zenpawn wrote:Ah, that explains how I could have missed it; I have WB 4.7.3 on one laptop and 4.8 on another. Thanks.
Well, part of that is my fault. I never released version 4.9 officially as WinBoard, because virtually all modifications in it compared to 4.8 were improvements of the XBoard front-end. (But the -epd option was an exception to that.) And at the time of 4.9.0 I was not entirely sure it was 100% stable. Officially the WinBoard-AA package contains a beta version, but it has some bug fixes compared to 4.9.0, while the only thing new in it is the possibility to use external piece bitmaps with alpha channel.
If there is anything about the EPD feature that you don't like, or think can be improved, please let me know.
zenpawn wrote:Ah, that explains how I could have missed it; I have WB 4.7.3 on one laptop and 4.8 on another. Thanks.
I generally run them through EPD2WB that comes with Gerbil.
I use GradualTest. For the STS (Strategic Test Suite), that gives partial score for moves that are decent (but not best), I wrote a small tool to parse GradualTest's output and determine Myrddin's final score.
Give partial scores? Probably not, because I don't even understand what you mean by that. WinBoard is not an engine, so how can't judge which moves are 'decent'? It only knows two classes of moves, legal and illegal. Seems to me you need an engine to decide what is 'decent'. It should of course be possible to let WinBoard use the second engine for that. Unless the EPD somehow lists what should be considered 'decent' through a field I am not aware of.
In a sense WinBoard currently considers every move that is not an 'avoid move' (am field) as 'decent', and awards a partial (= draw) score for it. Move from the bm field is a win, from the am field a loss.
hgm wrote:Give partial scores? Probably not, because I don't even understand what you mean by that. WinBoard is not an engine, so how can't judge which moves are 'decent'? It only knows two classes of moves, legal and illegal. Seems to me you need an engine to decide what is 'decent'. It should of course be possible to let WinBoard use the second engine for that. Unless the EPD somehow lists what should be considered 'decent' through a field I am not aware of.
In a sense WinBoard currently considers every move that is not an 'avoid move' (am field) as 'decent', and awards a partial (= draw) score for it. Move from the bm field is a win, from the am field a loss.
Sorry, I thought you might have been familiar with the test suite. All of the information is available in the EPD file. Here is an example:
1k2r3/1p1bP3/2p2p1Q/Ppb5/4Rp1P/2q2N1P/5PB1/6K1 b - - bm Kc7; id "STS(v11.0) King Activity.001"; c0 "Kc7=10, Kc8=4, Qa1+=4";
So the suite uses the "c0" comment field to state that Kc7 is awarded a full 10 points, and Kc8 and Qa1+ are awarded only 4 points.
OK, I see. So this is in fact an extension of the EPD standard, posing as a comment for backward compatibility.
I was not aware of this extension at all, so WinBoard doesn't do that. Doesn't sound very difficult to implement, though. WinBoard counts in the 'wins' counter of the match, and it would be easy enough to increment that by as much as the comment field specifies for the move.
hgm wrote:OK, I see. So this is in fact an extension of the EPD standard, posing as a comment for backward compatibility.
I was not aware of this extension at all, so WinBoard doesn't do that. Doesn't sound very difficult to implement, though. WinBoard counts in the 'wins' counter of the match, and it would be easy enough to increment that by as much as the comment field specifies for the move.
Excellent. I'd be happy to test if you like, but you can find the entire suite here:
