Some of you were asking for pre-compiled binaries for Windows. Well, you can find them on https://www.patreon.com/gonzochess75 for becoming a patron to help support more robust testing of NPS on multi-gpu systems via Vast.ai.
Eduard wrote: ↑Wed Feb 20, 2019 5:49 pm
Thank you! I see the engine play at CCC. Which NN is used?
I see it uses plain NN ID 32930 from LC0...
from CCC: (emphasizing by me)
Allie USA
Version 0.1 (net 32930)
Rating 3100 https://github.com/manyoso/allie
Developed by:
Adam Treat
Allie is a new and original chess engine heavily inspired by the seminal AlphaZero paper and the Lc0 project.
It uses the Lc0 networks, but with custom built search, hash, move generation, threading, and time management.
In the future, Allie is expected to use A/B search instead of MCTS.
Created 2018
Is the engine available to download? If so, I can't see the link.
When AllieStein played on Div 4 it was not performing as good using the MCTS and without EGTB, but on the second part it switched to A/B search and it won all of its games except one single lost.
Chessqueen wrote: ↑Sat Mar 16, 2019 4:54 am
When AllieStein played on Div 4 it was not performing as good using the MCTS and without EGTB, but on the second part it switched to A/B search and it won all of its games except one single lost.
You mean the engine was changed during the running division?
I don't think so, maybe between Div 4's start from divided groups and its final battle of the upward- climbers of the groups?
Chessqueen wrote: ↑Sat Mar 16, 2019 4:54 am
When AllieStein played on Div 4 it was not performing as good using the MCTS and without EGTB, but on the second part it switched to A/B search and it won all of its games except one single lost.
You mean the engine was changed during the running division?
I don't think so, maybe between Div 4's start from divided groups and its final battle of the upward- climbers of the groups?
In TCEC chat it was confirmed by both the author and an admin that there was no update of Allie+Stein between beginning of TCEC S15 Div4b and end of Div4 playoffs. So the better playoff results might be explainable by small sample size/different opponents/different openings etc.
Sven Schüle (engine author: Jumbo, KnockOut, Surprise)
Sven wrote: ↑Sat Mar 16, 2019 12:09 pm
In TCEC chat it was confirmed by both the author and an admin that there was no update of Allie+Stein between beginning of TCEC S15 Div4b and end of Div4 playoffs. So the better playoff results might be explainable by small sample size/different opponents/different openings etc.
Allie does not have AlphaBeta search enabled. That will be in a future release. There was no update of Allie between div4b and div4 playoff. The results were also consistent between div4b and div4 when taking into account small sample size and head2head matches against AS and pira in div4b.