KataGo V1.3

For discussing go computing, software announcements, etc.
go4thewin
Lives with ko
Posts: 150
Joined: Thu Jan 23, 2020 6:09 am
Rank: 25 kyu
GD Posts: 0
Has thanked: 200 times
Been thanked: 30 times

Re: KataGo V1.3

Post by go4thewin »

with the ladder code version of leela zero, it's much more even.

https://github.com/yssaya/leela-zero-ladder/releases

katago s243 vs lz 270 ladder version, both bots 1 thread 512 playouts:

lz wins 3 games kata wins 1. Katago lost the first three but won the last easily
set katago to play at your level https://docdro.id/sHZU1ti or experiment with gtp4zen ( https://rb.gy/kx2ilb )
lightvector
Lives in sente
Posts: 759
Joined: Sat Jun 19, 2010 10:11 pm
Rank: maybe 2d
GD Posts: 0
Has thanked: 114 times
Been thanked: 916 times

Re: KataGo V1.3

Post by lightvector »

Some new nets have been uploaded:

Roughly 50 Elo gain for the 20-block extended training net, and slightly more than 100 Elo gain for the 30 and 40-block nets. Enjoy!

https://github.com/lightvector/KataGo/releases
Vargo
Lives in gote
Posts: 337
Joined: Sat Aug 17, 2013 5:28 am
GD Posts: 0
Has thanked: 22 times
Been thanked: 97 times

Re: KataGo V1.3

Post by Vargo »

A last 50 game test for the 20b net : g170e-b20c256x2-s2430231552-d525879064
KataGo 1.3.3 g170e-b20c256x2-s2430231552-d525879064 v. LZ017#268 LADDER VERSION
Katago wins 35-15 = 70% (against the "regular" LZ017#268, KG won 74% and 68%)
Twogtp 1.5.1, 1600 visits for both (~ time parity), no error, no duplicate game, all games by resignation.

Details : (KG always appears as W, because of the command -alternate)
k1.jpg
k1.jpg (299.82 KiB) Viewed 9777 times
k2.jpg
k2.jpg (993.34 KiB) Viewed 9777 times
LADDER VERSION : https://github.com/yssaya/leela-zero-ladder/releases
goame
Lives with ko
Posts: 132
Joined: Sat Jan 28, 2017 12:25 am
GD Posts: 0
Been thanked: 7 times

Re: KataGo V1.3

Post by goame »

lightvector wrote:Some new nets have been uploaded:

Roughly 50 Elo gain for the 20-block extended training net, and slightly more than 100 Elo gain for the 30 and 40-block nets. Enjoy!

https://github.com/lightvector/KataGo/releases
Thx for the big nets :-)

b30c320x2
b40c256x2

From what I have seen so far, if the second number is bigger (320), then the net has a very good chance to be stronger and if the first number is bigger (40), then the net has a medium chance to be stronger.

Some ideas which could be tried:

30 and 320 vs 30 and 384
30 and 320 vs 30 and 512

40 and 256 vs 40 and 320
40 and 256 vs 40 and 384
40 and 256 vs 40 and 512
Vargo
Lives in gote
Posts: 337
Joined: Sat Aug 17, 2013 5:28 am
GD Posts: 0
Has thanked: 22 times
Been thanked: 97 times

Re: KataGo V1.3

Post by Vargo »

50 game test between the two newest 20b nets : g170e-b20c256x2-s2430231552-d525879064 v. g170e-b20c256x2-s2971705856-d633407024
The newest , g170e-b20c256x2-s2971705856-d633407024 wins 27-23 (54%)
Twogtp 1.5.1, 1600 visits for both (~ time parity), no error, no duplicate game, all games by resignation.

Details : (s2971 always appears as W, because of the command -alternate)
k3.jpg
k3.jpg (361.42 KiB) Viewed 9715 times
k4.jpg
k4.jpg (889.57 KiB) Viewed 9715 times
Vargo
Lives in gote
Posts: 337
Joined: Sat Aug 17, 2013 5:28 am
GD Posts: 0
Has thanked: 22 times
Been thanked: 97 times

Re: KataGo V1.3

Post by Vargo »

50 game test KG 1.3.3 g170e-b20c256x2-s2971705856-d633407024 v. LZ#270
KG wins 30-20 (60%, "only" , maybe 50 games is not enough, or maybe s2971 doesn't like #270 ;-) )
Again, Twogtp 1.5.1, 1600 visits for both (~ time parity), no error, no duplicate game, all games by resignation.

Details :
k1.jpg
k1.jpg (362.11 KiB) Viewed 9715 times
k2.jpg
k2.jpg (793.06 KiB) Viewed 9715 times
And
Gosei
Posts: 1464
Joined: Tue Sep 25, 2018 10:28 am
GD Posts: 0
Has thanked: 212 times
Been thanked: 215 times

Re: KataGo V1.3

Post by And »

Vargo in this case, maybe another 50?
go4thewin
Lives with ko
Posts: 150
Joined: Thu Jan 23, 2020 6:09 am
Rank: 25 kyu
GD Posts: 0
Has thanked: 200 times
Been thanked: 30 times

Re: KataGo V1.3

Post by go4thewin »

katago 1.3.3 20b s297 1 thread 12 playouts vs gtp4zen7 9d setting; 4-0

Wow I can't beieve it is possibe to have a bot this strong and fast on so few playouts.
Attachments
sgf.zip
sgfs
(3.81 KiB) Downloaded 442 times
set katago to play at your level https://docdro.id/sHZU1ti or experiment with gtp4zen ( https://rb.gy/kx2ilb )
Vargo
Lives in gote
Posts: 337
Joined: Sat Aug 17, 2013 5:28 am
GD Posts: 0
Has thanked: 22 times
Been thanked: 97 times

Re: KataGo V1.3

Post by Vargo »

50 more games KG 1.3.3 g170e-b20c256x2-s2971705856-d633407024 v. LZ#270
KataGo wins 37-13 = 74% (last test, it was 60%)

details : (KG always appears as W, because of the command -alternate)
kg1.jpg
kg1.jpg (320.81 KiB) Viewed 11663 times
kg2.jpg
kg2.jpg (829.08 KiB) Viewed 11663 times
And
Gosei
Posts: 1464
Joined: Tue Sep 25, 2018 10:28 am
GD Posts: 0
Has thanked: 212 times
Been thanked: 215 times

Re: KataGo V1.3

Post by And »

LZ will never catch up with the KataGo. Does anyone see the point in the further development of LZ? what are its advantages? different game style?
And
Gosei
Posts: 1464
Joined: Tue Sep 25, 2018 10:28 am
GD Posts: 0
Has thanked: 212 times
Been thanked: 215 times

Re: KataGo V1.3

Post by And »

to try to beat KataGo, you can play with a handicap or with a small playout. in which case does KataGo play more interesting?
splee99
Dies with sente
Posts: 101
Joined: Thu Nov 15, 2012 9:46 pm
Rank: KGS 2 D
GD Posts: 0
Has thanked: 2 times
Been thanked: 16 times

Re: KataGo V1.3

Post by splee99 »

It depends on your strength, of course. For me, I can't even beat it with 1 playout.
And
Gosei
Posts: 1464
Joined: Tue Sep 25, 2018 10:28 am
GD Posts: 0
Has thanked: 212 times
Been thanked: 215 times

Re: KataGo V1.3

Post by And »

in which of the 2 options is the game more interesting?
go4thewin
Lives with ko
Posts: 150
Joined: Thu Jan 23, 2020 6:09 am
Rank: 25 kyu
GD Posts: 0
Has thanked: 200 times
Been thanked: 30 times

Re: KataGo V1.3

Post by go4thewin »

I think the most interesting is to play against a bot exactly at your level. Find one from the list below. The ranks do not correspond to human ranks. Spend the money to get zen 6 or 7. In strength, Zen 6 ends right where katago starts, so between gtp4zen and katago, you have the full range of human play, except pure beginner. Sabaki allows for any standard size board, komi, and either common ruleset.
zen7 17k -t1 -r3 -s200 -n0 -o4.0 -p0.2

zen7 16k -t1 -r3 -s200 -n0 -o3.0 -p0.25

zen7 15k -t1 -r3 -s300 -n0 -o2.4 -p0.3

zen7 14k -t1 -r3 -s300 -n0 -o2.2 -p0.3

zen6 13k -z6 -r3 -t1 -s300 -n0 -o0.1 -p0.5

zen6 12k -z6 -r5 -t1 -s400 -n0 -o0.1 -p0.6

zen6 11k -z6 -r5 -t1 -s500 -n0 -o0.2 -p0.65

zen6 10k -z6 -r5 -t1 -s600 -n0 -o0.3 -p0.7

zen6 9k -z6 -t1 -s700 -n0 -o0.4 -p0.75

zen6 8k -z6 -t1 -s800 -n0 -o0.5 -p0.8

zen6 7k -z6 -t1 -s1000 -n0 -o0.7 -p0.8

zen6 6k -z6 -t1 -s1500 -n0 -o0.9 -p0.9

zen6 5k -z6 -t1 -s2000 -n0 -o1 -p1

zen6 4k -z6 -t1 -s2500 -n1 -o0.2 -p1

zen6 3k -z6 -t1 -s3000 -n1 -o0.3 -p1

zen6 2k -z6 -t1 -s3000 -n1 -o0.5 -p1

zen6 1k -z6 -r15 -t1 -s3000 -n1 -o1 -p1

Katas297 1d resign threshold 90% 1 thread 1 playout

Katas297 2d resign threshold 90% 1 thread 2 playouts

Katas297 3d resign threshold 90% 1 thread 4 playouts

Katas297 4d resign threshold 90% 1 thread 6 playouts

Katas297 5d resign threshold 90% 1 thread 8 playouts

Katas297 6d resign threshold 90% 1 thread 10 playouts

Katas297 7d resign threshold 90% 1 thread 12 playouts

Katas297 8d resign threshold 90% 1 thread 14 playouts

Katas297 9d resign threshold 90% 1 thread 16 playouts
set katago to play at your level https://docdro.id/sHZU1ti or experiment with gtp4zen ( https://rb.gy/kx2ilb )
Bill Spight
Honinbo
Posts: 10905
Joined: Wed Apr 21, 2010 1:24 pm
Has thanked: 3651 times
Been thanked: 3373 times

Re: KataGo V1.3

Post by Bill Spight »

go4thewin wrote:I think the most interesting is to play against a bot exactly at your level. Find one from the list below. The ranks do not correspond to human ranks. Spend the money to get zen 6 or 7.
One thing to consider is that humans and neural nets have different strengths and weaknesses. I don't know about today's bots, but kyu level bots before the advent of AlphaGo were bad at life and death and at the endgame, by comparison with humans of similar strength. This relative weakness of the bots showed up in particular at the end of play, when the dame get filled. If the same is true of today's bots, there will be opportunities at the end of play to swing the game, either honestly or be pulling off a swindle. ;) :)
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.
Post Reply