def datasetSplit_KS(self, X_master, X_slave, y, num=5):
'''
将导入的数据集采用KS算法进行划分,找到方差最大的样本
Param：
X_master：主仪器光谱数据
X_slave：从仪器光谱数据
y： 主从仪器光谱数据的响应值
num: 前半部分抽取的方差最大的样本数目
return：
X_master_prior：主仪器光谱数据划分后的前半部分
X_master_next：主仪器光谱数据划分后的后半部分
X_slave_prior：从仪器光谱数据划分后的前半部分
X_slave_next：从仪器光谱数据划分后的后半部分
y_prior：主从仪器光谱数据的响应值的前半部分
y_next：主从仪器光谱数据的响应值的后半部分
'''
KS_demo = KennardStone(X_master, num)
print X_slave.shape , num
CalInd_master, ValInd_master = KS_demo.KS()
# CalInd_master方差最大样本的下标 ValInd_master 剩余样本的下标
X_master_prior = X_master[CalInd_master]
X_master_next = X_master[ValInd_master]
X_slave_prior = X_slave[CalInd_master]
X_slave_next = X_slave[ValInd_master]
y_prior = y[CalInd_master]
y_next = y[ValInd_master]
return X_master_prior, X_master_next, X_slave_prior, X_slave_next, y_prior, y_next
def affine_pls2(self, j):
pls_demo = Partial_LS(self.X_m_cal,
self.y_m_cal,
folds=10,
max_comp=15)
W_m_cal, T_m_cal, P_m_cal, comp_best, coefs_cal, RMSECV = pls_demo.pls2_fit(
)
# print self.y_s_test.shape , gg
RMSEC, RMSEP, ytest_pre = pls_demo.pls_pre(self.X_m_test,
self.y_m_test, coefs_cal)
y_ = deal_4_2(self.y_m_cal, self.i, j)
coef = deal_4_2(coefs_cal, self.i, j)
y_s_test_ = deal_4_2(self.y_s_test, self.i, j)
y_s_cal_ = deal_4_2(self.y_s_cal, self.i, j)
T_m_std, T_s_std, y_m_pre, y_s_pre, T_s_test, ys_test_pre = Pre_deal(
self.X_m_cal, self.X_s_cal, W_m_cal, P_m_cal, coef, self.X_m_cal,
y_, self.X_s_test)
y0_m = y_m_pre[:, 0:1]
y1_m = y_m_pre[:, 1:2]
y0_s = y_s_pre[:, 0:1]
y1_s = y_s_pre[:, 1:2]
y0_s_test = ys_test_pre[:, 0:1]
y1_s_test = ys_test_pre[:, 1:2]
demo = Affine_trans1(y0_m, y0_s, y1_m, y1_s, comp_best)
bia, sin_x, cos_x, x, b_s, k_m, b_m = demo.AT_train()
KS_master_std = KennardStone(self.X_s_cal, 16)
CalInd_master_std, ValInd_master_std = KS_master_std.KS()
# print CalInd_master_std , ValInd_master
# KS_master_std = KennardStone(self.X_s_cal, 32)
# CalInd_master_std, ValInd_master_std = KS_master_std.KS()
# # print CalInd_master_std , ValInd_master
# y0_s_sure = y0_s[CalInd_master_std]
# y1_s_sure = y1_s[CalInd_master_std]
# y_sure_real = y_s_cal_[CalInd_master_std]
y0_s_sure = y0_s[48:64, :]
y1_s_sure = y1_s[48:64, :]
y_sure_real = y_s_cal_[48:64, :]
# print bia, sin_x, cos_x, b_s
RMSEC_y0, RMSEC_y1, y0_cal_pre = demo.AT_pre(y0_s, y1_s, bia, cos_x,
sin_x, y_s_cal_, b_s)
RMSEP_y0, RMSEP_y1, y0_pre = demo.AT_pre(y0_s_test, y1_s_test, bia,
cos_x, sin_x, y_s_test_, b_s)
RMSEP_sure_y0, RMSEP_sure_y1, y_cal_pre = demo.AT_pre(
y0_s_sure, y1_s_sure, bia, cos_x, sin_x, y_sure_real, b_s)
print y_cal_pre.shape, "shapesss"
RMSEP_no_trans_0, RMSEP_no_trans_1, y_no_pre = demo.AT_pre_no_trans(
y0_s_test, y1_s_test, y_s_test_)
# print y0_s_test[:,0]
print RMSEP_sure_y0, RMSEP_sure_y1, "SURE ", RMSEP_y0
# print RMSEP_no_trans_0 ,"no_trans"
# return comp_best,RMSEP_y0,y0_pre,y_cal_pre , y0_s_test , y1_s_test, RMSEP_y0_
# print y_sure_real.shape , y_cal_pre.shape , "shape"
return comp_best, RMSEP_y0, y0_pre, RMSEC_y0, y_cal_pre, y0_s_test, y1_s_test, RMSEP_no_trans_0, y_sure_real
def getStd(X_m_cal, X_s_cal, y_m_cal, y_s_cal, std_num):
KS_master_std = KennardStone(X_m_cal, std_num)
CalInd_master_std, ValInd_master_std = KS_master_std.KS()
# print CalInd_master_std , ValInd_master
X_m_std = X_m_cal[CalInd_master_std]
y_m_std = y_m_cal[CalInd_master_std]
X_s_std = X_s_cal[CalInd_master_std]
y_s_std = y_s_cal[CalInd_master_std]
return X_m_std, y_m_std, X_s_std, y_s_std
def Data_KS_split(X_master, X_slave, y, num):
KS_master = KennardStone(X_slave, num)
CalInd_master, ValInd_master = KS_master.KS()
X_m_cal = X_master[CalInd_master]
y_cal = y[CalInd_master]
X_m_test = X_master[ValInd_master]
y_test = y[ValInd_master]
X_s_cal = X_slave[CalInd_master]
X_s_test = X_slave[ValInd_master]
return X_m_cal, y_cal, X_s_cal, X_s_test, X_m_test, y_test
def Dataset_KS_split_std(X_master, X_slave, y, num, std_num=8):
KS_master = KennardStone(X_master, num)
CalInd_master, ValInd_master = KS_master.KS()
# print CalInd_master , ValInd_master
X_m_cal = X_master[CalInd_master]
y_m_cal = y[CalInd_master]
X_m_test = X_master[ValInd_master]
y_m_test = y[ValInd_master]
X_s_cal = X_slave[CalInd_master]
X_s_test = X_slave[ValInd_master]
y_s_cal = y[CalInd_master]
y_s_test = y[ValInd_master]
KS_master_std = KennardStone(X_m_cal, std_num)
CalInd_master_std, ValInd_master_std = KS_master_std.KS()
# print CalInd_master_std , ValInd_master
X_m_std = X_m_cal[CalInd_master_std]
y_m_std = y_m_cal[CalInd_master_std]
X_s_std = X_s_cal[CalInd_master_std]
y_s_std = y_s_cal[CalInd_master_std]
return X_m_cal, y_m_cal, X_s_cal, y_s_cal, X_s_test, y_s_test, X_m_test, y_m_test, X_m_std, y_m_std, X_s_std, y_s_std
def TCA(self, num, i, snv=0):
#################
DP_demo = datasetProcess(dataType=i, bool=snv)
X_master, X_slave, y = DP_demo.datasetImport()
X_master_cal, X_master_test, X_slave_cal, X_slave_test, y_cal, y_test = DP_demo.datasetSplit_KS(
X_master, X_slave, y, num)
X_m_train, X_m_val, y_m_train, y_m_val = train_test_split(
X_master_cal, y_cal, test_size=0.5, random_state=0)
KS_demo = KennardStone(X_slave, num)
CalInd, ValInd = KS_demo.KS()
X_s = X_slave[CalInd] # �������б�ǩ������
X_s_o = X_slave[ValInd] # ������û�б�ǩ������
y_s = y[CalInd] # �����������ı�ǩ
# ѡ�����
TPC_demo = tca_param_choose(X_m_train, y_m_train, X_m_val, y_m_val,
X_s, y_s, X_s_o)
m_op, k_op = TPC_demo.choose_Param(15)
print m_op, "m_op"
my_tca = TCA(dim=m_op)
T_m_cal, T_s_o, T_slave_test, T_s = my_tca.fit_transform(
X_master_cal, X_slave_cal, X_slave_cal, X_s)
T = np.vstack((T_m_cal, T_s))
y = np.vstack((y_cal, y_s))
k1 = np.linalg.lstsq(T, y)[0]
# print np.shape(k1), np.shape(T_slave_test)
y_cal_pre = np.dot(T_slave_test, k1)
RMSEC = np.sqrt(
np.sum(np.square(np.subtract(y_cal, y_cal_pre)), axis=0) /
y_cal_pre.shape[0])
# print RMSEC , "RMSEC"
T_m_cal_, T_s_o, T_slave_test_, T_s_ = my_tca.fit_transform(
X_master_cal, X_slave_cal, X_slave_test, X_s)
T_ = np.vstack((T_m_cal_, T_s_))
y_ = np.vstack((y_cal, y_s))
k1_ = np.linalg.lstsq(T_, y_)[0]
# print np.shape(k1), np.shape(T_slave_test_)
y_test_pre_ = np.dot(T_slave_test_, k1_)
RMSEP = np.sqrt(
np.sum(np.square(np.subtract(y_test, y_test_pre_)), axis=0) /
y_test_pre_.shape[0])
return m_op, RMSEP, y_test_pre_, RMSEC, y_cal_pre
def Dataset_split(self):
num = int(round(self.X_master.shape[0] * self.cal_size))
############## 划分数据集
KS_master = KennardStone(self.X_master, num)
CalInd_master, ValInd_master = KS_master.KS()
X_m_cal = self.X_master[CalInd_master]
y_m_cal = self.y[CalInd_master]
X_m_test = self.X_master[ValInd_master]
y_m_test = self.y[ValInd_master]
# print CalInd_master , ValInd_master
X_s_cal = self.X_slave[CalInd_master]
X_s_test = self.X_slave[ValInd_master]
y_s_cal = self.y[CalInd_master]
y_s_test = self.y[ValInd_master]
KS_master_std = KennardStone(X_m_cal, self.std_num)
CalInd_master_std, ValInd_master_std = KS_master_std.KS()
X_m_std = X_m_cal[CalInd_master_std]
y_m_std = y_m_cal[CalInd_master_std]
X_s_std = X_s_cal[CalInd_master_std]
y_s_std = y_s_cal[CalInd_master_std]
return X_m_cal, y_m_cal, X_m_std, y_m_std, X_s_std, y_s_std, X_s_test, y_s_test, X_s_cal, y_s_cal, X_m_test
def TCA(self , num , i , X_master , X_slave , y , snv = 0 ):
#################
X_m_train, X_m_val, y_m_train, y_m_val = train_test_split(self.X_m_cal, self.y_m_cal[:,[i]], test_size=0.5, random_state=0)
KS_demo = KennardStone(X_slave, num)
CalInd, ValInd = KS_demo.KS()
X_s = X_slave[CalInd] # �������б�ǩ������
X_s_o = X_slave[ValInd] # ������û�б�ǩ������
y_s = y[CalInd] # �����������ı�ǩ
# ѡ�����
TPC_demo = tca_param_choose(X_m_train, y_m_train, X_m_val, y_m_val, X_s, y_s, X_s_o)
m_op, k_op = TPC_demo.choose_Param(20)
print m_op , "m_op"
my_tca = TCA(dim=m_op)
# print y_s.shape
T_m_cal, T_s_o, T_slave_test, T_s = my_tca.fit_transform(self.X_m_cal, self.X_s_cal, self.X_s_cal, X_s)
T = np.vstack((T_m_cal, T_s))
y = np.vstack((self.y_s_cal[:,[i]], y_s))
k1 = np.linalg.lstsq(T, y)[0]
# print np.shape(k1), np.shape(T_slave_test)
y_cal_pre = np.dot(T_slave_test, k1)
RMSEC = np.sqrt(np.sum(np.square(np.subtract(self.y_s_cal[:,[i]], y_cal_pre)), axis=0) / y_cal_pre.shape[0])
# print RMSEC , "RMSEC"
T_m_cal_, T_s_o, T_slave_test_, T_s_ = my_tca.fit_transform(self.X_m_cal, self.X_s_cal, self.X_s_test, X_s)
T_ = np.vstack((T_m_cal_, T_s_))
y_ = np.vstack((self.y_s_cal[:,[i]], y_s))
k1_ = np.linalg.lstsq(T_, y_)[0]
# print np.shape(k1), np.shape(T_slave_test_)
y_test_pre_ = np.dot(T_slave_test_, k1_)
RMSEP = np.sqrt(np.sum(np.square(np.subtract(self.y_s_test[:,[i]], y_test_pre_)), axis=0) / y_test_pre_.shape[0])
return m_op,RMSEP,y_test_pre_,RMSEC,y_cal_pre
if __name__ == '__main__':
import numpy as np
from tca_param_choose import tca_param_choose
rmsep_list = []
m_list = []
for i in range(7):
# 导入数据
print i
DP_demo = datasetProcess(dataType=i)
X_master, X_slave, y = DP_demo.datasetImport()
X_master_cal, X_master_test, X_slave_cal, X_slave_test, y_cal, y_test = DP_demo.datasetSplit_Random(X_master, X_slave, y, next_size=0.2)
X_m_train, X_m_val, y_m_train, y_m_val = train_test_split(X_master_cal, y_cal, test_size=0.5, random_state=0)
KS_demo = KennardStone(X_slave, 64)
CalInd, ValInd = KS_demo.KS()
X_s = X_slave[CalInd] # 从仪器有标签的样本
X_s_o = X_slave[ValInd] # 从仪器没有标签的样本
y_s = y[CalInd] # 从仪器样本的标签
# 选择参数
TPC_demo = tca_param_choose(X_m_train, y_m_train, X_m_val, y_m_val, X_s, y_s, X_s_o)
m_op, k_op = TPC_demo.choose_Param(15)
print m_op , "m_op"
m_list.append(m_op)
my_tca = TCA(dim=m_op)
T_m_cal, T_s_o, T_slave_test, T_s = my_tca.fit_transform(X_master_cal, X_slave_cal, X_slave_cal, X_s)