TalkChess.com

I understand your concern, but I suspect in practice it won't matter. Looking at enough examples, the network can figure out that strong positions for white end up being either won by white or draws the vast majority of the time, even if it doesn't know that loss < draw < win.

You don't propose an alternative, so it's hard to discuss pros and cons. However, these matters always come down to an empirical question. Having a scheme that produces distinct probabilities for the outcomes has the advantage that they can be combined however necessary when we use the network. For ...

Was a set of input planes representing a position at time (t - T + 1) oriented for the side to move at time t or time (t - T + 1)? All the history input planes are oriented for the current side to move, i.e. at time t. Would it matter if they weren’t? (That’s a serious question btw) I suppose not b...

Here's a hacked-together Perl implementation: cat kk.pgn | perl -ne ' if (/^\[/ || /^$/) {print; next;} for $token (split " ", $_) { if ($token =~ /^[1-9]/) {$wtm = 1; print "$token "; next;} if ($wtm == 0) {$token =~ s/[!\?]//g;} print "$token "; $wtm = 1-$wtm; } print "\n" ' That removes markings ...

smatovic wrote: ↑

Fri Feb 01, 2019 6:03 pm

Myself still ponders on how to combine an AlphaBeta searcher with gpgpu ann eval...

I have a promising idea for that, but I don't really have the time to implement it. My job, my wife and my kids are getting in the way of the really important stuff!

If you have access to Perl, you can have much more control, like this:

You should concentrate on move ordering, because it's much more important than lazy generation. In my engine RuyDos I roughly do the following: * generate captures and sort them by MVV/LVA, * filter the losing captures (SEE<0) and save them for later, * generate non-captures and sort them by history...

Input of a training example consists of position value plus probability for each legal move And how do you know the probability for each legal move from a PGN? You don't. In the normal training of AlphaZero and LC0, your training samples are written at the end of a search, and the visit count of ea...

Henk, you are adding noise to the conversation.

What you are describing is the sort of reinforcement learning that was described in one stage of the initial AlphaGo paper. We are not talking about that here.

Input of a training example consists of position value plus probability for each legal move And how do you know the probability for each legal move from a PGN? You don't. In the normal training of AlphaZero and LC0, your training samples are written at the end of a search, and the visit count of ea...