Does anyone know if the AlphaGo team is saving the probability of winning assignments that AlphaGo gave its moves?
It would be fascinating to see how early AlphaGo assigned very high probability of it winning. It would also be interesting to see if there were particular moves which changed this assignment a lot. For instance, are there moves that Lee Sedol made for which the win probability is very different for the AlphaGo moves before and after?
In the introduction to the third match, the project lead talked about how Alphago evaluated the controversial 37th move in the second match. Apparently the policy network decided that it had a one in 10,000 chance of a human making that move, But the value network overrode this. So it seems like they are recording this kind of data and looking at it in cases where it's interesting.
Thanks, I'm only just watching match 2 and hadn't seen that. I really hope they publish this data. It would be incredibly interesting to construct a "review of the game by AlphaGo" from this data.
It would be fascinating to see how early AlphaGo assigned very high probability of it winning. It would also be interesting to see if there were particular moves which changed this assignment a lot. For instance, are there moves that Lee Sedol made for which the win probability is very different for the AlphaGo moves before and after?