cdani wrote:I published a new version of Andscacs, 0.87. Is just a little stronger than the one playing at TCEC stage 2. It has too many changes and I did not want to wait any more. Also should be clearly stronger. An example is the current results on Graham "Super tournament".
Changes:
* Improved various evaluation parameters.
* Improvements on king safety.
* Improved imbalances and space evaluation.
* Improved passed pawns.
* Improved some endgames.
* Improved lmr.
* Speed improvements.
* Solved little bugs.
* Tries to avoid to give draw too easily.
* Improved SMP scaling.
* Improved time management.
* Improved move sorting.
I will explain on detail the changes when Andscacs is out of TCEC.
Very nice - here are some games against my latest endeavor, Mikhail 1.10 (unreleased). Should call this version 0.10 really.
I just found a bug that make it play bad at ltc. I think I have the solution. I will publish the new version soon. Don't use this 0.87 version. I will change the number. Sorry for the inconvenience.
Finally nothing was wrong, other than the test that showed something wrong
So you can use the downloaded 0.87 without problem.
Sorry for the inconveniences.
The previous version had an uninitialized variable that can evaluate randomly a parameter related to king safety. So sure was a few elo loss. Please, update your version!!
To be able to diferentiate between the versions, this one is named 0.871.
Kyodai wrote:
Thanks for the new version of Andscacs. I wonder if you have some
sort of estimation of the difference in elo between version 0.86 and 0.87?
Probably more than 20, but not sure as I have not done a serious tests against 0.86.
