def pls2_predict_cv(self):
x_train, x_test, y_train, y_test = self.cv()
y_allPredict = np.ones((1, self.max_components))
pls = _NIPALS(self.max_components)
for i in range(self.n_fold):
# print 'i:', i
y_predict = np.zeros((y_test[i].shape[0], self.max_components))
# print 'y',np.shape(y_predict)
x_trainMean = np.mean(x_train[i], axis=0)
y_trainMean = np.mean(y_train[i], axis=0)
x_testCenter = np.subtract(x_test[i], x_trainMean)
# y_testCenter = np.subtract(y_test,y_trainMean)
w, t, p, list_coef_B = pls.fit(x_train[i], y_train[i],
self.max_components)
for j in range(self.max_components):
# print 'j:', j
# print "textCenter", np.shape(x_testCenter), np.shape(list_coef_B[j])
y_pre = np.dot(x_testCenter, list_coef_B[j])
y_pre = y_pre + y_trainMean
# print 'y_pre:', np.shape(y_pre)
y_pre_sum = np.sum(y_pre, axis=1)
# print 'y_pre_sum:' np.shape(y_pre_sum)
y_predict[:, j] = y_pre_sum.ravel()
y_allPredict = np.vstack((y_allPredict, y_predict))
y_allPredict = y_allPredict[1:]
return y_allPredict, self.y
def pls_fit(self):
pls_cv = PLSCV(self.x_train, self.y_train)
y_predict_all, y_measure = pls_cv.cv_predict(self.n_folds,
self.max_components)
RMSECV, min_RMSECV, comp_best = pls_cv.cv_mse(y_predict_all, y_measure)
# print RMSECV
pls = _NIPALS(comp_best)
W, T, P, lists_coefs = pls.fit(self.x_train, self.y_train, comp_best)
# print self.y_train
coefs_B = lists_coefs[comp_best - 1]
# drawings.rmsecv_comp_line(self.max_components, RMSECV)
return W, T, P, coefs_B, RMSECV, min_RMSECV, comp_best
def pls_fit(self):
pls_cv_demo = Cross_Validation(self.X_train, self.y_train, self.folds, self.max_comp)
# print np.shape(self.X_train),np.shape(self.y_train)
y_allPredict, y_measure = pls_cv_demo.predict_cv()
# print 'shape:', np.shape(y_allPredict), np.shape(y_measure)
# print y_measure
RMSECV, min_RMSECV, comp_best = pls_cv_demo.mse_cv(y_allPredict, y_measure)
# print RMSECV , "rmsecv"
# RMSECV, min_RMSECV, comp_best = pls_cv_demo.cv_mse_F(y_allPredict, y_measure, alpha=0.05, num_tr=self.X_train.shape[0])
# print "comp_best" , comp_best
pls_demo = _NIPALS(comp_best)
weight_W, score_T, loading_P, List_coef_B = pls_demo.fit(self.X_train, self.y_train, comp_best)
coefs_B = List_coef_B[comp_best - 1]
# print np.shape(coefs_B), np.shape(List_coef_B), comp_best
return weight_W, score_T, loading_P, comp_best, coefs_B, RMSECV
def pls2_fit(self):
pls2_cv_demo = Cross_Validation(self.X_train, self.y_train, self.folds, self.max_comp)
y_allPredict, y_measure = pls2_cv_demo.pls2_predict_cv()
y = np.sum(y_measure, axis=1)
y = y.reshape(self.X_train.shape[0], 1)
# print 'measure:', np.shape(y), y
RMSECV, min_RMSECV, comp_best = pls2_cv_demo.pls2_mse_cv(y_allPredict, y)
# RMSECV, min_RMSECV, comp_best = pls_cv_demo.cv_mse_F(y_allPredict, y_measure, alpha=0.05, num_tr=self.X_train.shape[0])
pls2_demo = _NIPALS(comp_best)
weight_W, score_T, loading_P, List_coef_B = pls2_demo.fit(self.X_train, self.y_train, comp_best)
coefs_B = List_coef_B[comp_best - 1]
# print 'infor:', comp_best, np.shape(self.X_train), np.shape(self.y_train), np.shape(List_coef_B), np.shape(coefs_B)
# print comp_best , "comp_best"
# print coefs_B.shape , "shape"
return weight_W, score_T, loading_P, comp_best, coefs_B, RMSECV
def cv_predict(self, n_folds, max_components):
x_tr, x_te, y_tr, y_te = baseCV.CV(self, self.x, self.y, n_folds)
# print x_train, x_test, y_train, y_test
# print np.shape(y_predict)
y_predict_all = np.ones((1, max_components))
# print np.shape(Y_predict)
pls = _NIPALS(max_components)
for i in range(n_folds):
y_predict = np.zeros((x_te[i].shape[0], max_components))
xtrainmean = np.mean(x_tr[i], axis=0)
ytrainmean = np.mean(y_tr[i], axis=0)
xte_center = np.subtract(x_te[i], xtrainmean)
yte_center = np.subtract(y_te[i], ytrainmean)
w, T, P, lists_coefs_B = pls.fit(x_tr[i], y_tr[i], max_components)
# print lists_coefs_B
for j in range(max_components):
y_pre_center = np.dot(xte_center, lists_coefs_B[j])
Y_pre = y_pre_center + ytrainmean
# print np.shape(Y_pre)
y_predict[:, j] = Y_pre.ravel()
y_predict_all = np.vstack((y_predict_all, y_predict))
y_predict_all = y_predict_all[1:]
# print np.shape(y_predict_all)
# print np.shape(self.y)
# print self.y
return y_predict_all, self.y