||||
return self.preds
|
||||
92
RULSTSVM.py
Executable file
92
RULSTSVM.py
Executable file
@@ -0,0 +1,92 @@
|
||||
"""
|
||||
Article: Reduced Universum Least Squares Support Vector Machine
|
||||
Link : New
|
||||
Author : Saeed Khosravi
|
||||
"""
|
||||
|
||||
import numpy as np
|
||||
import math
|
||||
|
||||
class RULSTSVM:
|
||||
def __init__(self, X, y, C, eps):
|
||||
self.X = X
|
||||
self.y = y
|
||||
self.C = C
|
||||
self.eps = eps
|
||||
|
||||
def fit(self):
|
||||
self.plane1(self.X, self.y, self.C[0], self.C[1], self.C[2], self.eps)
|
||||
self.plane2(self.X, self.y, self.C[3], self.C[4], self.C[5], self.eps)
|
||||
|
||||
def predict(self, x_test):
|
||||
distance_1 = np.abs(np.dot(x_test, self.w1) + self.b1)
|
||||
distance_2 = np.abs(np.dot(x_test, self.w2) + self.b2)
|
||||
y_pred = np.zeros_like(distance_1).reshape((-1, 1))
|
||||
for i in range(y_pred.shape[0]):
|
||||
if (distance_1[i] < distance_2[i]):
|
||||
y_pred[i][0] = 1;
|
||||
else:
|
||||
y_pred[i][0] = -1;
|
||||
self.preds = y_pred
|
||||
|
||||
def plane1(self, X, y, C1, C2, C3, eps):
|
||||
S, T_, O_, e1, eg = self.definitions1(X, y)
|
||||
STS = np.dot(S.T, S)
|
||||
T_TT_ = np.dot(T_.T, T_)
|
||||
O_TO_ = np.dot(O_.T, O_)
|
||||
I = np.eye(STS.shape[0], STS.shape[1])
|
||||
v1 = -np.dot(np.linalg.inv(STS + C1*T_TT_ + C2*I + C3*O_TO_), np.dot(C1*T_.T, e1) + (1-eps)*C3*np.dot(O_.T, eg))
|
||||
self.w1 = v1[:-1, :]
|
||||
self.b1 = v1[ -1, :]
|
||||
|
||||
def plane2(self, X, y, C4, C5, C6, eps):
|
||||
S, T, O, e1, ed = self.definitions2(X, y)
|
||||
TTT = np.dot(T.T, T)
|
||||
STS = np.dot(S.T, S)
|
||||
OTO = np.dot(O.T, O)
|
||||
I = np.dot(TTT.shape[0], TTT.shape[0])
|
||||
v2 = np.dot(np.linalg.inv(TTT + C4*STS + C5*I + C6*OTO), C4*np.dot(S.T, e1) - C6*np.dot(O.T, (1-eps)*ed))
|
||||
self.w2 = v2[:-1, :]
|
||||
self.b2 = v2[ -1, :]
|
||||
|
||||
def definitions1(self, X, y):
|
||||
X1 = X[np.ix_(y[:,0] == 1),:][0,:,:]
|
||||
X2 = X[np.ix_(y[:,0] == -1),:][0,:,:]
|
||||
r, n = X1.shape
|
||||
s, n = X2.shape
|
||||
np.random.shuffle(X2)
|
||||
X2_ = X2[:r, :]
|
||||
U = X2[r: , :]
|
||||
d, n = U.shape
|
||||
g = math.ceil(r/2)
|
||||
U_ = U[np.random.choice(np.arange(1, d), g), :]
|
||||
e1 = np.ones((X1.shape[0], 1))
|
||||
eg = np.ones((U_.shape[0], 1))
|
||||
S = np.concatenate((X1 , e1), axis = 1)
|
||||
T_ = np.concatenate((X2_, e1), axis = 1)
|
||||
O_ = np.concatenate((U_ , eg), axis = 1)
|
||||
return S, T_, O_, e1, eg
|
||||
|
||||
def definitions2(self, X, y):
|
||||
X1 = X[np.ix_(y[:,0] == 1),:][0,:,:]
|
||||
X2 = X[np.ix_(y[:,0] == -1),:][0,:,:]
|
||||
r, n = X1.shape
|
||||
s, n = X2.shape
|
||||
np.random.shuffle(X2)
|
||||
X2_ = X2[:r, :]
|
||||
U = X2[r: , :]
|
||||
d, n = U.shape
|
||||
g = math.ceil(r/2)
|
||||
e1 = np.ones((X1.shape[0], 1))
|
||||
e2 = np.ones((X2.shape[0], 1))
|
||||
ed = np.ones((U.shape[0] , 1))
|
||||
S = np.concatenate((X1 , e1), axis = 1)
|
||||
T = np.concatenate((X2 , e2), axis = 1)
|
||||
O = np.concatenate((U , ed), axis = 1)
|
||||
return S, T, O, e1, ed
|
||||
|
||||
def get_params(self):
|
||||
return self.w1, self.b1, self.w2, self.b2
|
||||
|
||||
def get_preds(self):
|
||||
return self.preds
|
||||
156
RUTSVM.py
Executable file
156
RUTSVM.py
Executable file
@@ -0,0 +1,156 @@
|
||||
"""
|
||||
Article : A reduced universum twin support vector machine for class imbalance learning
|
||||
Link : https://sci-hub.tw/https://www.sciencedirect.com/science/article/abs/pii/S0031320319304510
|
||||
Author : Saeed Khosravi
|
||||
"""
|
||||
|
||||
import numpy as np
|
||||
from cvxopt import solvers, matrix
|
||||
import math
|
||||
|
||||
class RUTSVM:
|
||||
|
||||
def __init__(self, X, y, C1, C2, CU, eps):
|
||||
|
||||
self.X = X
|
||||
self.y = y
|
||||
self.C1 = C1
|
||||
self.C2 = C2
|
||||
self.CU = CU
|
||||
self.eps = eps
|
||||
|
||||
def fit(self):
|
||||
self.w1, self.b1 = self.plane1(self.X, self.y, self.C1, self.CU, self.eps)
|
||||
self.w2, self.b2 = self.plane2(self.X, self.y, self.C2, self.CU, self.eps)
|
||||
|
||||
def predict(self, x_test):
|
||||
norm2_w1 = np.linalg.norm(self.w1)
|
||||
norm2_w2 = np.linalg.norm(self.w2)
|
||||
distance_1 = np.abs(np.dot(x_test, self.w1) + self.b1)/norm2_w1
|
||||
distance_2 = np.abs(np.dot(x_test, self.w2) + self.b2)/norm2_w2
|
||||
y_pred = np.zeros_like(distance_1).reshape((-1, 1))
|
||||
for i in range(y_pred.shape[0]):
|
||||
if (distance_1[i] < distance_2[i]):
|
||||
y_pred[i][0] = 1;
|
||||
else:
|
||||
y_pred[i][0] = -1;
|
||||
|
||||
self.preds = y_pred
|
||||
|
||||
def plane1(self, X, y, c1, cu, eps):
|
||||
S, T, O, T_, O_, e1, e2, eg, ed = self.split_dataset(X, y)
|
||||
m1 = S.shape[0]
|
||||
m2 = T_.shape[0]
|
||||
mg = O_.shape[0]
|
||||
|
||||
STS = np.dot(S.T, S)
|
||||
I = np.eye(STS.shape[0], STS.shape[1])
|
||||
STS_inv = np.linalg.inv(1e-4*I + STS)
|
||||
|
||||
_P = np.dot(np.dot(T_, STS_inv), T_.T)
|
||||
_P = np.concatenate((_P, -np.dot(np.dot(T_, STS_inv), O_.T)), axis = 1)
|
||||
_P2 = -np.dot(-np.dot(O_, STS_inv), T_.T)
|
||||
_P2 = np.concatenate((_P2, np.dot(np.dot(O_, STS_inv), O_.T)), axis = 1)
|
||||
_P = np.concatenate((_P, _P2), axis = 0) # (m1 + mg , m1 + mg)
|
||||
|
||||
_q = np.concatenate((-e1.T, (1-eps)*eg.T), axis = 1).T # (m1 + mg , 1)
|
||||
|
||||
_G1 = np.concatenate(( np.eye(m1, m1), np.zeros((m1, mg))), axis = 1)
|
||||
_G2 = np.concatenate((-np.eye(m1, m1), np.zeros((m1, mg))), axis = 1)
|
||||
_G3 = np.concatenate(( np.zeros((mg, m1)), np.eye(mg, mg)), axis = 1)
|
||||
_G4 = np.concatenate((np.zeros((mg, m1)), -np.eye(mg, mg)), axis = 1)
|
||||
_G = np.concatenate((_G1, _G2), axis = 0)
|
||||
_G = np.concatenate((_G , _G3), axis = 0)
|
||||
_G = np.concatenate((_G, _G4), axis = 0) # (2m1 + 2mg , m1 + mg)
|
||||
|
||||
_h = np.zeros((2*m1 + 2*mg, 1))
|
||||
_h[:m1, :] = c1
|
||||
_h[2*m1:2*m1+mg, :] = cu
|
||||
|
||||
_P = matrix(_P, tc= 'd')
|
||||
_q = matrix(_q, tc = 'd')
|
||||
_G = matrix(_G, tc = 'd')
|
||||
_h = matrix(_h, tc = 'd')
|
||||
|
||||
qp_sol = solvers.qp(_P, _q, _G, _h, kktsolver='ldl', options={'kktreg':1e-9, 'show_progress':False})
|
||||
qp_sol = np.array(qp_sol['x'])
|
||||
alphas = qp_sol[:m1, 0]
|
||||
mus = qp_sol[m1:, 0]
|
||||
vp = -np.dot(STS_inv, np.dot(T_.T, alphas) - np.dot(O_.T, mus))
|
||||
w = vp[:-1]
|
||||
b = vp[-1]
|
||||
return w, b
|
||||
|
||||
|
||||
def plane2(self, X, y, c2, cu, eps):
|
||||
S, T, O, T_, O_, e1, e2, eg, ed = self.split_dataset(X, y)
|
||||
|
||||
m1 = S.shape[0]
|
||||
m2 = T.shape[0]
|
||||
md = O.shape[0]
|
||||
|
||||
TTT = np.dot(T.T, T)
|
||||
I = np.eye(TTT.shape[0], TTT.shape[1])
|
||||
TTT_inv = np.linalg.inv(1e-4*I + TTT)
|
||||
_P = np.dot(np.dot(S, TTT_inv), S.T)
|
||||
_P = np.concatenate((_P, -np.dot(np.dot(S, TTT_inv), O.T)), axis = 1)
|
||||
_P2 = -np.dot(-np.dot(O, TTT_inv), S.T)
|
||||
_P2 = np.concatenate((_P2, np.dot(np.dot(O, TTT_inv), O.T)), axis = 1)
|
||||
_P = np.concatenate((_P, _P2), axis = 0) # (m1 + md , m1 + md)
|
||||
|
||||
_q = np.concatenate((-e1.T, (eps - 1)*ed.T), axis = 1).T # (m1 + md , 1)
|
||||
|
||||
_G1 = np.concatenate(( np.eye(m1, m1), np.zeros((m1, md))), axis = 1)
|
||||
_G2 = np.concatenate((-np.eye(m1, m1), np.zeros((m1, md))), axis = 1)
|
||||
_G3 = np.concatenate(( np.zeros((md, m1)), np.eye(md, md)), axis = 1)
|
||||
_G4 = np.concatenate((np.zeros((md, m1)), -np.eye(md, md)), axis = 1)
|
||||
_G = np.concatenate((_G1, _G2), axis = 0)
|
||||
_G = np.concatenate((_G , _G3), axis = 0)
|
||||
_G = np.concatenate((_G, _G4), axis = 0) # (2m1 + 2md , m1 + md)
|
||||
|
||||
_h = np.zeros((2*m1 + 2*md, 1))
|
||||
_h[:m1, :] = c2
|
||||
_h[2*m1:2*m1+md, :] = cu
|
||||
|
||||
|
||||
_P = matrix(_P, tc= 'd')
|
||||
_q = matrix(_q, tc = 'd')
|
||||
_G = matrix(_G, tc = 'd')
|
||||
_h = matrix(_h, tc = 'd')
|
||||
|
||||
qp_sol = solvers.qp(_P, _q, _G, _h, kktsolver='ldl', options={'kktreg':1e-9, 'show_progress':False})
|
||||
qp_sol = np.array(qp_sol['x'])
|
||||
alphas = qp_sol[:m1, 0]
|
||||
mus = qp_sol[m1:, 0]
|
||||
vn = -np.dot(TTT_inv, np.dot(S.T, alphas) - np.dot(O.T, mus))
|
||||
w = vn[:-1]
|
||||
b = vn[-1]
|
||||
return w, b
|
||||
|
||||
def split_dataset(self, X, y):
|
||||
X1 = X[np.ix_(y[:,0] == 1),:][0,:,:]
|
||||
X2 = X[np.ix_(y[:,0] == -1),:][0,:,:]
|
||||
r, n = X1.shape
|
||||
s, n = X2.shape
|
||||
X2_ = X2[:r, :]
|
||||
U = X2[r: , :]
|
||||
d, n = U.shape
|
||||
g = math.ceil(r/2)
|
||||
tmp = np.random.choice(np.arange(1, d), g)
|
||||
U_ = U[tmp, :]
|
||||
e1 = np.ones((X1.shape[0], 1))
|
||||
e2 = np.ones((X2.shape[0], 1))
|
||||
eg = np.ones((U_.shape[0], 1))
|
||||
ed = np.ones((U.shape[0] , 1))
|
||||
S = np.concatenate((X1 , e1), axis = 1)
|
||||
T_ = np.concatenate((X2_, e1), axis = 1)
|
||||
T = np.concatenate((X2 , e2), axis = 1)
|
||||
O_ = np.concatenate((U_ , eg), axis = 1)
|
||||
O = np.concatenate((U , ed), axis = 1)
|
||||
return S, T, O, T_, O_, e1, e2, eg, ed
|
||||
|
||||
def get_params(self):
|
||||
return self.w1, self.b1, self.w2, self.b2
|
||||
|
||||
def get_preds(self):
|
||||
return self.preds
|
||||
82
TSVM.py
Executable file
82
TSVM.py
Executable file
@@ -0,0 +1,82 @@
|
||||
"""
|
||||
Article : Twin Support Vector Machine
|
||||
Link : https://sci-hub.tw/https://ieeexplore.ieee.org/document/4135685
|
||||
Author : Saeed Khosravi
|
||||
"""
|
||||
|
||||
import numpy as np
|
||||
from cvxopt import solvers, matrix
|
||||
|
||||
class TSVM:
|
||||
|
||||
def __init__(self, X, y, C1, C2, eps=1e-4):
|
||||
|
||||
self.A = X[np.ix_(y[:,0] == 1),:][0,:,:]
|
||||
self.B = X[np.ix_(y[:,0] == -1),:][0,:,:]
|
||||
self.C1 = C1
|
||||
self.C2 = C2
|
||||
self.eps = eps
|
||||
|
||||
def fit(self):
|
||||
self.w1, self.b1 = self.plane1(self.A, self.B, self.C1, self.eps)
|
||||
self.w2, self.b2 = self.plane2(self.A, self.B, self.C2, self.eps)
|
||||
|
||||
def predict(self, x_test):
|
||||
norm2_w1 = np.linalg.norm(self.w1)
|
||||
norm2_w2 = np.linalg.norm(self.w2)
|
||||
distance_1 = np.abs(np.dot(x_test, self.w1) + self.b1)/norm2_w1
|
||||
distance_2 = np.abs(np.dot(x_test, self.w2) + self.b2)/norm2_w2
|
||||
y_pred = np.zeros_like(distance_1)
|
||||
for i in range(y_pred.shape[0]):
|
||||
if (distance_1[i] < distance_2[i]):
|
||||
y_pred[i][0] = 1;
|
||||
else:
|
||||
y_pred[i][0] = -1;
|
||||
|
||||
self.preds = y_pred
|
||||
|
||||
def plane1(self, A, B, c, eps):
|
||||
e1 = np.ones((A.shape[0],1))
|
||||
e2 = np.ones((B.shape[0],1))
|
||||
H = np.concatenate((A,e1), axis=1)
|
||||
G = np.concatenate((B,e2), axis=1)
|
||||
HTH = np.dot(H.T, H)
|
||||
if np.linalg.matrix_rank(H)<H.shape[1]:
|
||||
HTH += eps*np.eye(HTH.shape[0], HTH.shape[1])
|
||||
|
||||
_P = matrix(np.dot(np.dot(G, np.linalg.inv(HTH)),G.T), tc = 'd')
|
||||
_q = matrix(-1 * e2, tc = 'd')
|
||||
_G = matrix(np.concatenate((np.identity(B.shape[0]),-np.identity(B.shape[0])), axis=0), tc = 'd')
|
||||
_h = matrix(np.concatenate((c*e2,np.zeros_like(e2)), axis=0), tc = 'd')
|
||||
qp_sol = solvers.qp(_P, _q, _G, _h, kktsolver='ldl', options={'kktreg':1e-9, 'show_progress':False})
|
||||
qp_sol = np.array(qp_sol['x'])
|
||||
z = -np.dot(np.dot(np.linalg.inv(HTH), G.T), qp_sol)
|
||||
w = z[:z.shape[0]-1]
|
||||
b = z[z.shape[0]-1]
|
||||
return w, b[0]
|
||||
|
||||
def plane2(self, A, B, c, eps):
|
||||
e1 = np.ones((A.shape[0],1))
|
||||
e2 = np.ones((B.shape[0],1))
|
||||
H = np.concatenate((A,e1), axis=1)
|
||||
G = np.concatenate((B,e2), axis=1)
|
||||
GTG = np.dot(G.T, G)
|
||||
if np.linalg.matrix_rank(G)<G.shape[1]:
|
||||
GTG += eps*np.eye(GTG.shape[0], GTG.shape[1])
|
||||
#solving the qp by cvxopt
|
||||
_P = matrix(np.dot(np.dot(H, np.linalg.inv(GTG)), H.T), tc = 'd')
|
||||
_q = matrix(-1 * e1, tc = 'd')
|
||||
_G = matrix(np.concatenate((np.identity(A.shape[0]),-np.identity(A.shape[0])), axis=0), tc = 'd')
|
||||
_h = matrix(np.concatenate((c*e1,np.zeros_like(e1)), axis=0), tc = 'd')
|
||||
qp_sol = solvers.qp(_P, _q, _G, _h, kktsolver='ldl', options={'kktreg':1e-9, 'show_progress':False})
|
||||
qp_sol = np.array(qp_sol['x'])
|
||||
z = -np.dot(np.dot(np.linalg.inv(GTG), H.T), qp_sol)
|
||||
w = z[:z.shape[0]-1]
|
||||
b = z[z.shape[0]-1]
|
||||
return w, b[0]
|
||||
|
||||
def get_params(self):
|
||||
return self.w1, self.b1, self.w2, self.b2
|
||||
|
||||
def get_preds(self):
|
||||
return self.preds
|
||||
146
UTSVM.py
Executable file
146
UTSVM.py
Executable file
@@ -0,0 +1,146 @@
|
||||
"""
|
||||
Article : Twin Support Vector Machine with Universum data
|
||||
Link : https://sci-hub.tw/https://www.sciencedirect.com/science/article/abs/pii/S0893608012002304
|
||||
Author : Saeed Khosravi
|
||||
"""
|
||||
|
||||
import numpy as np
|
||||
from cvxopt import solvers, matrix
|
||||
|
||||
class UTSVM:
|
||||
|
||||
def __init__(self, X, y, U, C1, C2, CU, eps):
|
||||
|
||||
self.X = X
|
||||
self.y = y
|
||||
self.U = U
|
||||
self.C1 = C1
|
||||
self.C2 = C2
|
||||
self.CU = CU
|
||||
self.eps = eps
|
||||
|
||||
def fit(self):
|
||||
self.w1, self.b1 = self.plane1(self.X, self.y, self.U, self.C1, self.CU, self.eps)
|
||||
self.w2, self.b2 = self.plane2(self.X, self.y, self.U, self.C2, self.CU, self.eps)
|
||||
|
||||
def predict(self, x_test):
|
||||
norm2_w1 = np.linalg.norm(self.w1)
|
||||
norm2_w2 = np.linalg.norm(self.w2)
|
||||
distance_1 = np.abs(np.dot(x_test, self.w1) + self.b1)/norm2_w1
|
||||
distance_2 = np.abs(np.dot(x_test, self.w2) + self.b2)/norm2_w2
|
||||
y_pred = np.zeros_like(distance_1).reshape((-1, 1))
|
||||
for i in range(y_pred.shape[0]):
|
||||
if (distance_1[i] < distance_2[i]):
|
||||
y_pred[i][0] = 1;
|
||||
else:
|
||||
y_pred[i][0] = -1;
|
||||
|
||||
self.preds = y_pred
|
||||
|
||||
def plane1(self, X, y, U, c, cu, eps):
|
||||
A = X[np.ix_(y[:,0] == 1),:][0,:,:]
|
||||
B = X[np.ix_(y[:,0] == -1),:][0,:,:]
|
||||
m1 = A.shape[0]
|
||||
m2 = B.shape[0]
|
||||
ep = np.ones((m1, 1))
|
||||
en = np.ones((m2, 1))
|
||||
mu = U.shape[0]
|
||||
eu = np.ones((mu, 1))
|
||||
H = np.concatenate((A, ep), axis = 1)
|
||||
G = np.concatenate((B, en), axis = 1)
|
||||
O = np.concatenate((U, eu), axis = 1)
|
||||
HTH = np.dot(H.T, H)
|
||||
I = np.eye(HTH.shape[0], HTH.shape[1])
|
||||
HTH_inv = np.linalg.inv(1e-4*I + HTH)
|
||||
|
||||
_P = np.dot(np.dot(G, HTH_inv), G.T)
|
||||
_P = np.concatenate((_P, -np.dot(np.dot(G, HTH_inv), O.T)), axis = 1)
|
||||
_P2 = -np.dot(np.dot(O, HTH_inv), G.T)
|
||||
_P2 = np.concatenate((_P2, np.dot(np.dot(O, HTH_inv), O.T)), axis = 1)
|
||||
_P = np.concatenate((_P, _P2), axis = 0) # (en + eu , en + eu)
|
||||
|
||||
|
||||
_q = np.concatenate((-en.T, (1-eps)*eu.T), axis = 1).T # (en + eu , 1)
|
||||
|
||||
|
||||
_G1 = np.concatenate(( np.eye(m2, m2), np.zeros((m2, mu))), axis = 1)
|
||||
_G2 = np.concatenate((-np.eye(m2, m2), np.zeros((m2, mu))), axis = 1)
|
||||
_G3 = np.concatenate(( np.zeros((mu, m2)), np.eye(mu, mu)), axis = 1)
|
||||
_G4 = np.concatenate(( np.zeros((mu, m2)), -np.eye(mu, mu)), axis = 1)
|
||||
_G = np.concatenate((_G1, _G2), axis = 0)
|
||||
_G = np.concatenate((_G , _G3), axis = 0)
|
||||
_G = np.concatenate((_G, _G4), axis = 0) # (4 * m2, m2 + mu)
|
||||
|
||||
_h = np.zeros((2*m2 + 2*mu, 1))
|
||||
_h[:m2, 0] = c
|
||||
_h[2*m2:2*m2+mu, :] = cu
|
||||
|
||||
_P = matrix(_P, tc= 'd')
|
||||
_q = matrix(_q, tc = 'd')
|
||||
_G = matrix(_G, tc = 'd')
|
||||
_h = matrix(_h, tc = 'd')
|
||||
qp_sol = solvers.qp(_P, _q, _G, _h, kktsolver='ldl', options={'kktreg':1e-9, 'show_progress':False})
|
||||
qp_sol = np.array(qp_sol['x'])
|
||||
alphas = qp_sol[:m2, 0]
|
||||
mus = qp_sol[m2:, 0]
|
||||
vp = -np.dot(HTH_inv, np.dot(G.T, alphas) - np.dot(O.T, mus))
|
||||
w = vp[:vp.shape[0]-1]
|
||||
b = vp[vp.shape[0]-1]
|
||||
return w, b
|
||||
|
||||
def plane2(self, X, y, U, c, cu, eps):
|
||||
A = X[np.ix_(y[:,0] == -1),:][0,:,:]
|
||||
B = X[np.ix_(y[:,0] == 1),:][0,:,:]
|
||||
m1 = A.shape[0]
|
||||
m2 = B.shape[0]
|
||||
en = np.ones((m1, 1))
|
||||
ep = np.ones((m2, 1))
|
||||
mu = U.shape[0]
|
||||
eu = np.ones((mu, 1))
|
||||
Q = np.concatenate((A, en), axis = 1)
|
||||
P = np.concatenate((B, ep), axis = 1)
|
||||
S = np.concatenate((U, eu), axis = 1)
|
||||
QTQ = np.dot(Q.T, Q)
|
||||
I = np.eye(QTQ.shape[0], QTQ.shape[1])
|
||||
QTQ_inv = np.linalg.inv(1e-4*I + QTQ)
|
||||
|
||||
_P = np.dot(np.dot(P, QTQ_inv), P.T)
|
||||
_P = np.concatenate((_P, -np.dot(np.dot(P, QTQ_inv), S.T)), axis = 1)
|
||||
_P2 = -np.dot(np.dot(S, QTQ_inv), P.T)
|
||||
_P2 = np.concatenate((_P2, np.dot(np.dot(S, QTQ_inv), S.T)), axis = 1)
|
||||
_P = np.concatenate((_P, _P2), axis = 0) # (ep + eu , ep + eu)
|
||||
|
||||
|
||||
_q = np.concatenate((-ep.T, (1-eps)*eu.T), axis = 1).T # (ep + eu , 1)
|
||||
|
||||
|
||||
_G1 = np.concatenate(( np.eye(m2, m2), np.zeros((m2, mu))), axis = 1)
|
||||
_G2 = np.concatenate((-np.eye(m2, m2), np.zeros((m2, mu))), axis = 1)
|
||||
_G3 = np.concatenate(( np.zeros((mu, m2)), np.eye(mu, mu)), axis = 1)
|
||||
_G4 = np.concatenate(( np.zeros((mu, m2)), -np.eye(mu, mu)), axis = 1)
|
||||
_G = np.concatenate((_G1, _G2), axis = 0)
|
||||
_G = np.concatenate((_G , _G3), axis = 0)
|
||||
_G = np.concatenate((_G, _G4), axis = 0) # (4 * m2, m2 + mu)
|
||||
|
||||
_h = np.zeros((2*m2 + 2*mu, 1))
|
||||
_h[:m2, 0] = c
|
||||
_h[2*m2:2*m2+mu, :] = cu
|
||||
|
||||
_P = matrix(_P, tc= 'd')
|
||||
_q = matrix(_q, tc = 'd')
|
||||
_G = matrix(_G, tc = 'd')
|
||||
_h = matrix(_h, tc = 'd')
|
||||
qp_sol = solvers.qp(_P, _q, _G, _h, kktsolver='ldl', options={'kktreg':1e-9, 'show_progress':False})
|
||||
qp_sol = np.array(qp_sol['x'])
|
||||
alphas = qp_sol[:m2, 0]
|
||||
mus = qp_sol[m2:, 0]
|
||||
vp = -np.dot(QTQ_inv, np.dot(P.T, alphas) - np.dot(S.T, mus))
|
||||
w = vp[:vp.shape[0]-1]
|
||||
b = vp[vp.shape[0]-1]
|
||||
return w, b
|
||||
|
||||
def get_params(self):
|
||||
return self.w1, self.b1, self.w2, self.b2
|
||||
|
||||
def get_preds(self):
|
||||
return self.preds
|
||||
Reference in New Issue
Block a user