LDraw.org Discussion Forums

Full Version: New Keywords For Precision Search (very radical idea)
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Not sure where to post this, but I guess here isn't the worst option at least.

Trying to identify a minifig head or torso can be very difficult. I have been thinking of a system of standardized, unique keywords to narrow down the search in a very easy way.

So far, I've just made a pilot test with the 51 first minifig torsos that comes up in my Parts library, and added three new keywords. The point is to be able to filter for example, not only whether there are buttons or not on a torso pattern, but exactly how many. The three "variables" I've been experimenting with so far is:

btn for button count (btn7 for 7 buttons and so on)
pkt for pocket count
zpr for zipper count

What would be the most efficient search criteria, could of course be subject of discussions; these three were just the first that came to my mind. But yet, the combination of these three filters looks very promising. Very often, only one torso remains in the search. (But then again, I have only implemented the system on 51 torsos so far...) We could probably add other search criteria, such as base colour[s], belt [clip count], and... you name it, but on the other hand, we can still use the rest of the part descriptions and already existing keywords in our search. So the key is to find the best keywords that has the highest potential for effective filtering. So far, I've found that 'btn' is by far the most efficient variable, but combined with other criteria, the filtering effectiveness increases exponentially.

So, for example, now when I enter "btn7" in LDList (for 7 buttons), I get only one hit: 973p1f. I'm sure there will be more hits if I ever get through all torsos, but then I may enter police or badge in the search, and it will probably be very few hits.

What do you think? A massive amount of work, yes, but IMO extremely beneficial.
I like the idea of a theme-agnostic way of searching for minifig related items since, more often than not, it isn't very obvious what theme a torso/head/legs came from.
(2017-10-30, 18:29)Orion Pobursky Wrote: [ -> ]I like the idea of a theme-agnostic way of searching for minifig related items since, more often than not, it isn't very obvious what theme a torso/head/legs came from.
I like the idea too. Could be the opportunity to add a keyword that links to other pattern classification database, such as Bricklink, to ease the search by set inventory.
That sounds like a good idea to describe patterns like this.

Like I mentioned here, it would be good to have slightly larger pictures in the patterned part ref to be better able to distinguish between the different torsi or other e.g. heads.
Thank you for the positive responses.

I have just spent an awful lot of time updating keywords for 406 patterned torsos, expanding ldrawlist.txt with 22 Kb.
But it was worth it; the result is very promising.

One problem I soon stumbled on was how to define a button! Sounds simple, but some space alien torsos can be virtually impossible. My solution is to have multiple numbers in some tricky cases. The same goes for some jackets where one can assume that there is a zipper. Just like some 3039p* slopes in Classic Space, you can have different opinions about which circles and quad's are lamps and which are buttons. I've been generous and counted some "looks like a button/zipper/pocket" as hits.

I've also introduced the keyword 'Dual' - but not from when I started this work. I realized the need after a hundred torsos or so. But I will go back and check the first ones. I strongly recommend this keyword for minifig heads and bricks with double-sided print also.

I started out on belt and "belt clip", but didn't like it. Belt and buckle is better. Best would dedicated variable names be, like the three I've tried out already: zpr, btn, and pkt. Why? Because then you can get more powerful filtering even in those environments that miss the option to put a minus sign in front of search words.

Another good word would be wrinkles. But also a bit hard to define at times. Are these lines wrinkles or pockets...?

Maybe also neck color.

Base color has been mentioned before. This can be a very effective filter if implemented, especially later for non-yellow minifig heads. Very few patterned torsos have come in more than two colors. They can be left without that data. I would prefer to write it as one word, as a "Blue" standing alone will make a false hit with the keyword "blue pen" in most search engines. So, BCLBlue or PCLBlue wouldn't look quite so nice, but would make much easier and better precision even with already existing search tools, as LDList.

So, when I'm finished with this first step, we have zpr(n), btn(n), pkt(n) mandatory for torsos. And Dual, tie, bowtie, belt, buckle, and navel consistently when they appear. Did I forget anything here? I certainly did.

I did not realize there were so many torsos!