Eddie Huang

Hi RohanM, I'm wondering how you would implement this Zobrist hashing. This codebase is largely reliant on numpy/scipy and I would prefer if your zobrist hashing used numpy/scipy if possible....

I might not have the best understanding of Zobrist hashing but since you're working with small boards anyway. Why not just compare the board states directly? I doubt it would...

Yeah. 5 is quite a common score. Maybe try using more varied images

I also don't think the `expand_as` in `weight = (weight - weight_mean) / std.expand_as(weight)` is necessary since `std` should have the same number of dimensions as `weight`

I actually think Peter’s suggestion is even better. It would avoid the tendency to make this a chat and still serve its purpose