TalkChess.com

Entered "Beth Harmon" in My KO tourney...engine lost in first round 1-3...may have some time control (time mgmt) issues Thx AR :) :wink: https://www.mediafire.com/file/u1pcterttwimtn6/2021Super_Engines_KO.7z/file The net is quite a bit weaker than the other nnue nets — -200 to night nurse and -300 ...

MikeB wrote: ↑

Sun Mar 07, 2021 1:24 am

mclane wrote: ↑

Sun Mar 07, 2021 12:17 am

Is honey a Stockfish clone ?!

Yes

Except my nets. They have nothing to do with Stockfish.

1. Teaching other people how simple it is to write a basic mcts/nn engine with a0lite python.
2. Experimenting with new nn architectures and non-RL training approaches.
3. Combining ab/nnue and mcts/nn in a hybrid approach. (Was a0lite julia, renamed Bender)
4. Reach 3300 ccrl
5. Have fun.

how is it possible that in 2021 nobody has been yet taken to court for copying something NN related? so at lease we'd know how the judge ruled that case Companies have been claiming copyrights and licensing pretrained models (read nn weights) for years. They’ve been watermarking them and applying o...

After all of this Sturm und Drang I think two things are clear:

1) everyone has heard of FF2
2) the SF/Leela/lichess communities’ words and actions speak for themselves.

hgm wrote: ↑

Sun Feb 28, 2021 8:06 pm

But with the usual method of learning (= weight adjusting), what is the same will stay the same. The old and the new half would never diverge from each other.

I hope Chris gives it a shot. Would love to see the results of his 512x2x16x16 net.

Did you really try that? Backpropagation through an NN to adjust the weights according to gradient descent of the total square error would adjust weight in the proportion to which they contributed to the error. If they are all zero, they don't contribute anything. So they would never be adjusted. F...

Did you really try that? Backpropagation through an NN to adjust the weights according to gradient descent of the total square error would adjust weight in the proportion to which they contributed to the error. If they are all zero, they don't contribute anything. So they would never be adjusted. F...

How to make a double-sized net as good as SF NNUE in a few easy steps. Say we have a NNUE with topography 512 x 32 x 32 x 1 and we want to make a better (we hope) NNUE with topography 1024 x 32 x 32 x 1 Step 1. Initialise our new 1024 neuron network with all weights = 0.0 Step 2. Unload all SF NNUE...

In today computer chess correspondence GM's are oveerated and so are their statements. There is no human that can beat SF13 with no opening book running at 16 core modern CPU at ICCF TC using whatever they have at their disposal. The very best they can do is to draw. So let’s run the next TCEC in a...