Stockfish classical going downhill now

[Jouni]

Current SF classical vs latest version before NNUE:

ELO: -19.66 +-4.3 (95%) LOS: 0.0%
Total: 6440 W: 771 L: 1135 D: 4534
Ptnml(0-2): 84, 760, 1833, 522, 21

I am not surprised at all!

[Dicaste]

They should separate the development of classical stockfish like komodo did i think. NNUE is much more superior to classical stockfish i only tried once and i instantly hate it.

[Madeleine Birchfield]

They should separate the development of classical stockfish like komodo did i think. NNUE is much more superior to classical stockfish i only tried once and i instantly hate it.

AFAIK, the classical evaluation is still used when UseNNUE is set to true. Above a certain evaluation threshold, the classical evaluation is used instead of NNUE because it is faster to calculate than NNUE evaluation and there isn't much difference between the two when the evaluation is about +5 or +10.

[Dicaste]

Faster calculation doesn't mean everything.

[Pi4Chess]

Of course classical modifications are only tested now in situations where NNUE is not used (in default hybrid mode). So it will only become stronger where NNUE is less stronger than classical.

[Jouni]

But what's this test now measuring?

https://tests.stockfishchess.org/tests/ ... 9192d3958b

I don't get it. Classical vs "classical", what !?!?

[Alayan]

Of course classical modifications are only tested now in situations where NNUE is not used (in default hybrid mode). So it will only become stronger where NNUE is less stronger than classical.

This is wrong. Classical eval patches are tested with NNUE off.

[Pi4Chess]

Of course classical modifications are only tested now in situations where NNUE is not used (in default hybrid mode). So it will only become stronger where NNUE is less stronger than classical.

This is wrong. Classical eval patches are tested with NNUE off.

Ok then I don't understand why classical elo is going down as stated by OP.

[jhellis3]

What is happening is the superior NNUE eval allows for more aggressive search optimizations (narrower search) than what was/is optimal with only hand-coded eval.

Using classical eval still has the more aggressive pruning, which is not optimal for it.

[Pi4Chess]

What is happening is the superior NNUE eval allows for more aggressive search optimizations (narrower search) than what was/is optimal with only hand-coded eval.

Using classical eval still has the more aggressive pruning, which is not optimal for it.

Thanks for the explanation.