act and learn input ranges

[salamanders]

1. Should all state inputs to act be 0<=stateX<1?
2. Should all reward inputs be 0<=reward<1?
3. Is there any way to get out "nope, that wasn't a good reply. I want a second opinion!" (second place answer, etc)

[gb96]

1. I have been having some success with categorical state inputs. Since my categories are represented as strings I ended up implementing a string hashcode for state inputs since I quickly learned that state inputs needed to be numeric. My string hashcodes are integers in the thousands and they seem to be working with the library, so I think your suggested [0, 1) range for inputs is overly constrained.
2. I have been providing integer rewards between -100 and 100, and these seem to be working too, however I have been encountering NaN in the net weights after calls to learn(), so maybe I should try a smaller range?