为什么支持向量回归器预测所有相同的数字？

Question

我正在尝试一种新的回归方法，并且一直对使用支持向量机来做这件事感兴趣。问题是我正在为它提供训练数据，以便它可以预测测试数据。当它预测训练数据时，它预测如下 -

array([ 8.00000947,  8.10000947,  7.90000947,  8.40000947,  8.50000947,
        8.10000947,  7.90000947,  8.20000947,  8.40000947,  8.20000948,
        8.40000948,  8.40000947,  8.00000947,  8.10000947,  8.50000948,
        8.40000947,  8.60000947,  8.40000948,  8.40000948,  8.00000947,
        8.50000948,  8.30000948,  7.99922823, ...

但是，当我预测测试数据时，它预测如此 -

array([ 7.92969697,  7.92969697,  7.92969697,  7.92969697,  7.92969697,
        7.92969697,  7.92969697,  7.92969697,  7.92969697,  7.92969697,
        7.92969697,  7.92969697,  7.92969697, ...

代码如下 -

import sklearn
import pandas as pd
import numpy as np
from sklearn import svm

df = pd.read_excel('featureset.xlsx')
df.shape
(280, 23)

cols = pd.factorize(df.columns)
X= df
X.columns = cols[0]
Y.columns = [51]

X_train = X[:196]
X_test = X[196:]
y_train = Y[:196]
y_test = Y[196:]

y_train.shape
(196L,)

y_test.shape
(84L,)

clf = svm.SVR(kernel='rbf', C=1e3, degree=2)
clf.fit(X_train, y_train).predict(X_test)

因此，导致

array([ 7.92969697,  7.92969697,  7.92969697,  7.92969697,  7.92969697, ...