These results make me think that I should be using STS to help tune my evaluation function directly. I hate trying to get the constants in my evaluation tuned correctly (how much of a bonus for a centralized knight on an outpost square, etc), because there are so many degrees of freedom and testing with a statistically significant number of games takes a long time for each idea you have.swami wrote: I'm flattered to know that improvement of knight handling in this version is clearly reflected on STS scores (Knight outposts/ Bishop vs Knight)
Yes, It indeed was a great improvement, Aaron. The engine is already playing at Division 4 level which averages around 2600 elo! (since it scored the total of 491 points)
For info:
http://www.talkchess.com/forum/viewtopi ... 7&start=10
I think I should use STS to help identify the promising ideas and filter out the ones that definitely aren't helpful, then only do full testing on the ideas that look good in STS. That would make my turnaround time a lot faster and let me test more ideas. So thanks for your hard work on this suite; I think it will make my testing faster and more productive for 1.7.
