-
One of the issue with the current Leelaz is that it is too strong for human players, esp amateur players. While there is a great progress for the handicap games, I am wondering if the following appro…
-
Will update Windows build here.
**Complete package: https://drive.google.com/file/d/1bdIlVDJ3x6FZtX5fmuG6wNbb57GFU8S0/view (6/27/2019, use new cudnn DLL)**
-------------------------
Original co…
-
LZ is at least as strong as Elf since we switched to 40 blocks, ie LZ174.
However, despite significant improvements (15 new nets, with LZ189 now), performance of LZ against Elf is still relatively we…
-
https://github.com/gcp/leela-zero/blob/9e903eda6e93fb65375ea03127e4f62268913175/src/UCTNode.cpp#L264
In some cases fpu_eval may result negative, which I think is unintended behaviour, as this should …
-
@roy7: Watched the match of Leela Zero against AZ last night, where Leela Zero
lost 2 large semiais, but still pretending it's leading until the very end:
https://online-go.com/game/11903792, got…
-
UCT algorithm for choosing which branch to explore next appears to be somewhat sub optimal in practice. As an easy proof of concept I trained a very small neural network to replace it and in the synth…
-
Before ELF self-play games were introduced, LZ was great at passing correctly when the game was finished and all captured stones were removed from the board. However, after ELF self-play games were in…
-
I think it would be better to replace "winrate" by "points ahead/behind". Maybe it would be possible to create such a neural network from current Leela Zero self play? and then replace the value netwo…
Diapa updated
5 years ago
-
Hi, I'm interested in trying your bot on my account which has 30mil salt. How does your bot perform? How much does it gain on average per day, etc.? Any data on this at all would be very appreciated! …
-
I have following function beginning:
[EntryPoint]
public static void ComputeKohonen(float[] plainNetwork, float[] learningExamples, int[] winRates, int xDim, int yDim, int examplesCount, int…