We were talking in the office the other day about a fun little project for twitter. Basically just looking at what pairs of hash tags get used together. After getting one and a half hours sleep last night, waking up and being unable to get back to sleep I had some time to kill on my hands, so thought I’d throw it together.
Getting and munging the data into a form that gave tweet similarity was easy enough. But what to do with it then? The obvious thing to do is to visualise the resulting graph.
We have our own visualisation software at trampoline (which I did try on this data. It does fine), but I wanted something smaller and more stand alone. I’d heard people saying good things about IBM’s many eyes (in retrospect I may have to challenge them to duels for the sheer affrontery of this claim), so I thought I’d give it a go.
Let me start by saying that there is one single feature that would change my all encompassing burning loathing for Many Eyes into a mild dislike. It alleges to have the ability to let you edit data you have uploaded. Except that the button is disabled with the helpful tooltip “Editing of data is currently disabled”.
This renders the entire fucking site useless, because it takes what should be a trivial operation (editing the data you’ve uploaded to see how it changes the visualisation) into a massive ordeal. You need to create an entirely new dataset, label it, add it, recreate the visualisation…
Fortunately recreating the visualisation isn’t that hard. After all, Many Eyes doesn’t actually give you any functionality with which to customise your visualisation (maybe it does for some of the others. It sure doesn’t for the network viewer).
So why did I need to tinker with the data? Isn’t it just a question of upload a data set, feed it into IBM’s magic happy wonderground of visualisation and go “Oooooh” at the pretty pictures?
Well, it sortof is.
Actually what it is is upload the data set, feed it into IBM’s magic happy wonderground of visualisation and go “Aaaaaaaargh” as my browser grinds to a halt and then crashes.
It’s understandable really. I did just try to feed their crapplet an entire one point six MEGA bytes of data (omgz).
Wait, no. That’s not understandable at all. In fact it’s completely shit. That corresponds to about 12K nodes and about 60K edges. This is *not* a particularly large number (metascope happily lays it out in a few tens of seconds). This is a goddamn data visualisation tool. The whole point of it is that you’re supposed to be able to feed it large amounts of data. At the very least it should tell you “Sorry, I was written by monkeys so probably can’t handle the not particularly large amount of data you have just fed me”.
So, I spent some time trying to prune the data down to a size many eyes could manage to not fail dismally at but where the graph was still large enough to be interesting. This was a hilarious process. Consider:
- The only way to edit the data is to create an entirely new data set and recreate the visualisation.
- The only way to determine that I’ve got too much is to observe that my browser crashes.
After about half a dozen iterations of this I decided enough was enough and declared many eyes to be an unusable piece of shite that was not worth my time. Life’s too short for bad software.
That data set is not all that large, but it’s probably too much for a Java or Flash applet running in a browser (which are very memory-constrained). I agree that it should fail more gracefully, though, that is the one thing I would really fault them for. (It also shouldn’t crash your browser, but that’s a browser issue, not the applet’s fault – the browser should be able to contain a crashing applet)
Many Eyes’s focus is more on trying out new things in terms of visualization (they do a lot of text analysis with visualization) and crowdsourcing/interaction. Most datasets people want to play with are probably in the dozens or hundreds of data points, with a few dimensions, so scale is just not an issue for the majority of their users.
The whole thing is still an experiment really, more of a prototype than a production service. It’s hosted on alphaworks, after all …