Of course the private Rybka FRC version still tops the best versions list, but since you can't buy this, it is just an indicator of what will inevitably come when Rybka 3.0 is released with FRC capaility http://www.computerchess.org.uk/ccrl/404FRC/
Still 200 games in progress (vs Movei 438 and Rybka). Shredder 11 seems to be doing well in both, so I'm certain the +90 ELO will hold, and maybe improve slightly.
Now I wonder what the Hiarcs team have in store for us with Hiarcs 12
Uri Blass wrote:Note that movei00_8_438 progress 10 10 10 performed better in tests in normal chess and I think that this personality may also perform better in FRC.
My guess is 30 elo improvement relative to the default version based on results that I saw but I did not do testing.
The advantage is 28 elo in the 40/4 normal list and it also performs better in the infinite loop that has
Uri Blass wrote:Note that movei00_8_438 progress 10 10 10 performed better in tests in normal chess and I think that this personality may also perform better in FRC.
My guess is 30 elo improvement relative to the default version based on results that I saw but I did not do testing.
The advantage is 28 elo in the 40/4 normal list and it also performs better in the infinite loop that has
Progress is one of the uci parameters of movei and it is possible to change it.
Uri
Was there a reason for not making it the default value?
Yes
I did not know that 10 10 10 are stronger.
I thought that path dependent evaluation is not going to work if
I use hash for pruning but it seems that I was wrong.
Note also that I did not do a lot of testing to get better values in the magic numbers and I guess that it is possible to get 50 elo improvement by tuning and testing.
Some values are simply 0 by default that mean that there is some knowledge that movei does not use and I did not test if using it is productive.
