multiclass-utils

[percyliang]

This program is very inefficient because it's written in Ruby. Make it faster.

Also, it should reject datasets that:

• have repeated features
• have features which are not sorted
• have not exactly one space between features and labels

Says Benoit Favre:
The dataset checker for mutliclass does not check that the dataset
is well formed, it just counts features/labels: it should check that
labels go from 1 to N, features go from 1 to M in ascendent manner,
with valid values (no infinity), that empty examples do not exist
(only a label, no features), that feature ids are not sparse (most
classifiers fail on that one).