About Xiangqi NNUE Structure #491

PikaCat-OuO · 2022-06-01T12:25:04Z

PikaCat-OuO
Jun 1, 2022

I'm a novice in NNUE and want to make some exploration of Xiangqi NNUE Structure. I've done reading some code of the NNUE part of the fairy-stockfish, but there are some points that confuse me. I would be more than grateful if someone could help me answer these questions. ヽ(✿ﾟ▽ﾟ)ノ

In the current XiangQi NNUE Structure, the index of the piece is WHITE ROOK = 0, BLACK ROOK = 1, WHITE FERS = 2, BLACK FERS = 3, WHITE CANNON = 4, BLACK CANNON = 5, WHITE SOLDIER = 6, BLACK SOLDIER = 7, WHITE HORSE = 8, BLACK HORSE = 9, WHITE ELEPHANT = 10, BLACK ELEPHANT = 11, WHITE KING = BLACK KING = 12 is that correct?
If I print the kingSquareIndex[i] / nnuePieceIndices for i in range(0, 90), does it looks like this?

0  0  0  6  7  8  0  0  0
0  0  0  3  4  5  0  0  0
0  0  0  0  1  2  0  0  0
0  0  0  0  0  0  0  0  0
0  0  0  0  0  0  0  0  0
0  0  0  0  0  0  0  0  0
0  0  0  0  0  0  0  0  0
0  0  0  0  1  2  0  0  0
0  0  0  3  4  5  0  0  0
0  0  0  6  7  8  0  0  0

Answered by ianfab

Jun 2, 2022

Regarding the piece index, I don't know the order by heart, but it just follows the order of piece types in the enumeration at https://github.com/ianfab/Fairy-Stockfish/blob/088d4da202fec1792c336671813a2521c600f28b/src/types.h#L390-L396 So what you wrote looks correct.
For the king square index you need to consider that the king square first always is oriented according to the side to move, and the opposing palace isn't really relevant as only the own king's index is used. The indices are enumerated starting from a1 (i.e., the left bottom), i.e., d1 = 0, e1 = 1, ....

View full answer

ianfab · 2022-06-02T11:52:19Z

ianfab
Jun 2, 2022
Maintainer

Regarding the piece index, I don't know the order by heart, but it just follows the order of piece types in the enumeration at https://github.com/ianfab/Fairy-Stockfish/blob/088d4da202fec1792c336671813a2521c600f28b/src/types.h#L390-L396 So what you wrote looks correct.
For the king square index you need to consider that the king square first always is oriented according to the side to move, and the opposing palace isn't really relevant as only the own king's index is used. The indices are enumerated starting from a1 (i.e., the left bottom), i.e., d1 = 0, e1 = 1, ....

13 replies

ianfab Jun 2, 2022
Maintainer

Thanks for the info. I didn't mean to request anything, I was just curious, because I of course always strive to improve my engine. Sorry if this came across wrongly.

PikaCat-OuO Jun 2, 2022
Author

(●'◡'●)，That's totally fine, No problem at all. I'll always be there to help you and the open-source community if I can. Once I tested the network strength and proved that it is stronger than xiangqi-6f64c55fcb28.nnue . I'll upload it to https://forms.gle/8Am9LTqXQJo43ps79 immediately.
It's me that is the one who is going to be grateful to you for you providing so many valuable tools and answers to me.

PikaCat-OuO Jun 2, 2022
Author

By the way, according to the engine tournament test result 3 days ago, the release version of https://github.com/ianfab/Fairy-Stockfish-NNUE Release for variant xiangqi with built-in NNUE network xiangqi-6f64c55fcb28.nnue fairy-stockfish_x86-64-bmi2.exe is stronger than https://github.com/leedavid/NewGG in 5 seconds per move, and with a high probability of winning the game on my machine.

Belzedar94 Jun 2, 2022

It's open-source under WTFPL on: https://github.com/PikaCat-OuO/ChineseChess . It has a GUI attached to it written by my own but the engine is embedded in the GUI and the support of the UCI Protocol is not implemented yet.

Guess what, the first reaction that I see this repository is to make some contribution and improve the best nnue file now. It was about 3 days ago, I've used your data generator to generate 1000000 training data using xiangqi-6f64c55fcb28.nnue(the latest network) with setoption name EvalFile value xiangqi-6f64c55fcb28.nnue. Then I cloned your trainer and build it in Google Colab and run it with
python serialize.py --features='HalfKAv2' xiangqi-6f64c55fcb28.nnue startingpointfortraining.pt
python train.py --resume-from-model startingpointfortraining.pt ...
for about 30 epochs and the loss seems to go down further. I've not tested it yet using https://github.com/ianfab/fairyfishtest. Because I'm busy writing c++ code right now. I decided to do the test once I finish the code writing. If you want the nnue file right now, I'll send it to you right now XD.

Do you say you have trained a net on 1M positions? For reference, to improve the latest net by 15-20 elo, I used 1B (1.000.000.000) positions (depth 9). Anyways, despite the result, it's a good first contact.

I hope you can make some progress and achieve the desired results with your engine :)

PikaCat-OuO Jun 2, 2022
Author

It's open-source under WTFPL on: https://github.com/PikaCat-OuO/ChineseChess . It has a GUI attached to it written by my own but the engine is embedded in the GUI and the support of the UCI Protocol is not implemented yet.
Guess what, the first reaction that I see this repository is to make some contribution and improve the best nnue file now. It was about 3 days ago, I've used your data generator to generate 1000000 training data using xiangqi-6f64c55fcb28.nnue(the latest network) with setoption name EvalFile value xiangqi-6f64c55fcb28.nnue. Then I cloned your trainer and build it in Google Colab and run it with
python serialize.py --features='HalfKAv2' xiangqi-6f64c55fcb28.nnue startingpointfortraining.pt
python train.py --resume-from-model startingpointfortraining.pt ...
for about 30 epochs and the loss seems to go down further. I've not tested it yet using https://github.com/ianfab/fairyfishtest. Because I'm busy writing c++ code right now. I decided to do the test once I finish the code writing. If you want the nnue file right now, I'll send it to you right now XD.
Do you say you have trained a net on 1M positions? For reference, to improve the latest net by 15-20 elo, I used 1B (1.000.000.000) positions (depth 9). Anyways, despite the result, it's a good first contact.

I hope you can make some progress and achieve the desired results with your engine :)

The training data is generated using your latest net, and training is continued from your latest net. I think it's kinda like some bootstrapping thing. So either I make mistakes in the training process or there is some magic in it.
Here is the output of the training code:

Feature set: HalfKAv2^
Num real features: 10530
Num virtual features: 1260
Num features: 11790
Training with xiangqi.bin validating with xiangqi.bin
Global seed set to 42
Seed 42
Using batch size 16384
Smart fen skipping: True
Random fen skipping: 3
limiting torch to 1 threads.
Using log dir logs/
ModelCheckpoint(save_last=True, save_top_k=-1, monitor=None) will duplicate the last checkpoint saved.
GPU available: True, used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
/usr/local/lib/python3.7/dist-packages/pytorch_lightning/trainer/trainer.py:2086: LightningDeprecationWarning: `Trainer.root_gpu` is deprecated in v1.6 and will be removed in v1.8. Please use `Trainer.strategy.root_device.index` instead.
  "`Trainer.root_gpu` is deprecated in v1.6 and will be removed in v1.8. "
Using c++ data loader
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Ranger optimizer loaded. 
Gradient Centralization usage = False

  | Name         | Type                          | Params
---------------------------------------------------------------
0 | input        | DoubleFeatureTransformerSlice | 6.1 M 
1 | layer_stacks | LayerStacks                   | 152 K 
---------------------------------------------------------------
6.3 M     Trainable params
0         Non-trainable params
6.3 M     Total params
25.134    Total estimated model params size (MB)
Epoch 0:  95% 1220/1283 [01:45<00:05, 11.51it/s, loss=0.00173, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 0:  97% 1240/1283 [01:47<00:03, 11.53it/s, loss=0.00173, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.15it/s]
Epoch 0:  98% 1260/1283 [01:49<00:01, 11.54it/s, loss=0.00173, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.63it/s]
Epoch 0: 100% 1280/1283 [01:50<00:00, 11.56it/s, loss=0.00173, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.65it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.65it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.63it/s]
Epoch 0: 100% 1283/1283 [01:51<00:00, 11.56it/s, loss=0.00172, v_num=1]
Epoch 1:  95% 1220/1283 [01:42<00:05, 11.88it/s, loss=0.00158, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 1:  97% 1240/1283 [01:44<00:03, 11.89it/s, loss=0.00158, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.29it/s]
Epoch 1:  98% 1260/1283 [01:45<00:01, 11.90it/s, loss=0.00158, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.91it/s]
Epoch 1: 100% 1280/1283 [01:47<00:00, 11.92it/s, loss=0.00158, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.76it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.76it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.78it/s]
Epoch 1: 100% 1283/1283 [01:47<00:00, 11.92it/s, loss=0.00158, v_num=1]
Epoch 2:  95% 1220/1283 [01:42<00:05, 11.93it/s, loss=0.00148, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 2:  97% 1240/1283 [01:43<00:03, 11.93it/s, loss=0.00148, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 11.28it/s]
Epoch 2:  98% 1260/1283 [01:45<00:01, 11.92it/s, loss=0.00148, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 11.85it/s]
Epoch 2: 100% 1280/1283 [01:47<00:00, 11.93it/s, loss=0.00148, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.18it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.18it/s]
Validation DataLoader 0: 100% 62/62 [00:05<00:00, 12.20it/s]
Epoch 2: 100% 1283/1283 [01:47<00:00, 11.93it/s, loss=0.00147, v_num=1]
Epoch 3:  95% 1220/1283 [01:41<00:05, 11.97it/s, loss=0.00141, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 3:  97% 1240/1283 [01:43<00:03, 11.98it/s, loss=0.00141, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.16it/s]
Epoch 3:  98% 1260/1283 [01:45<00:01, 11.99it/s, loss=0.00141, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.79it/s]
Epoch 3: 100% 1283/1283 [01:46<00:00, 12.00it/s, loss=0.00141, v_num=1]
Epoch 4:  95% 1220/1283 [01:41<00:05, 12.01it/s, loss=0.00135, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 4:  97% 1240/1283 [01:43<00:03, 12.03it/s, loss=0.00135, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.40it/s]
Epoch 4:  98% 1260/1283 [01:44<00:01, 12.04it/s, loss=0.00135, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.99it/s]
Epoch 4: 100% 1280/1283 [01:46<00:00, 12.05it/s, loss=0.00135, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.79it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.79it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.78it/s]
Epoch 4: 100% 1283/1283 [01:46<00:00, 12.05it/s, loss=0.00135, v_num=1]
Epoch 5:  95% 1220/1283 [01:41<00:05, 12.02it/s, loss=0.00129, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 5:  97% 1240/1283 [01:42<00:03, 12.04it/s, loss=0.00129, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.35it/s]
Epoch 5:  98% 1260/1283 [01:44<00:01, 12.05it/s, loss=0.00129, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.97it/s]
Epoch 5: 100% 1280/1283 [01:46<00:00, 12.06it/s, loss=0.00129, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.89it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.89it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.89it/s]
Epoch 5: 100% 1283/1283 [01:46<00:00, 12.06it/s, loss=0.00129, v_num=1]
Epoch 6:  95% 1220/1283 [01:41<00:05, 12.01it/s, loss=0.00125, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 6:  97% 1240/1283 [01:43<00:03, 12.03it/s, loss=0.00125, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.44it/s]
Epoch 6:  98% 1260/1283 [01:44<00:01, 12.04it/s, loss=0.00125, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.98it/s]
Epoch 6: 100% 1280/1283 [01:46<00:00, 12.06it/s, loss=0.00125, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.98it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.98it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 13.00it/s]
Epoch 6: 100% 1283/1283 [01:46<00:00, 12.06it/s, loss=0.00126, v_num=1]
Epoch 7:  95% 1220/1283 [01:40<00:05, 12.11it/s, loss=0.00123, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 7:  97% 1240/1283 [01:42<00:03, 12.13it/s, loss=0.00123, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.30it/s]
Epoch 7:  98% 1260/1283 [01:43<00:01, 12.13it/s, loss=0.00123, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.95it/s]
Epoch 7: 100% 1280/1283 [01:45<00:00, 12.14it/s, loss=0.00123, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.83it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.83it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.83it/s]
Epoch 7: 100% 1283/1283 [01:45<00:00, 12.14it/s, loss=0.00123, v_num=1]
Epoch 8:  95% 1220/1283 [01:40<00:05, 12.10it/s, loss=0.0012, v_num=1] 
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 8:  97% 1240/1283 [01:42<00:03, 12.09it/s, loss=0.0012, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 11.23it/s]
Epoch 8:  98% 1260/1283 [01:44<00:01, 12.10it/s, loss=0.0012, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.09it/s]
Epoch 8: 100% 1280/1283 [01:45<00:00, 12.11it/s, loss=0.0012, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.41it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.41it/s]
Validation DataLoader 0: 100% 62/62 [00:05<00:00, 12.45it/s]
Epoch 8: 100% 1283/1283 [01:45<00:00, 12.11it/s, loss=0.0012, v_num=1]
Epoch 9:  95% 1220/1283 [01:41<00:05, 12.06it/s, loss=0.00119, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 9:  97% 1240/1283 [01:42<00:03, 12.08it/s, loss=0.00119, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.51it/s]
Epoch 9:  98% 1260/1283 [01:44<00:01, 12.09it/s, loss=0.00119, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.10it/s]
Epoch 9: 100% 1280/1283 [01:45<00:00, 12.10it/s, loss=0.00119, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.79it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.79it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.78it/s]
Epoch 9: 100% 1283/1283 [01:46<00:00, 12.10it/s, loss=0.00119, v_num=1]
Epoch 10:  95% 1220/1283 [01:41<00:05, 12.01it/s, loss=0.00116, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 10:  97% 1240/1283 [01:43<00:03, 12.03it/s, loss=0.00116, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.22it/s]
Epoch 10:  98% 1260/1283 [01:44<00:01, 12.04it/s, loss=0.00116, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.93it/s]
Epoch 10: 100% 1280/1283 [01:46<00:00, 12.05it/s, loss=0.00116, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.76it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.76it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.74it/s]
Epoch 10: 100% 1283/1283 [01:46<00:00, 12.04it/s, loss=0.00116, v_num=1]
Epoch 11:  95% 1220/1283 [01:41<00:05, 12.03it/s, loss=0.00113, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 11:  97% 1240/1283 [01:42<00:03, 12.05it/s, loss=0.00113, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.59it/s]
Epoch 11:  98% 1260/1283 [01:44<00:01, 12.06it/s, loss=0.00113, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.04it/s]
Epoch 11: 100% 1280/1283 [01:46<00:00, 12.07it/s, loss=0.00113, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.92it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.92it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.91it/s]
Epoch 11: 100% 1283/1283 [01:46<00:00, 12.07it/s, loss=0.00113, v_num=1]
Epoch 12:  95% 1220/1283 [01:41<00:05, 12.03it/s, loss=0.00111, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 12:  97% 1240/1283 [01:42<00:03, 12.05it/s, loss=0.00111, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.39it/s]
Epoch 12:  98% 1260/1283 [01:44<00:01, 12.06it/s, loss=0.00111, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.75it/s]
Epoch 12: 100% 1280/1283 [01:46<00:00, 12.07it/s, loss=0.00111, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.82it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.82it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.81it/s]
Epoch 12: 100% 1283/1283 [01:46<00:00, 12.07it/s, loss=0.00111, v_num=1]
Epoch 13:  95% 1220/1283 [01:41<00:05, 12.02it/s, loss=0.00111, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 13:  97% 1240/1283 [01:42<00:03, 12.04it/s, loss=0.00111, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.53it/s]
Epoch 13:  98% 1260/1283 [01:44<00:01, 12.05it/s, loss=0.00111, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.03it/s]
Epoch 13: 100% 1280/1283 [01:46<00:00, 12.06it/s, loss=0.00111, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.93it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.93it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.93it/s]
Epoch 13: 100% 1283/1283 [01:46<00:00, 12.06it/s, loss=0.00111, v_num=1]
Epoch 14:  95% 1220/1283 [01:40<00:05, 12.09it/s, loss=0.00109, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 14:  97% 1240/1283 [01:42<00:03, 12.11it/s, loss=0.00109, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.53it/s]
Epoch 14:  98% 1260/1283 [01:44<00:01, 12.11it/s, loss=0.00109, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.93it/s]
Epoch 14: 100% 1280/1283 [01:45<00:00, 12.12it/s, loss=0.00109, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.77it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.77it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.76it/s]
Epoch 14: 100% 1283/1283 [01:45<00:00, 12.12it/s, loss=0.00108, v_num=1]
Epoch 15:  95% 1220/1283 [01:41<00:05, 12.05it/s, loss=0.00108, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 15:  97% 1240/1283 [01:42<00:03, 12.07it/s, loss=0.00108, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.46it/s]
Epoch 15:  98% 1260/1283 [01:44<00:01, 12.08it/s, loss=0.00108, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.12it/s]
Epoch 15: 100% 1280/1283 [01:45<00:00, 12.09it/s, loss=0.00108, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.00it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.00it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 13.01it/s]
Epoch 15: 100% 1283/1283 [01:46<00:00, 12.09it/s, loss=0.00107, v_num=1]
Epoch 16:  95% 1220/1283 [01:40<00:05, 12.16it/s, loss=0.00107, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 16:  97% 1240/1283 [01:41<00:03, 12.18it/s, loss=0.00107, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.36it/s]
Epoch 16:  98% 1260/1283 [01:43<00:01, 12.19it/s, loss=0.00107, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.05it/s]
Epoch 16: 100% 1280/1283 [01:44<00:00, 12.20it/s, loss=0.00107, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.93it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.93it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.94it/s]
Epoch 16: 100% 1283/1283 [01:45<00:00, 12.20it/s, loss=0.00106, v_num=1]
Epoch 17:  95% 1220/1283 [01:40<00:05, 12.19it/s, loss=0.00106, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 17:  97% 1240/1283 [01:41<00:03, 12.21it/s, loss=0.00106, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.56it/s]
Epoch 17:  98% 1260/1283 [01:43<00:01, 12.22it/s, loss=0.00106, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.09it/s]
Epoch 17: 100% 1280/1283 [01:44<00:00, 12.23it/s, loss=0.00106, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.91it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.91it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.49it/s]
Epoch 17: 100% 1283/1283 [01:45<00:00, 12.22it/s, loss=0.00106, v_num=1]
Epoch 18:  95% 1220/1283 [01:40<00:05, 12.13it/s, loss=0.00103, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 18:  97% 1240/1283 [01:42<00:03, 12.16it/s, loss=0.00103, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.69it/s]
Epoch 18:  98% 1260/1283 [01:43<00:01, 12.17it/s, loss=0.00103, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.20it/s]
Epoch 18: 100% 1280/1283 [01:45<00:00, 12.18it/s, loss=0.00103, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.07it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.07it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 13.06it/s]
Epoch 18: 100% 1283/1283 [01:45<00:00, 12.18it/s, loss=0.00103, v_num=1]
Epoch 19:  95% 1220/1283 [01:40<00:05, 12.17it/s, loss=0.00103, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 19:  97% 1240/1283 [01:41<00:03, 12.19it/s, loss=0.00103, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.37it/s]
Epoch 19:  98% 1260/1283 [01:43<00:01, 12.20it/s, loss=0.00103, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.00it/s]
Epoch 19: 100% 1280/1283 [01:44<00:00, 12.20it/s, loss=0.00103, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.74it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.74it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.76it/s]
Epoch 19: 100% 1283/1283 [01:45<00:00, 12.20it/s, loss=0.00103, v_num=1]
Epoch 20:  95% 1220/1283 [01:40<00:05, 12.12it/s, loss=0.00102, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 20:  97% 1240/1283 [01:42<00:03, 12.13it/s, loss=0.00102, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.40it/s]
Epoch 20:  98% 1260/1283 [01:43<00:01, 12.14it/s, loss=0.00102, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.05it/s]
Epoch 20: 100% 1280/1283 [01:45<00:00, 12.16it/s, loss=0.00102, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.01it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.01it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.97it/s]
Epoch 20: 100% 1283/1283 [01:45<00:00, 12.16it/s, loss=0.00102, v_num=1]
Epoch 21:  95% 1220/1283 [01:40<00:05, 12.18it/s, loss=0.00101, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 21:  97% 1240/1283 [01:41<00:03, 12.20it/s, loss=0.00101, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.57it/s]
Epoch 21:  98% 1260/1283 [01:43<00:01, 12.21it/s, loss=0.00101, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.15it/s]
Epoch 21: 100% 1280/1283 [01:44<00:00, 12.22it/s, loss=0.00101, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.07it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.07it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 13.08it/s]
Epoch 21: 100% 1283/1283 [01:44<00:00, 12.22it/s, loss=0.00101, v_num=1]
Epoch 22:  95% 1220/1283 [01:40<00:05, 12.13it/s, loss=0.00101, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 22:  97% 1240/1283 [01:42<00:03, 12.15it/s, loss=0.00101, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.19it/s]
Epoch 22:  98% 1260/1283 [01:43<00:01, 12.16it/s, loss=0.00101, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.08it/s]
Epoch 22: 100% 1280/1283 [01:45<00:00, 12.17it/s, loss=0.00101, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.98it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.98it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.97it/s]
Epoch 22: 100% 1283/1283 [01:45<00:00, 12.17it/s, loss=0.00101, v_num=1]
Epoch 23:  95% 1220/1283 [01:40<00:05, 12.16it/s, loss=0.000986, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 23:  97% 1240/1283 [01:41<00:03, 12.18it/s, loss=0.000986, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.68it/s]
Epoch 23:  98% 1260/1283 [01:43<00:01, 12.20it/s, loss=0.000986, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:02<00:01, 13.30it/s]
Epoch 23: 100% 1280/1283 [01:44<00:00, 12.20it/s, loss=0.000986, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.07it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.07it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 13.05it/s]
Epoch 23: 100% 1283/1283 [01:45<00:00, 12.20it/s, loss=0.000982, v_num=1]
Epoch 24:  95% 1220/1283 [01:40<00:05, 12.14it/s, loss=0.000995, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 24:  97% 1240/1283 [01:41<00:03, 12.16it/s, loss=0.000995, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.33it/s]
Epoch 24:  98% 1260/1283 [01:43<00:01, 12.16it/s, loss=0.000995, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.91it/s]
Epoch 24: 100% 1280/1283 [01:45<00:00, 12.17it/s, loss=0.000995, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.92it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.92it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.93it/s]
Epoch 24: 100% 1283/1283 [01:45<00:00, 12.17it/s, loss=0.000994, v_num=1]
Epoch 25:  95% 1220/1283 [01:40<00:05, 12.10it/s, loss=0.000976, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 25:  97% 1240/1283 [01:42<00:03, 12.12it/s, loss=0.000976, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.57it/s]
Epoch 25:  98% 1260/1283 [01:43<00:01, 12.13it/s, loss=0.000976, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.08it/s]
Epoch 25: 100% 1280/1283 [01:45<00:00, 12.15it/s, loss=0.000976, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.00it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.00it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 13.01it/s]
Epoch 25: 100% 1283/1283 [01:45<00:00, 12.15it/s, loss=0.000977, v_num=1]
Epoch 26:  95% 1220/1283 [01:40<00:05, 12.17it/s, loss=0.00096, v_num=1] 
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 26:  97% 1240/1283 [01:41<00:03, 12.19it/s, loss=0.00096, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.55it/s]
Epoch 26:  98% 1260/1283 [01:43<00:01, 12.20it/s, loss=0.00096, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.13it/s]
Epoch 26: 100% 1280/1283 [01:44<00:00, 12.21it/s, loss=0.00096, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.07it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.07it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 13.02it/s]
Epoch 26: 100% 1283/1283 [01:45<00:00, 12.21it/s, loss=0.000961, v_num=1]
Epoch 27:  95% 1220/1283 [01:40<00:05, 12.09it/s, loss=0.000961, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 27:  97% 1240/1283 [01:42<00:03, 12.12it/s, loss=0.000961, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:02, 14.17it/s]
Epoch 27:  98% 1260/1283 [01:43<00:01, 12.12it/s, loss=0.000961, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.18it/s]
Epoch 27: 100% 1280/1283 [01:45<00:00, 12.13it/s, loss=0.000961, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.02it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.02it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 13.03it/s]
Epoch 27: 100% 1283/1283 [01:45<00:00, 12.13it/s, loss=0.000964, v_num=1]
Epoch 28:  95% 1220/1283 [01:40<00:05, 12.17it/s, loss=0.000958, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 28:  97% 1240/1283 [01:41<00:03, 12.18it/s, loss=0.000958, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.14it/s]
Epoch 28:  98% 1260/1283 [01:43<00:01, 12.19it/s, loss=0.000958, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 12.76it/s]
Epoch 28: 100% 1280/1283 [01:44<00:00, 12.19it/s, loss=0.000958, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.70it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 12.70it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 12.73it/s]
Epoch 28: 100% 1283/1283 [01:45<00:00, 12.19it/s, loss=0.000961, v_num=1]
Epoch 29:  95% 1220/1283 [01:39<00:05, 12.29it/s, loss=0.000955, v_num=1]
Validation: 0it [00:00, ?it/s]
Validation:   0% 0/62 [00:00<?, ?it/s]
Epoch 29:  97% 1240/1283 [01:40<00:03, 12.31it/s, loss=0.000955, v_num=1]
Validation DataLoader 0:  32% 20/62 [00:01<00:03, 13.67it/s]
Epoch 29:  98% 1260/1283 [01:42<00:01, 12.32it/s, loss=0.000955, v_num=1]
Validation DataLoader 0:  65% 40/62 [00:03<00:01, 13.17it/s]
Epoch 29: 100% 1280/1283 [01:43<00:00, 12.33it/s, loss=0.000955, v_num=1]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.05it/s]
Validation DataLoader 0:  97% 60/62 [00:04<00:00, 13.05it/s]
Validation DataLoader 0: 100% 62/62 [00:04<00:00, 13.06it/s]
Epoch 29: 100% 1283/1283 [01:44<00:00, 12.33it/s, loss=0.000954, v_num=1]
Epoch 29: 100% 1283/1283 [01:45<00:00, 12.21it/s, loss=0.000954, v_num=1]

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

About Xiangqi NNUE Structure #491

{{title}}

{{editor}}'s edit

{{editor}}'s edit

Replies: 1 comment 13 replies

{{title}}

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

Select a reply

About Xiangqi NNUE Structure #491

PikaCat-OuO Jun 1, 2022

Replies: 1 comment · 13 replies

ianfab Jun 2, 2022 Maintainer

ianfab Jun 2, 2022 Maintainer

PikaCat-OuO Jun 2, 2022 Author

PikaCat-OuO Jun 2, 2022 Author

Belzedar94 Jun 2, 2022

PikaCat-OuO Jun 2, 2022 Author

PikaCat-OuO
Jun 1, 2022

Replies: 1 comment 13 replies

ianfab
Jun 2, 2022
Maintainer

ianfab Jun 2, 2022
Maintainer

PikaCat-OuO Jun 2, 2022
Author

PikaCat-OuO Jun 2, 2022
Author

PikaCat-OuO Jun 2, 2022
Author