Life In 19x19

I took it as a serious question - I was wondering the same myself, not just in terms of strength, but if the moves were "human" enough that it would work well as a review partner.

EdLee - definitely would be better if/when I can remember the moves of a game a bit better. That might be a while though, particularly for my games on DGS. Pretty hard for me to remember a move an opponent made a month ago or more...I found a reasonable workaround for me - view the game on my iPad as I'm analyzing it on Leela. Then I can see the game's next move, alongside a list of move(s) that work best. Two "clicks" instead of one each move, but better than forcing it to reanalyze every move. The best solution, obviously, would be a viewing option where you could tell it to display the next move if you so choose.

Try it out and decide for yourself if you like it as a review partner or not.

I like it very much: it gives me lots of new ideas to play, and it corrects my mistakes.

The only caveat: when it comes to life-and-death, you need to let it run longer since on quick analysis it may miss the status of some groups.

aTan wrote:Another example using your game

Thanks, the script is interesting. How did you interact with Leela is there an API or something? I remember that chess programs (e.g. Fritz, back in the day I had Fritz 8 when I was a chess player) used to be able to output analysed games to PDF. There would be 3 or 4 diagrams per game, and comments and variations. It would be interesting for the same kind of thing to be available as an output from a Go program; ask it to analyse the game, go to work, come home from work, and a fresh print-out of the game is by the printer ready to be looked through on a real board.

About how far to you need to let it run out to be confident that the move(s) it's giving you are indeed the best? Would that be measured in nodes?