IMD Scatter Matrix

Comment

Comment by Robin Gower on August 4, 2008 at 2:06pm

It seems pretty scalable with the datasets I've used.

When you say "features", I presume that's what I would call "fields" - i.e. columns of table data. If so, then the answer is probably yes but I imagine that the chart would become quite unreadable with any more than 25 fields. I would recommend that you estimate some summary statistics to reduce the number of fields.
I've used 8 continuous fields as matrix dimensions (each is compared pairwise).
I've also segmented 1 of the continuous fields (IMD Score) into deciles (median 80% combined) to give the colour range.

If by "features" you actually mean records - i.e. rows of table data - then the answer is an overwhelming yes! Each of the 64 cells in the graphic covers every one of the 32,482 "Super Output Areas" - i.e. streets - in England (though at this scale there is quite a bit of overlap).

Hope this helps.

Comment by Sobhan Hota on August 4, 2008 at 4:24am

Can this tool handle feature selection/representation among thousands of features?

Comment by Robin Gower on February 19, 2008 at 3:14pm

You can find a full explanation on my site:

http://infonomics.ltd.uk/?q=node/12