Tuning the eval .. again :-)

[Daniel Anulliero]

Hi all
Well if I have , for exemple , 4 criterias to score a piece :
Bonus1 = +14 , bonus2 = +23 , bonus3 = +13 , penalty = -12
Now , I want to add a New bonus , bonus4 = +34
Can I Keep the first four criterias unchanged or they must be ajusted to the bonus4?
Here the idea is the bonus4 is the main bonus so the others become:
Bonus1 = (34/4) -14 = +6 , bonus2 = +15 , bonus2 = +5 , penalty1 = -12 unchanged .
Ok may be a silly question but some advices are welcome
Bests
Dany

[AlvaroBegue]

It depends on the correlation between the other features and the new one. If they are uncorrelated, the old values don't need to change. Otherwise, you would need to repeat the regression with the new feature added.

[Daniel Anulliero]

It depends on the correlation between the other features and the new one. If they are uncorrelated, the old values don't need to change. Otherwise, you would need to repeat the regression with the new feature added.

Ok so , for a concrete exemple : passed pawns , if I add a New bonus for a passer , I must adjust ALL the bonuses , sure .
Thank you Alvaro !

[Henk]

In my dream world tuning should be last step. Incomplete and buggy evaluation are similar.

In reality we are too stupid to implement a complete and correct evaluation. So we have to tune again and again after each failure.

There might be a dependency on search too. Hopeless. Keeping you busy forever. Independent/uncorrelated terms are essential.