Author: Michel Langeveld
Date: 22:31:02 11/17/99
Go up one level in this thread
On November 17, 1999 at 18:04:58, Dann Corbit wrote: >On November 17, 1999 at 16:18:24, Michel Langeveld wrote: >[snip] >Three very good ideas. >>1) >>An other idea, (I possibly already said), is to do all the positions of the >>crafty book(s). So all positions have CAP Data. I already programmed an crafty >>that's pugged all the unique positions during book creation to the LOG.xxx file. >>In this way I new look at non-cap-ed positions and do them.... >I am especially fond of this one. Can you create the following table for >positions without analysis: >[Frequency of hits] [EPD position] > >e.g. >600213, r3kbnr/ppp2ppp/2p1bq2/8/4P3/3B4/PPP2PPP/RNBQK2R w KQkq - >600001, r1b1k1nr/2ppqppp/pbp5/4N3/3PP3/2P5/PP3PPP/RNBQ1RK1 b kq - >etc? > >That way, we can process the most important rows first. It would be especially >good to build an enormous book using a ton of PGN, if you know what I mean. >That way, we will get the best stats from real usage. > >>2) >>It's also possible to look at the current EPD's <= 30 ply from the initial >>position and look if the move crafy recommends is an EPD positon that's already >>in the CAP data. If's it's not it has to be analyzed. In this way new >>theoretical lines are analyzed to an certain level. >Another dandy. Do you have code to accomplish this already? > >>3) >>Find large score drops and do all succ. positions and look at all the possible >>positons if there's a better move. >> >>For example if we have the folowing data. >> >>1. e4) 0.00 >> >>1. e4 e5 -3.00 >> >>Then analyse 1. e4 a6, 1.e4 a7 .... 1.e4 h6 1.e4 h7 >We would also need to look at the plies analyzed. What if one is ten plies and >the other is 19 plies? Good point, I don't have a very good solve for this but there 're a few options. If the succ. positions differ in plydepth then it's not a very big problem yet. If the root has a higher depth then (one of) the succ. position then we can try to enhance the plydepth of the succ. position by backtracking. For example Position A) 1.e4 0.00/19 (0.00 at 19 ply) Position B) 1. e4 e5 -3.00/10 (-3.00 at 10 ply) Then we can analyze: C1) 1.e4 e5 2.a3 (??/17) C2) 1.e4 e5 2.a4 (??/17) C3) 1.e4 e5 2.b3 (??/17) C4) 1.e4 e5 2.b4 (??/17) .. .. Cx) 1.e4 e5 2.h6 (??/17) Cx) 1.e4 e5 2.h7 (??/17) Although I think if both Position A & B are analysed at CAP time control it will be enough to analyse C1 - Cx at CAP time control. But infact I think such ply difference like Position A & B can't exist in practise (at least not often). Kind regards, Michel Langeveld
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.