[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] trevor_1_14.5

From: Trevor Morris
Subject: Re: [gnugo-devel] trevor_1_14.5
Date: Sat, 17 Nov 2001 12:47:25 -0500

At 12:16 PM 11/16/2001 -0800, you wrote:
>> >
>> Why does this patch remove owl attack pattern A1?
>> That pattern looks needed to me. You could quibble with
>> the valuation.
>I added this patch to the CVS. But it removes four patterns,
>A1, A303, A304, A1001 and D211. I let the removal of A1001
>stand, but I didn't remove the other patterns. A1, A303 and A304
>definitely look needed, and the specific tuning of D211 
>(value 71) makes me think that yes, as the patch comment says,
>that move might be covered by other patterns but with lower valuation.

A1: Gunnar's defended already - thanks, so perhaps it can be salvaged.
A303: Could be useful, though one space to the left of a is prob. better.
A304: B should never try *, so no need to defend it.
A1004: I couldn't see why this made sense.
D211:  I saw the following match: 
 |..O       try before hanging connection
 |.*O   #This O ...
 |.O.   #  and this O are not present in the pattern.
  - This was such a bad move, that I removed the pattern.

On all of these, removing them broke no regression test.  Going with
the strategy that more is less, I still believe they should all be 
removed, until a suitable test case that requires them has been 
constructed.  We could debate the relative merits of various patterns 
for quite a while.  It seems to me more productive to let the test 
cases speak for themselves.

I didn't mention that this patch reduced the number of owl_nodes by
about 9% (from 19012 to 17351) in owl.tst.  Partly by removing these
patterns, and partly by improving the reading on other problems.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]