As one of the strongest public bots available, it would seem that ELF would rarely resign early if you didn't change it's resign settings. Therefore it would probably be beneficial to study games where ELF resigned early to see the masterful moves which forced its defeat. Here is a game between ELF v1 vs LeelaMaster GX59 (https://github.com/pangafu/LeelaMasterWeight) where ELF as black resigned after 111 moves with 7.5 komi. Enjoy.

_________________
http://tchan001.wordpress.com
A blog on Asian go books, go sightings, and interesting tidbits
Go is such a beautiful game.

How many playouts was each side getting in this game? Elf can suffer from blindspots (particularly on low playouts and with ladders), so I could imagine another player who can exploit that could win (e.g. 1000 playouts of a weaker but faster network vs 100 Elf). In this game at the end Elf initially thinks if black c15 white will f16 allowing black to d14 connect, but within about 300 playouts realises that white can cut and then black is stuffed due to white b16 afterwards. When there is an initial blindspot like this I've found you need quite a lot of playouts for Elf to realise this several moves prior (e.g. if it spots the opponents good next move with only 300 playouts then to see it when reading 5 moves ago might need 10000). But why did this Elf play e14? My Elf v1 from 0-800 (I'm on weak computer atm) doesn't want to play there but the much more natural looking f16, and can see white's tesuji in the corner which means black has to lose g15 anyway (but thinks black is still slightly leading despite that). Also that black didn't block at j3 before defending the corner suggests to me Elf didn't have enough playouts to realise the ladder works for black (I found Elf initially wanted to block, then wanted to c1 because principal variation shows white cutting and black not laddering, and I suspect with loads of playouts will go back to block once it finds the ladder).

But yes d9 wedge is indeed a cute move, which Elf thinks is good once played and only dedicated a few playouts to (up to 300 total now). EDIT: with about 1300 playouts total Elf now thinks d9 is best move and not a surprise. Does that mean it wouldn't play e9 jump with more time....?

I just set up Sabaki to play on default settings so not sure about the number of playouts.

_________________
http://tchan001.wordpress.com
A blog on Asian go books, go sightings, and interesting tidbits
Go is such a beautiful game.