Symmetry corrections (was: RE: [gnugo-devel] code maintenance pat ch)

[Portela Fernand]

Gunnar wrote:

> There's one plausible and probably also likely explanation. When
> removing the redundant symmetry for these patterns the matching order
> of patterns may change which in turn may lead to a different move
> ordering. Since only a limited number of moves are tried this could
> lead to a different result.
> 
> The way to verify this is to run --decide-dragon with and without the
> patch and compare the trace output (-t) and/or the generated sgf files
> (-o vars.sgf). If the point of divergence involves multiple moves with
> the same value, one of which comes from the modified patterns, the
> hypothesis can be assumed to be correct.

I'm verifying this and it seems you're right. For instance in owl:23,
pattern D1119 is getting matched before D1001 when the patch is applied.

I verified this for a couple other cases like strategy:29, lazarus:15, ...
At first, it seems to me that it would be wrong to try to solve these
problems by fine-tuning the values in the patterns. Can somebody confirm ?
Anyway, there are cases where this isn't even possible (in lazarus:15, the
move ordering changes because the same pattern matches at another place)

Should I verify the move ordering cause of all the deltas generated by this
patch (nando_3_9.5) ?

---- File [owl.tst]
23 FAILED: Correct '0', got '1 S10'               D1001 / D1119
---- File [strategy.tst]
29 FAILED: Correct 'R4', got 'E16'                D721 / D123
---- File [lazarus.tst]
15 FAILED: Correct 'Q15|T5|H5|H3', got 'S13'      D209 at S10 / S11
---- File [strategy2.tst]
82 PASSED
---- File [nngs.tst]
320 PASSED
1700 PASSED
---- File [13x13.tst]
63 PASSED
---- File [nngs2.tst]
110 FAILED: Correct 'B14|B13', got 'L15'
---- File [auto04.tst]
2 FAILED: Correct '!0', got '0'


/nando