What would make an engine a better analysis tool... ?

[as]

...when quality of the analysis is more important than the time used to produce it, even at the cost of placing lower in ranking lists at time controls?

i.e. persistent hash (and ability to save/manage it), less aggressive pruning, etc.

[Gian-Carlo Pascutto]

Less aggressive pruning certainly not. If being more aggressive makes the engine stronger, it also makes the analysis better, because, for example, it gets deeper in a fixed amount of time. Remember that even when analyzing time is finite!

However, the following things, while not directly affecting gameplay, do affect analysis:

- Score and variation (in)stability (*cough* Stockfish 1.7 *cough*)
- Proper support of MultiPV, searchmoves, etc
- Correct and complete PVs
- Persistent hash
- Good propagation of analysis when going back and forth in the game
- Information about moves other than the best (multipv, hash exploration, ...)

[Marek Soszynski]

Gian-Carlo, how long has it been since the last update of your own engine, Deep Sjeng WC2008?

Does it have "Proper support of MultiPV, searchmoves, etc."?

[metax]

Less aggressive pruning certainly not. If being more aggressive makes the engine stronger, it also makes the analysis better, because, for example, it gets deeper in a fixed amount of time. Remember that even when analyzing time is finite!

Also, when analyzing there is often more time than in games. Hence, tactical weaknesses (caused by aggressive pruning) take a back seat (because there is enough time anyway) while a low EBF is very beneficial for searching deep.

[Gian-Carlo Pascutto]

Less aggressive pruning certainly not. If being more aggressive makes the engine stronger, it also makes the analysis better, because, for example, it gets deeper in a fixed amount of time. Remember that even when analyzing time is finite!

Also, when analyzing there is often more time than in games. Hence, tactical weaknesses (caused by aggressive pruning) take a back seat (because there is enough time anyway) while a low EBF is very beneficial for searching deep.

I'm not sure a tactical trick missed at 25 ply is better than a tactical trick missed at 15 ply.

[Gian-Carlo Pascutto]

Gian-Carlo, how long has it been since the last update of your own engine, Deep Sjeng WC2008?

It was last updated in 2008.

Does it have "Proper support of MultiPV, searchmoves, etc."?

MultiPV yes. Searchmoves is in older versions but not in 3.0 and later, but will be back in 4.0 (thanks to Aquarium).

[Marek Soszynski]

Gian-Carlo, you mention Sjeng 4.0. When will that be released? Within a couple of months?

[Werewolf]

Gian-Carlo, you mention Sjeng 4.0. When will that be released? Within a couple of months?

I am also interested in this. I hope it comes with an opening book like Hiarcs & Shredder

[as]

So if one wanted to "stand in the shoulders of giants" by taking an open source engine to build a strong analysis tool (based on the above) ... which one would be the better pick?

[alpha123]

Gian-Carlo, you mention Sjeng 4.0. When will that be released? Within a couple of months?

I am also interested in this. I hope it comes with an opening book like Hiarcs & Shredder

I'm interested too.

An opening book would be cool, I guess, but not necessary. Clustering is cool enough .

Peter