Life In 19x19

I think it is playable

also Ke Jie:

Click Here To Show Diagram Code

[go]$$B
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . O X . . . . . . . . . . . . . . . . |
$$ | . O X O . . . . . . . . . O X . . . . |
$$ | . O X O . . . . . , W . . O . X . . . |
$$ | . X O O . . . . . . . . . . . . . . . |
$$ | X . X . . . . . . . . . . . . X . . . |
$$ | . X . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . X . . . . . . . X . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Dwyrin was afraid to play a three space extension because of his 9D Pro opponent. Instead of tenuki he could have played like Ke Jie or Cho Chikun. But he could not manage to do so, because of the Know-it-all's advice to never play a two space extension.

Ke(1) and Cho are short being defensive. Ke(2) is short creating thickness reducing Black's sphere of influence. They are different from the early opening position, in which overconcentration is considered bad.

Before I buy a new PC with GPU, I'd like to know if I use Leela Zero correctly.

Is my notebook computer (Intel i5 2.2G, Windows 10 64 bit, 4G RAM) really too slow in terms of visits per seconds?

I have tested it with Lizzie, the following results in 3-5 vps and 12 vps only:

C:/leela-zero-0.17-cpuonly-win64/leelaz.exe -g -b 0 -w C:/leela-zero-0.17-cpuonly-win64/266_40b_15926.gz

C:/leela-zero-0.17-cpuonly-win64/leelaz.exe -g -b 0 -w C:/leela-zero-0.17-cpuonly-win64/117_15b_10214.gz

and Some people told me should be 30 vps or higher, at least for 15 blocks.

Is there any parameter I can try? Thanks again.

If you have an i5 5200u 3.5 playouts a second seems about right. Even 3.5 visits seems ok. if you can use opencl, try using the gpu version https://software.intel.com/en-us/node/540387

Gomoto wrote:I think it is playable

What do LZ and KataGo say about the close extensions in these cases? It matters what is in the adjacent corner.

For the last position, 15-block Katago likes Ke Jie's move:
but other moves are playable as well, everything is within the margin of error (<1%).

15-block LeelaZero prefers the normal low extension, but thinks that Ke Jie's move is playable as well (-0.5%).

P.S. My notebook computer is CPU-only: Intel Core i7-8650u CPU @1.90GHz 2.11 GHz, 32G RAM, 64 bits. On this particular position I get about 80 playouts/second.

jlt wrote:but other moves are playable as well, everything is within the margin of error (<1%).

Careful with this thinking. In Katago review I noticed it often shows my game move as an acceptable alternative with only slightly lower winrate (like 49% instead of 51%). However, going forward in the game, both me and opponent plays the bots top choice for 1-2 more moves it suddenly drops by 10% or so. So its evaluations doesn't seem reliable for non-top moves (less so than LZ 15b at least).

jann wrote:In Katago review I noticed it often shows my game move as an acceptable alternative with only slightly lower winrate (like 49% instead of 51%). However, going forward in the game, both me and opponent plays the bots top choice for 1-2 more moves it suddenly drops by 10% or so. So its evaluations doesn't seem reliable for non-top moves (less so than LZ 15b at least).

I know this can happen in complex fights, because some branches are not explored enough at the first move, but in the particular case I explored above, there is no such complex fight, and alternative moves are explored with several thousand visits, so there is no risk of a sudden 10% drop.

jann wrote:

jlt wrote:but other moves are playable as well, everything is within the margin of error (<1%).

Careful with this thinking. In Katago review I noticed it often shows my game move as an acceptable alternative with only slightly lower winrate (like 49% instead of 51%). However, going forward in the game, both me and opponent plays the bots top choice for 1-2 more moves it suddenly drops by 10% or so. So its evaluations doesn't seem reliable for non-top moves (less so than LZ 15b at least).

What about the number of visits? Often when the human move is not the same as the bot's top choice, the human move gets relatively few visits by comparison with the top choice. Few visits means unreliable winrate estimates. In the GoGoD commentaries by Elf, human plays with fewer than 500 visits are not directly evaluated. Instead, they inherit their winrate estimates from Elf's next top choice. Even better, IMO, is to make the human play directly and see how the bot evaluates it. Then you have a comparable comparison.

If the winrate estimates of a bot's top choices change suddenly on another top choice, that indicates a horizon effect. The bot sees something that makes it change its winrate estimates. If that happens regularly, then the bot needs more rollouts to make reliable evaluations.

My vague recollection is this doesn't seem related to complex fighting or horizont effect. Few visits are more of a reason (I have 1060 and use 15b nets, and only casual walkthrough my games). But the point is, this doesn't seem to happen with LZ 15b but frequent with Katago 15b.

It would be interesting to see test matches where the value/policy nets are swapped independently between LZ and Kata. (I suspect this effect might be caused by LZ policy being better and more focused on key lines, thus allowing better evaluations with the first few dozen visits.)

go4thewin wrote:If you have an i5 5200u 3.5 playouts a second seems about right. Even 3.5 visits seems ok. if you can use opencl, try using the gpu version https://software.intel.com/en-us/node/540387

I upgraded my notebook computer's graphics driver from ASUS, which is per advice from Intel Driver Support & Assistant (a tool), it seems OpenCL 1.2 and 2.0 are in place, Leela Zero GPU version fails (see below), but Leela Zero CPU-only version still works.


C:\WINDOWS\system32>C:/leela-zero-0.17-win64/leelaz.exe -g -b 0 -w C:/leela-zero-0.17-win64/266_40b_15926.gz
Using OpenCL batch size of 5
Using 10 thread(s).
RNG seed: 10561116271007652040
Using per-move time margin of 0.00s.
BLAS Core: Haswell
Detecting residual layers...v1...256 channels...40 blocks.
Initializing OpenCL (autodetecting precision).
Detected 2 OpenCL platforms.
Platform version: OpenCL 1.2 CUDA 9.1.112
Platform profile: FULL_PROFILE
Platform name: NVIDIA CUDA
Platform vendor: NVIDIA Corporation
Device ID: 0
Device name: GeForce 920M
Device type: GPU
Device vendor: NVIDIA Corporation
Device driver: 388.57
Device speed: 954 MHz
Device cores: 2 CU
Device score: 1112
Platform version: OpenCL 2.0
Platform profile: FULL_PROFILE
Platform name: Intel(R) OpenCL
Platform vendor: Intel(R) Corporation
Device ID: 1
Device name: Intel(R) HD Graphics 5500
Device type: GPU
Device vendor: Intel(R) Corporation
Device driver: 20.19.15.4404
Device speed: 900 MHz
Device cores: 24 CU
Device score: 620
Device ID: 2
Device name: Intel(R) Core(TM) i5-5200U CPU @ 2.20GHz
Device type: CPU
Device vendor: Intel(R) Corporation
Device driver: 5.2.0.10094
Device speed: 2200 MHz
Device cores: 4 CU
Device score: 520
Selected platform: NVIDIA CUDA
Selected device: GeForce 920M
with OpenCL 1.2 capability.
Error creating OpenCL context: clCreateContext: 999

jann wrote:My vague recollection is this doesn't seem related to complex fighting or horizon effect. Few visits are more of a reason (I have 1060 and use 15b nets, and only casual walkthrough my games).

"Few visits" is how many? 10? 100? 1000?

jlt wrote:"Few visits" is how many? 10? 100? 1000?

As mentioned 1060 on 15b for casual reviews, so 10-15 sec per move, which gives around 10k visits in total. So inferior alternatives may not get more than 100 visits. That's low for sure, but again the point is the difference between LZ and Kata behavior.