
First, we non-boolean attributes (ie, attributes that can take more than
two values) we have that the measure of information
I(p_i/(p_i+n_i),n_i/(p_i+n_i)) becomes:
I(A) = \sum_i -P(v_i)lg(P(v_i))

We then recall:

Gain = Information_before - Reminder_information

Hence a possibility for a ratio can be

Ratio = Reminder_information/Information_before

