I've wondered that sometimes. I think it's because a lambda opens to the left, and starts with "l". Rhonda opens to the right, and is a similar word starting with "r". It's not supposed to make sense; it just cracks me up :P
I managed it, though it thought it was Omega first (I had a curve to my Pi though). When I used straight-line Pi it got it first go. Maybe it's more trained now.
This is awesome, and it works. My only quibble is that it doesn't seem to handle accented characters, ie it finds \aa, presumably because that's a single symbol, but it doesn't find \"{a} and it doesn't even seem to show up in the training list.
Missing feature: the ability, after training a symbol, to click a button whose meaning is "Whoops, I didn't mean to do that". (I gave it some training examples for a couple of arrow-type symbols. I forgot to click to switch from one sort of arrow to the other. It now has a left-pointing arrow as a training case for a right-pointing arrow symbol. Or maybe the other way around; I forget which.)
Of course that was an utterly boneheaded mistake. But most people -- even among the likely users of this thing -- are boneheaded occasionally.
(Looking at the source, it seems that un-training would be quite easy; the app is doing nearest-neighbour classification, and what it stores is just all the examples it's been given.)
This might be nitpicking but I feel like the "Clear" button is misplaced. There must be another position for it that allows me to quickly reset and start drawing another character. Come to think of it, it would be great if there was a key for it. Click and drag with the mouse, reset with a keystroke. Almost like playing a first person shooter!
Well hopefully the author of this will give a real summary, but I had a quick squiz at the code at its nothing like a CAPTCHA solver.
Basically its a nearest neighbour classifier based on turning the actual stroke data in sets of equally spaced points after being centred and scaled. The direction between points is recoded as compass directions (so sequences of N,E,NE). These strings are matched against the training data to find the closest matches.
Yes, noticed this also for the Ampersand symbol. LaTeX (which I'd not previously been familiar with) must have a specific purpose, which seems great and appears to be made greater by this nifty tool.
LaTeX is a typesetting program (related to TeX which was made by Knuth). Its quite popular in academia (sciences at least) because it handles equations really well.
Its analogous to a markup language in that it is not a WYSIWYG typesetting program.
\leq and \pi work consistently now, just trained it to do \geq, and also did \otimes and \circledast. Be careful what you say about it, it looks like a fast learner ;)
It would be awesome if this were integrated into a LaTeX editor, so I could (e.g.) click a "draw" button in the toolbar. Then I'd get a little drop-down canvas, and when I get the suggestions back from the server, I could click on the right macro, and it would be automatically inserted into my diagram.
Have you guys seen Windows 7's math input thingie? It does the same thing for formulas (as opposed to symbols) and MathML. I was amazed to see this in a default install of a Microsoft OS.
38 comments
[ 2.7 ms ] story [ 97.2 ms ] threadedit: It allows you to submit training examples, which is a great idea. Just did two for lower-case sigma.
Of course that was an utterly boneheaded mistake. But most people -- even among the likely users of this thing -- are boneheaded occasionally.
(Looking at the source, it seems that un-training would be quite easy; the app is doing nearest-neighbour classification, and what it stores is just all the examples it's been given.)
Basically its a nearest neighbour classifier based on turning the actual stroke data in sets of equally spaced points after being centred and scaled. The direction between points is recoded as compass directions (so sequences of N,E,NE). These strings are matched against the training data to find the closest matches.
Great idea and great implementation. I have seen so much LaTeX frustration. This will definitely ease a lot of pain.
Its analogous to a markup language in that it is not a WYSIWYG typesetting program.
[1] OK, not actual useful, but it can be done.
Instead just give me a "done" button.
and good luck with /leq ...