Gaviota TBs

[Stein]

Hi,

The Gaviota TBs are available in different compression formats: https://github.com/michiguel/Gaviota-Ta ... master/gtb
There is .cp1, .cp2, .cp3 and .cp4.
https://github.com/michiguel/Gaviota-Ta ... ompression lists four compression methods.
Do you know which compression method is used for which of those .cps? Or e.g. .cp4?

https://python-chess.readthedocs.io/en/ ... viota.html shows that the .cp4 format can be read directly (see below).
Also that it is possible to search for a FEN there.
My questions are:
Is there a way to unpack on of the .cp formats?
I think that the unpacked format is .epd. Is that correct?
Is there a way to probe .epd or the unpacked format directly instead of FEN?

If someone has experience working with the Gaviotas, then it would be great to get in touch.

add_directory(directory: str) → None[source]
Adds .gtb.cp4 tables from a directory. The relevant files are lazily opened when the tablebase is actually probed.

Thank you and have a nice day!

[MOBMAT]

Now that every other EGTB format is going deeper, can Gaviota TBs be generate for 6-man? Or is it already done and I didn't notice.

[Nordlandia]

The developer said he may consider do it when time allowed it. 1-2 years ago afaik.

[syzygy]

Hi,

The Gaviota TBs are available in different compression formats: https://github.com/michiguel/Gaviota-Ta ... master/gtb
There is .cp1, .cp2, .cp3 and .cp4.
https://github.com/michiguel/Gaviota-Ta ... ompression lists four compression methods.
Do you know which compression method is used for which of those .cps? Or e.g. .cp4?

The answer is probably given by this table:
https://sites.google.com/site/gaviotach ... ne/schemes

I think that the unpacked format is .epd. Is that correct?

Not correct. It would be incredibly inefficient.
A tablebase file does not contain any FENs or other encodings of individual positions. Instead, when you probe a position you first need to map the position to a number N, where N runs from 0 to the number of positions in the table minus 1. The uncompressed table will now store the value of the position in the Nth byte.

[phhnguyen]

Now that every other EGTB format is going deeper, can Gaviota TBs be generate for 6-man? Or is it already done and I didn't notice.

I guess that 6-men should be about 1 TB, request a strong computer with at least 64 GB, run in 1-2 months. Doable but still expensive, download and store them are still big troubles.

[Dann Corbit]

The gaviota generator cannot make 6 man files unless it has been updated

[Nordlandia]

Nalimov 5-men 6.89 GB | Gaviota 5-men 6.54 GB | Difference = 5%

Total fire difference is going to be Nalimov 6-men 1.11TB | Gaviota 6-men ~ 1.0534 TB

Unless compression is modified, there is little gain. I'll consider 5% as trivial.

[hgm]

The gain is that you do not need a license.

[syzygy]

Now that every other EGTB format is going deeper, can Gaviota TBs be generate for 6-man? Or is it already done and I didn't notice.

I guess that 6-men should be about 1 TB, request a strong computer with at least 64 GB, run in 1-2 months. Doable but still expensive, download and store them are still big troubles.

The Gaviota TB scheme uses a somewhat complex indexing scheme. It is a lot of work to implement that scheme for the various possible combinations of 6 pieces (you need separate schemes for KQRvKQR, KQQvKQR, KQQvKQQ, KQQQvKQ, KQQQQ, repeat for pawnful tables).

I have written generators with similar schemes (with 5000+ lines of code for indexing 5-piece suicide tables) and in the end realised it is better to keep the indexing relatively simple (which then gives more room for trying out various permutations of the pieces and see which gives the smallest table).