[Top][All Lists]

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

Re: [Bug-gnubg] Gnubg's Cache and Plies > 3 - problem?

From: Christian Anthon
Subject: Re: [Bug-gnubg] Gnubg's Cache and Plies > 3 - problem?
Date: Tue, 1 Sep 2009 11:22:24 +0200

I thought (believe) that the cache isn't added to for noisy
evaulations. Let's make sure.

The reduction code is rather unlikely to work as it hasn't been
compiled in for a couple of years. We probably want to reimplement the
idea in terms of rollout anyway. Let's remove the reduction code. In
any case pruning and the reduction code a mutually exclusive in the
current implementations. Though it doesn't have to be so.


2009/9/1 Massimiliano Maini <address@hidden>:
> Isn't the reduction code the Snowie-like thing, where you can decide
> to examine only a subset of the rolls ?
> E.g. 50% = only consier half of the possible rolls for analysis/rollout.
> If it's the case, I do think we can throw it away.
> For information, my Win builds have REDUCTION_CODE  *not* defined.
> MaX.
> 2009/9/1 Øystein Johansen <address@hidden>:
>> On Tue, Sep 1, 2009 at 10:45 AM, Jonathan Kinsey <address@hidden>
>> wrote:
>>> Note that all 32 bits are used if REDUCTION_CODE is defined. I'm not sure
>>> what
>>> the "reduction code" does exactly (something about reducing the number of
>>> rolls
>>> used in evaluations for a quicker less accurate calculation?), is it a
>>> failed
>>> experiment or is it actually used in general?
>> It was something Nis Jorgensen came up with. I don't remember exactly how it
>> worked, but it was cool enough!
>> However it had to be #defined away to suite the pruning neural networks.
>> No wait.... maybe I remember incorrectly. Maybe it was David Montgomery who
>> added the reduction code and Nis who made a twist by letting some branches
>> in the go n-ply and some other go (n-1) ply. I think it actually worked, at
>> least when n=2.
>>> I'm all for removing this code if it was found to have little use.
>> Well, if someone wants  to take up this algorithm it would be interesting,
>> however if someone wants that they can find the code in the cvs archives
>> even if it's removed.
>> -Øystein
>> _______________________________________________
>> Bug-gnubg mailing list
>> address@hidden
>> http://lists.gnu.org/mailman/listinfo/bug-gnubg
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg

reply via email to

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