[INDOLOGY] Artificial Neural Network for Sanskrit programming

[hellwig7 at gmx.de]

Dear Dhaval,

thanks for sharing this interesting experiment!

Two questions come to my mind:

1. Converting letters (nominal scale) to floating point values (metrical 
scale) in the function weightage() looks a bit unfamiliar to me. Actually, 
the classifier will consider "a" and "A" as more similar than, for instance, 
"a" and "u" in this way. Is this conversion grounded theoretically?

2. Could you provide per-class error rates? Does the NN perform much better 
for "simple" classes such as avyayibhava?

But apart from this, it seems to work!

Best, Oliver

---
Oliver Hellwig, University of Düsseldorf, Germany