[yb2-Meta] MeowthBot/Generator clustering || List/Table Changes

[reiella]

The hand-gen move-gen clustering is getting somewhat frustrating to me to see...

Just did a minor change to the way meowthbot selects specials (same statistical probability, just should be less likly to 'cluster' specials) by changing the 1 in 20 chance from an only active on a 1 roll to a only active on some different number (in fact, it should be a different number for each 'special').

Also... I will probably be working at fixing the clustering with the list fields later by simply shifting the entry generated out of the list and to either the begining or the ending or the list (psuedo-randomly determined unless move originally came from either of those two).

 

[reiella]

Finished updating my test-gen with an easier to read and possibly lower clustering version of the code.

I also went through and fixed some typos in the listfields so at least they're proper spelling (Groto -> Grotto; cherry blossum -> cherry blossom; lotus blossum -> lotus blossom; vaccuum -> vacuum).

The web-gen spelling corrections so far are just in the the move-gen on my test-gen... MeowthBot should have them throughout.

Sometime later today I'll prolly go about fixing the insult and masters for my test-gen as well.

[ And of interesting note... while testing this, I got the 'undefined' field error again... I'm getting close to fixing it, unless no one really cares or notices. I think I may still be the only one who has actually seen it. ]

 

[reiella]

I will also be changing the three troublesome fields in the defensive target list sometime this weekend.

For those who have missed them, they are Thorn (it is also a style), River (it is also a location), and Spirit (it is listed twice).

Right now it looks like I'll be changing them to void, scarf, and tomb (drifted through OA to find some words and came up with those). Would be nice to hear some other suggestions though.

 

[graydoom]

Void sounds good to me.
Tomb and scarf, however, don't. I would suggest "lightning" (could make for some good syngery moves along with (or defending from) the thunder style) and perhaps "mind" to replace those two.

 

[graydoom]

One more thing that I just remember. In chart 9, "in addition, it" needs to be changed to "and in addition it". The current way it is makes moves like this, that really don't sound right (taken from (YBA)Sharlea enters Wicht's Lair Alone).

The fist sweeps atop the plain in addition, it thrusts the thorn; dodges the rising of the rodent as it waves away the onslaught of the orchid!

 

[reiella]

Fixed the the conjugation grammar error on my test-gen and mb.

Waiting to update the def targets for a while in case others have suggestions.

 

[Kalanyr]

I suggest plant instead of thorn for the def targets.

 

[reiella]

Just shifted my test-gen and MeowthBot's generator over.

Thorn->Plant
River->Lightning
2nd Spirit->Mind

Vacuum->Void (this just reads better imo, not as much a sciencey feel, I also remember the generator commitee agreeign to this change when we were fixing the 'new' generator a while back... Not sure why it never made it though.

 

[Lady Diamond]

Meowth! Reowth!

Would ya mind explaining in layman's terms what you're doing? I have several years experience in mainframe program code, and what you've said here doesn't make sense to me much.

How 'bout slowin' down a bit hm? It'd also be nice that since folks are using both the web generator and Meowthbot that the two stay similar so we don't get into conflict over who uses what.

 

[reiella]

With the gen, there is the possibility of pulling one style (thorn) and one location (river) in a place where they really aren't expected.

By the letter of the rules currently, they would result in the appropiate benefits (including immunity for thorn)... This is not actually intended, and was something that was purposefully removed from the generator except apparantly a few were missed. Spirit was also listed twice for no real reason... Vacuum was an artifact that was agreed removed in the last discussion, but apparantly KD never got around to it.

The clustering change in itself is simple.

Just change the number dependant for every single special to not be the same. Same statistical probability, less likly to get cluster (which are more noticable with the hand-gen functions) on the specials.

There's also some of a way to extend that to styles and other datafields, but would take a bit of a semi-major shift in the coding paradigm used for the web-gen at least and will prolly beisomething I hold off on (effectively shifting all the lists to be global and when one item is pulled, shifting it out of the list and then reinserting randomly at either the head or tail), only real difference this makes is to try to skirt away from the clustering that is somewhat evident with the hand-gen functions.

The rest were fixes to spelling and grammar.

[ Clustering ... The psuedo-random behavior/nature of the generator becoming evident by multiple (30 actually) checks at the same general time. Resulting in a non-randomly liklihood of numbers to be 'clustered' around the same region... Partially resulting in an oddity of what appears to be a gob of moves that have abnormally many specials or an excessive strand of 'normal' moves... It is alot more noticable in the listfields and a couple people have actually mentioned it occuring there and I do believe you made one comment about that some time back on the old boards... A set of moves that use similiar fields in areas. This becomes alot more obvious with rougher random number generators (ask Gray about the roaring phoenix bug mb used to have before I updated the random core)... It's alot easier to fix in terms of the specials as they just compare against fixed numbers, change those numbers and the clustering behavior is diminshed.]