我一直在使用sklearn的模型(SVM,Logistic回归,MLP,......)直到昨天我没有任何问题,但我不知道为什么,目前当我尝试适合模型时,这需要夸大其词多少时间 .
例如,对于具有6个特征的551个样本,尝试使用多项式内核拟合支持向量机并改变参数:
-
C = 1.00度= 1.00 Coef = 0.000 Gamma = 0.25 15.124秒 .
-
C = 1.00度= 1.00 Coef = 0.000 Gamma = 0.75 22.937秒 .
-
C = 1.00度= 1.00 Coef = 3.000 Gamma = 0.25 11.703秒
-
C = 1.00度= 1.00 Coef = 3.000 Gamma = 0.75 18.81秒
-
C = 1.00度= 2.00 Coef = 0.000 Gamma = 0.25 316.115秒
-
C = 1.00度= 2.00 Coef = 0.000 Gamma = 0.75 74.530秒
-
C = 1.00度= 2.00 Coef = 3.000 Gamma = 0.25 270.514秒
-
C = 1.00度= 2.00 Coef = 3.000 Gamma = 0.75 357.194秒
from sklearn import svm
for C in C_values:
for degree in degree_values:
for coef in coef_values:
for gamma in gamma_values:
print('C = {0:.2f} Degree = {1:.2f} Coef = {2:.3f} Gamma = {3:.2f} '.format(C,degree,coef,gamma))
clf = svm.SVC(C = C, degree = degree, gamma = gamma, coef0 = coef, kernel = 'poly', decision_function_shape = decision_function_shape )
time_i = time()
clf.fit(X_train, Y_train)
time_f = time()
print('str(time_f - time_i) '+'seconds')
我从昨天到今天没有任何改变,我认为5分钟适合550个样本和6个功能的模型被夸大了 .
什么可以产生这种增加的时间?
谢谢
EDIT :
我一直在尝试使用random_state参数不同的时间,结果是类似的:
-
C = 1.00度= 1.00 Coef = 0.000 Gamma = 0.25 ...... 7.359 scnds
-
C = 1.00度= 1.00 Coef = 0.000 Gamma = 0.75 ...... 26.156 scnds
-
C = 1.00度= 1.00 Coef = 3.000 Gamma = 0.25 ...... 8.781 scnds
-
C = 1.00度= 1.00 Coef = 3.000 Gamma = 0.75 ...... 16.437 scnds
-
C = 1.00度= 2.00 Coef = 0.000 Gamma = 0.25 ..... 259.248 scnds
-
C = 1.00度= 2.00 Coef = 0.000 Gamma = 0.75 ...... 219.170 scnds
-
C = 1.00度= 2.00 Coef = 3.000 Gamma = 0.25 ...... 302.201 scnds
-
C = 1.00度= 2.00 Coef = 3.000 Gamma = 0.75 ...... 345.163 scnds