It is a bit roundabout, since it involves converting maia models to onnx before loading into pytorch and some outdated versions of libraries (maia/lc0 are a little old). We were using this for transfer learning for a competition, so we needed some flexibility that we didn't know how to do quickly/easily in TF.
Hope this helps.
------------------
Personal note: given your interest in chess ai and your starcraft username, I think we would have a lot of shared interests. Feel free to reach out (info is in my profile).
If you are interested in this field (exploring the limits of neural models using formal language theory), I help run a weekly seminar on it, Formal Languages and Neural Networks:
I also found the color scheme to be difficult to understand. A friend suggested interpolating between red and green depending on how close you are to the correct key. This is not too hard, since red is (255, 0, 0) and green is (0, 255, 0), so you can compute a distance (normalized to [0, 1]) and output (255 x d, 255 x (1-d), 0) to get the interpolated color.
Hey, I just released a new update that adds multiple color schemes to help with colorblindness! At some point I may add a color blind mode that does away with the need for colors altogether, but for now this should make things a bit better. Also, there is now a gauge next to the board so you can more easily see the range of colors (unfortunately, it doesn't yet work on Firefox).
Check it out!
I also skipped forward a day, so if you already played today you can play again with the next word to try it out.
Ah, the other color schemes look really good (especially I like the heatmap one). However, the default stoplight one is nearly impossible for me to disambiguate. For example, in [1], I can't tell which of mine are closer at all (I don't think I am colorblind, but maybe I'm just in for a surprise today).
Anyway, overall, a very fun variant. Thanks for sharing!
The mean transduction in the sense of sequence-to-sequence models (transducing one sequence to another).
Attention based models don't necessarily need to be sequence to sequence. They can be classifiers, decoder only, etc. Attention is just one tool in the ML architecture toolkit.
