bob wrote:Uri Blass wrote:bob wrote:Michel wrote:Very strong. Very buggy.
Very strange. After 330 games have not observed a single crash yet. Perhaps the crashes only happen at very fast time controls?
I have been experimenting with RobboLito since the source looks cleaner but people here say it is just a refactored ippolit.
I tried all types of time controls. But 330 games is a tiny number. You might not have hit the circumstances yet. I may try another run and report how many games end abnormally out of 8,000...
or maybe you used a different source
I understood that
RobboLito is not the same source as Ippolit
Uri
I assumed they were essentially the same. If not, that could explain the issue.
The new version out today has support for endgame tablebases, and compiling the code produces a second binary, RobboBuild, to build the tablebases. From the read-me:-
RobboTotalBase uses by design 64K blocks of RLE+BWT+MTF+Huffman-compressed DTC
data, with wins just having one bit. This compares to other schemes in the end.
Full Size RobboSize Index Total PerCent
3s: 410382 37176+ 272 37448 9.1%
4s: 157440864 15885237+ 10780 15935813 10.1%
5s: 37429189056 3659510583+2289088 3661799671 9.8% (3.41GB)
The Nalimov takes 7.2GB for 5s, but they are DTM. For DTC, they are still
4.5GB, largely for the RobboEdge in its one-sided nature. The FEG are 5.6GB
with DTM, and should compare the RobboScheme if outfitted with DTC.