RangeTransform of SVMProblem generated on the training is not always applicable to the validation or test

[mkommend]

An exception is thrown if the range transform was created on different set of input variables than its applied on. This is a direct result of the use of the SVM helper class, which removes input variables that have not at least two different non-NaN non-Infinity values. Thus the validation can contain more input variables than the training and an IndexOutOfRangeException is thrown in RangeTransform.Transform(double,int) line 134.

[gkronber]

r6002: fixed minor issues in LDA, LR, SVC and SVR to make sure everything works correctly in presence of NaN and infinity values.

[mkommend]

Thx for correting this.