[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

A Somewhat Tricky 43 - Rollout & Comments

Posted By: Nack Ballard
Date: Wednesday, 17 October 2018, at 6:58 p.m.

In Response To: A Somewhat Tricky 43 - Rollout & Comments (Stick)

First, people shouldn't toss around the term 'computer bug' when talking about this sort of thing. To me, this is not that. A computer bug is some kind of flaw in the software that would cause the program to not act normally. XG on its lower settings makes bad plays when playing this out, that's all. I don't see how that can be considered a bug.

(By the way, I've been running a 1k rollout at XGR+ level since last night -- had been planning to post but I guess your 159-trial ++ rollout is sufficient to make the point. After 1058 trials, mine has 5/1 4/1 best and 6/3 5/1 behind by .015, with other plays in the mix).

I readily admit that you know more about computer programs than I do. Whatever we call this bug / software flaw / whatever, though, it seems to me that there's something seriously wrong that needs to be addressed. It would behoove us (or Xavier, if he's around) to figure out (or more closely glean) what's wrong, so that we will have a better sense of what to trust or not trust in the future.

This large error hasn't shown up merely in a rollout of 1-ply or even 2-ply; it has happened at 3-ply (with XGR cube). And it hasn't happened in a post-containment or mass-backgame situation (we know that we need to be highly skeptical of such results, if not ignore them completely). It has happened in a simple holding game.

In simple positions, we've seen smallish XGR++ mistakes at 3-ply. When I've investigated these small ones, the culprit has been in one or more erroneous cube or checker plays immediately done by the other side, or by the original player on the next turn. That approach can be adopted here by anyone who wants to spend the time; if anything it will be easier because either there is some sort of (3-ply) octuple-whopper being perpetrated, or a bunch of blunders spread out over multiple checker-play or cube decisions. (You need find only one for encouragement that you're on the right track.) That's what it would take to cause a .066 (or so) outage.

What sort of blunder might 6/3 5/1 induce from the other bot-player? Perhaps White will split off the 18pt when behind in the race, Black will improve his board (instead of hitting) and White will be left in just a bad race or worse. That might be a place to start. There are other possibilities. Or if no such large error(s) are found, I guess the problem is systemic (if I haven't bungled another computer term).

Nack

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.