Re: [Bug-gnubg] Strange New Contact Net Build backgame Evaluation

[Øystein O Johansen]

> POS ID:         vHsXAADCNhQCNw
> MATCH ID:   MAEAAAAAAAAA
>
> OLD CONTACT NET:
> 2 ply 0.399/0.011/0.000--.601/.393/.122 (-0.202 (-0.915)
> NEW CONTACT NET:
> 2 ply 0.295/0.007/0.000--.705/.478/.175 (-1.056 (-1.244)

I see! This is a position where it seems like the odd plies makes a better
estimation. I don't think we should be surprised that the contact net may
be worse on some position. (I guess the Snowie 4 network is weaker than the
Snowie 3 network in some positions as well.)

A question to Joseph: When breeding new nets, you add positions to the
database when 0-ply and 2-ply disagrees on a move, by adding the two
resulting positions after the two different moves. Can it be an idea to
also add the resulting positions _with the other player_ on roll? This way
there will be four new positions for each move 0-ply and 2-ply disagrees.
This will maybe help some of the even/odd-ply disagreement in some of the
positions, since it will train on a similar (not equal, since it's the
"wrong" player on roll) position seen from the other side. Do you think
that will improve the training?

 I'm really looking forward to see how Nis' half-ply algorithm works on
positions like this.

-Øystein


-------------------------------------------------------------------
The information contained in this message may be CONFIDENTIAL and is
intended for the addressee only. Any unauthorised use, dissemination of the
information or copying of this message is prohibited. If you are not the
addressee, please notify the sender immediately by return e-mail and delete
this message.
Thank you.