everyone knows about the cost of mpv, but it is a very nice (i think essential to any analysis engine) feature. most high level chess players i know crank mpv up quite a bit when looking at their games (i think depth & score is less relevant the quicker you can spot ideas behind moves); i'm not high level but still usually prefer 3-5 pvs, at least to begin with.Robert Houdart wrote:Anthony,
"If I have seen a little further it is by standing on the shoulders of Giants."
Multi-PV comes at a cost. If you want 3 principal variations, the engine may need to do up to 3 times the work to compute them. A secondary effect is that the hash table will be filled with positions for the second and third PV, most of which are useless for the real PV.
Basically it's like playing on a computer with less computational power and a smaller hash table.
Houdini uses a variable contempt value throughout the game (a bit more in the opening, less in the end game), but currently disables the contempt in analysis mode. I will add an ANALYSIS_CONTEMPT option so you can enable contempt in the analysis mode too.
You can expect real soon Houdini 1.02 which is basically a bug fix release, the multi-PV analysis will be added in Houdini 1.03.
Cheers,
Robert
btw, i still have my problems with you about what i feel is unnecessary disingenuousness from you on your sources -- i hope i am wrong & would like to see solid evidence on way or another -- but i will hand it to you that houdini is an excellent engine whatever the case. i will finish sometime shortly a ponder-on, FIDE TC, nunn suite match with houdini against R4, 1 cpu each, 1GB hash & will share the results here. i think you'll be pleased; i am certainly surprised so far.
(attributed to Fields Medalist Gerd Faltings, who is known for making such swashbuckling remarks, partially in jest).