Life In 19x19

q30 wrote:Compilation was the same (PC, system soft and compilation parameters).
Because other proofs (with bigger %, i.e. in big plays number matches with GPU enabled) are absent, on this moment the approval "version 14 is stronger, that 15 one" is true...

You can't claim it to be true as long as it is not proven.

Talking about the result of 208:216... If both opponents have the same strength, the chance that one will score at least 216 points in a 424 games match is 73%.

You have got the result 216:208 in favour of Phoenix LZ v.14?
And how many playouts there was in games beginning?

q30 wrote:You have got the result 216:208 in favour of Phoenix LZ v.14?
And how many playouts there was in games beginning?

It's from the link you posted.

I had not posted any link to any game results between Phoenix v.14 and v.15, except mine.
So, on this moment the only proof is, that Phoenix LZ v.14 is stronger, than v.15, with (from basics of probability theory) the probability of their equality (1/2)^4 and the probability of stronger v.15 <(1/2)^4, i.e. v.14 is stronger with >87,5%.

If A and B are equally strong, A will beat B 4-0 with probability 1/16. This doesn't mean that if A beats B 4-0, the probability they are equally strong is 1/16.

q30 wrote:I had not posted any link to any game results between Phoenix v.14 and v.15, except mine.

This is becoming really ridiculous... You post something, I answer, you say, you didn't post it.

I promise this is my last post in this thread...

You posted this:

q30 wrote:In these results 0.14 wins 0.15 by black and by white.

Here:
https://www.lifein19x19.com/forum/viewt ... 44#p235944

Welcome on my blacklist.

dfan wrote:If A and B are equally strong, A will beat B 4-0 with probability 1/16. This doesn't mean that if A beats B 4-0, the probability they are equally strong is 1/16.

OK. And what the probability they are equally strong is in this case?

as0770 wrote:

q30 wrote:I had not posted any link to any game results between Phoenix v.14 and v.15, except mine.

This is becoming really ridiculous... You post something, I answer, you say, you didn't post it.

I promise this is my last post in this thread...

You posted this:

q30 wrote:In these results 0.14 wins 0.15 by black and by white.

Here:
https://www.lifein19x19.com/forum/viewt ... 44#p235944

Welcome on my blacklist.

This is really ridiculous, how You read posts...
Where You had found in the post by this link word "Phoenix" (or "Феникс")?!

q30 wrote:

dfan wrote:If A and B are equally strong, A will beat B 4-0 with probability 1/16. This doesn't mean that if A beats B 4-0, the probability they are equally strong is 1/16.

OK. And what the probability they are equally strong is in this case?

We don't have enough information to say; it depends on our prior guess of how likely various strength differences are.

I had determined the 2018 year winners of different sizes LeelaZero weights (details):
"bantamweight" <|= 2 ^ 23 B (< 12 MiB) - best_5b.txt;
"featherweight" 2 ^ 24 B (12 - 24 MiB) - I haven't;
"lightweight" 2 ^ 25 B (24 - 48 MiB) - LeelaMaster_E08.txt;
"welterweight" 2 ^ 26 B (48 - 96 MiB) - LeelaMaster_W11.txt;
"middleweight" 2 ^ 27 B (96 - 192 MiB) - LeelaMaster_GX35.txt;
"light heavyweight" 2 ^ 28 B (192 - 384 MiB) - elf_converted_weights.txt;
"heavyweight" 2 ^ 29 B (384 - 768 MiB) - leelaz-model-swa-9-72000.txt;
"super heavyweight" >|= 2 ^ 30 B (> 768 MiB) - I haven't.

The best of the 2018 year LeelaZero weights are next (details):
1) "light heavyweight" 2 ^ 28 B (192 - 384 MiB) - elf_converted_weights.txt;
2) "heavyweight" 2 ^ 29 B (384 - 768 MiB) - leelaz-model-swa-9-72000.txt;
3) "middleweight" 2 ^ 27 B (96 - 192 MiB) - LeelaMaster_GX35.txt | leelaz-model-swa-8-64000_quantized.txt

The top list of GTP engines without using GPU (details):
Highest level
1) LeelaZero
High level
2) Leela
Middle level
3) Rayon
4) Hiratuka
5) Pachi_DCNN
6) Pachi
7) Ray
8) Ray(RLO)
9) MoGo

The top list of GTP engines without using GPU (details):
Highest level
1) LeelaZero
High level
2) Leela
Middle level
3) Rayon
4) Hiratuka
5) Zenith
6) Pachi_DCNN
7) Pachi
8) Ray
9) MoGo

q30 wrote:The top list of GTP engines without using GPU.

Sorry q30,
You posted 2 different lists with the same description.
Did you mean to make it with and without GPU?

Yes, the description is the same and the lists are different (the list was updated: was added Zenith and was removed Ray(RLO) because of about sense). What is the problem?
On my PC without modern NVidia card it isn't possible to test with GPU...