在scikit学习中训练不同特征维数的逻辑回归模型

Question

在Windows上使用Python 2.7。想要对一个分类问题拟合一个使用特征T1和T2的逻辑回归模型，目标是T3。

我显示了T1和T2的值，以及我的代码。问题是，既然T1有维度5,T2有维度1，我们应该如何预处理它们，以便它可以被scikit-learn逻辑回归训练正确地利用?

BTW，我的意思是对于训练样本1，其T1的特征为[ 0 -1 -2 -3], T2的特征为[0]，对于训练样本2，其T1的特征为[ 1 0 -1 -2], T2的特征为[1]，…

import numpy as np
from sklearn import linear_model, datasets
arc = lambda r,c: r-c
T1 = np.array([[arc(r,c) for c in xrange(4)] for r in xrange(5)])
print T1
print type(T1)
T2 = np.array([[arc(r,c) for c in xrange(1)] for r in xrange(5)])
print T2
print type(T2)
T3 = np.array([0,0,1,1,1])
logreg = linear_model.LogisticRegression(C=1e5)
# we create an instance of Neighbours Classifier and fit the data.
# using T1 and T2 as features, and T3 as target
logreg.fit(T1+T2, T3)

,

[[ 0 -1 -2 -3]
 [ 1  0 -1 -2]
 [ 2  1  0 -1]
 [ 3  2  1  0]
 [ 4  3  2  1]]

T2,

[[0]
 [1]
 [2]
 [3]
 [4]]

Answer 1

需要使用numpy.concatenate连接特征数据矩阵。

import numpy as np
from sklearn import linear_model, datasets
arc = lambda r,c: r-c
T1 = np.array([[arc(r,c) for c in xrange(4)] for r in xrange(5)])
T2 = np.array([[arc(r,c) for c in xrange(1)] for r in xrange(5)])
T3 = np.array([0,0,1,1,1])
X = np.concatenate((T1,T2), axis=1)
Y = T3
logreg = linear_model.LogisticRegression(C=1e5)
# we create an instance of Neighbours Classifier and fit the data.
# using T1 and T2 as features, and T3 as target
logreg.fit(X, Y)
X_test = np.array([[1, 0, -1, -1, 1],
                   [0, 1, 2, 3, 4,]])
print logreg.predict(X_test)