act and learn input ranges about reinforcejs

[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?