World Draughts Forum

Rein, I assume that at least from a memory point of view the 9p would be doable-ish. My guestimate is that 256 GByte would be sufficient, and at least the Threadripper could address this. Also with Threadripper mother boards with 8 DIMM slots and 32 GByte Dimm seems not to be impossible. Last but n...

Rein, interesting posts. I tend to agree that in the end nn or nnue is a better abstraction for an evaluation function. However the current HW (although improving) imposes a too big nps penalty for nnue, which still favors the Scan-like patterns. Im however convinced that the balance will change in...

Ok, I did some more calculations. With a 64-core Threadripper (~$4K) and leading squares indexing, you can do 9 pieces in 90 days and 192 Gb of RAM. You can also do 10 pieces in 2 years and 640 Gb of RAM. All assuming that a threadripper core is pound for pound the equivalent of a Xeon core. If that...

Rein, I assume that at least from a memory point of view the 9p would be doable-ish. My guestimate is that 256 GByte would be sufficient, and at least the Threadripper could address this. Also with Threadripper mother boards with 8 DIMM slots and 32 GByte Dimm seems not to be impossible. Last but n...

I wonder over something like that.I just don't know anything about it.Is there any online service - rent a supercomputer to do it, for example, in three months. Of course: https://aws.amazon.com/ec2/pricing/on-demand/ For the peak memory load you want r5b.24xlarge with 96 cores and 768 Gb RAM which...

Krzysztof, the test from Ed, and also my test made clear that Scan is still better than Scan nnue. As we only start to digging into nnue, there is room for improvement. I personally don't believe we will surpass (with all other boundary conditions the same) Scan. But at which speed we will improve ...

The W-3245 CPU is 3.2 GHz, 16 cores. Yes, certainly it is stronger than many programs that Krzysztof runs in his tournaments. Of course a big part of that is that it has the scan search, which is very good, perhaps the best of all draughts programs. As to whether it can eventually do as well as pat...

IIRC, the 10-pc dbs are in the same ballpark as the chess 7-pc dbs. One key bottleneck would be generating the 5v5 kings sub db, which has 2.6 trillion (10^12) positions. That requires 640+Gb of RAM with 2 bits per position for a single side to move. For 6v4 kings you need at least 1 Tb of RAM. Tha...

IIRC, the 10-pc dbs are in the same ballpark as the chess 7-pc dbs. One key bottleneck would be generating the 5v5 kings sub db, which has 2.6 trillion (10^12) positions. That requires 640+Gb of RAM with 2 bits per position for a single side to move. For 6v4 kings you need at least 1 Tb of RAM. That...

They use pandas, this should work with pyTorch too. Currently I use numpy.fromfile to read binary files, this is as fast as reading binary files with C++. I prepare the binary data in C++ Python is way to slow for this. I'm thinking about using libTorch and doing everything in C++, the syntax is di...

In practice 1.4B positions are difficult to handle (one needs fast amounts of memory for this) so I used only 240M. I have gotten distracted with many other interesting machine learning projects (non-draughts) that kind of had to happen right now before anything else. But I will soon start working ...

[hit quote instead of edit by accident]

In practice 1.4B positions are difficult to handle (one needs fast amounts of memory for this) so I used only 240M. I have gotten distracted with many other interesting machine learning projects (non-draughts) that kind of had to happen right now before anything else. But I will soon start working ...

Guess Rein knows the details. Hardly! I think getting zero-knowledge self-play reinforcement learning (RL) correct is quite tricky. Obviously, it worked for AlphaZero / LeelaZero etc. But those were massive distributed efforts, with hundreds (even thousands) of CPU years of training. I haven't look...

I should also add that NNUE could have any number of fully connected layers on top. There is an chess engine called Halogen that uses an NNUE with only two fully connected layers on top, and another engine called Seer that uses a NNUE with four connected layers on top. Yes, the possible variations ...