Rybka 1.1 FAQ

March 16, 2006

 

I am using one of the Rybka Beta versions. Should I upgrade to Rybka 1.1?

 

Absolutely – this is a very good time to take a moment to upgrade.

 

What is improved since the latest Rybka Beta version (Beta 13d)?

 

1) Better handling of transitions from the middlegame to the endgame

2) New heuristics for endgame play

3) More middlegame evaluation knowledge

4) More efficient search

5) Bug fixes

 

What didn’t make it into the Rybka 1.1 release?

 

The top-priority remaining items are:

 

1) Full UCI & UCI2 compliance

2) Better support for human-driven analysis

3) Sensible configurable engine parameters

4) More chess knowledge

5) More efficient search

6) Multi-processor support

 

Can you elaborate on what you mean by „support for human-driven analysis”?

 

This means two things.

 

In the first place, it means some fixes to the cosmetic output of the engine. For example:

 

1) Displaying the full main variation (currently, it is truncated in some cases)

2) Displaying the variation all the way to mate if a mate score is reported

3) Continuing to search even after finding a mate, to look for a shorter mate

 

More importantly, it means searching in a way which is more useful for a human analyst, and remembering this analysis in a seamless way. This topic is thus far completely unexplored in the world of computer chess, and will involve some more fundamental changes to the search. For a bit more technical explanation, please see here:

 

http://www.rybkachess.com/free/AnalysisModeComments.rtf

 

How many more releases will be part of the Rybka 1 series?

 

One or two. There will definitely be an end-of-April release which will conclude the series. There may also be an intermediate early-April release, depending on the engine status at that point.

 

What from the above to-do list is guaranteed to be included in Rybka 1?

 

Full UCI & UCI2 compliance has been promised and will definitely be included. The other items are planned but not promised.

 

What about underpromotions?

 

Included in Rybka 1.1. However, underpromotions to bishops are always ignored – you can think of it as a sort of pruning. So, studies involving underpromotions to bishops will join studies involving middlegame zugzwang as unsolvable for Rybka.

 

How good is the new endgame knowledge?

 

It’s a good start, but certainly a lot more work is needed. Humans are naturally very good at understanding how to transition their middlegame advantage into the endgame - a human will have a very good grasp of where the drawing chances exist and how to avoid or seek them. However, it is very difficult to formulate these things and teach them to an engine. At the very least, Rybka is better now than the other engines in this department.

 

Some of the configurable parameters from the Beta versions have disappeared? What happened to them?

 

Some of them (ie. search direction, focus and soundness) were removed because the changes to the Rybka 1.1 search made the previous settings meaningless. For example, Rybka now searches with a focus much higher than the previously highest value of 6, but the search decisions are covered by a number of new heuristics which make this feasible.

 

The optimism parameters have been removed because they are complex and I have seen enough data to be convinced that tinkering with them is not a path to making Rybka play better.

 

Why has UCI & UCI2 compliance slipped so much in the schedule?

 

Search and evaluation changes require a lot of time to think through, implement, test and validate. I try to keep those moving forward at all times, and fit the other stuff in between.

 

Can you release Rybka versions more often? Surely you could have released an improved version in late February.

 

In theory, a new engine build could be made every day. Some software projects do work like this. There is a tradeoff between keeping customers up-to-date, and having enough of a „wow” effect around engine releases.

 

Is Rybka 2 going to be released using the same continual release model as Rybka 1?

 

I am very much leaning that way. My customer survey (http://www.rybkachess.com/survey/) shows a huge support for it. The approximate date range for Rybka 2 would be Aug 1 to Dec 31, 2006, but please don’t quote me on this.