HCha/Stockfish
1
0
Fork 0
mirror of https://github.com/HChaZZY/Stockfish.git synced 2025-12-26 12:06:22 +08:00
Code Issues Packages Projects Releases Wiki Activity

Use plain nnue eval for validation loss calculation instead of first performing qsearch

Browse Source
This commit is contained in:
Tomasz Sobczyk
2020-12-21 20:35:23 +01:00
committed by nodchip
parent 6853b4aac2
commit 8ca82646a9
2 changed files with 2 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/instrumented_learn.sh
View File

@@ -124,7 +124,7 @@ cat << EOF > learn01.exp
send "uci\n"
send "setoption name SkipLoadingEval value true\n"
send "setoption name Use NNUE value true\n"
send "setoption name Use NNUE value pure\n"
send "setoption name Threads value $threads\n"
send "isready\n"
send "learn targetdir training_data epochs 1 sfen_read_size 100 thread_buffer_size 10 batchsize 100 use_draw_in_training 1 use_draw_in_validation 1 lr 1 eval_limit 32000 nn_batch_size 30 newbob_decay 0.5 eval_save_interval 30 loss_output_interval 10 validation_set_file_name validation_data/validation_data.bin\n"