In , share and fluidly explore data in many dimensions.
Paint a new datascape
To make an interactive datascape, sign in to Pcaso and click the Paint new datascape button. Then just upload a .csv file*, pick fields to show as axes, or ID/other metadata, add a caption if you like, and set the datascape as public (viewable by anyone) or private (viewable only by you and anyone else you invite). You’ll instantly get a datascape to explore and share as you wish.*Use a Windows-formatted .csv, e.g., as exported from Mac or Windows Excel via the Save as dropdown. And keep numeric data in standard decimal (not scientific/exponential) notation.
Explore a datascape
After opening a datascape, click any small plot to see it big. Click any metaclass (listed next to the plots) to recolor points by that kind of underlying data. Hover on a point in the big plot to see its metadata. And to highlight particular points that interest you, either
- click a value for the currently chosen metaclass
- search by text box to find points by any matching metadata
- click or click-drag one or more points in the big plot
Share a datascape
You can drop any public datascape’s URL into an email, text, tweet, research manuscript (great for preprints), carrier pigeon, etc. Giving a talk with data? You can fluidly walk through the public datascape, on- or (once loaded in your browser window) off-line, and show your listeners the URL to explore live on their own.
To more selectively share a private datascape in your Pcaso.io portfolio, you can invite someone to come see it by adding their email address to the Share privately with list for that datascape.
Remix or delete a datascape
To change the basic axes, metadata, or caption for a datascape in your Pcaso.io portfolio, click the gear icon by its title, change settings as desired, and click Update datascape to see the new version. To change the title (which determines a datascape's URL), for now you must reupload the underlying data, to paint it with a new title.
To delete your datascape, click the gear icon by its title, then click the Delete datascape button.
Why beats paper
We people see, and think, in too few dimensions to easily understand complex data with many variables.
Conventionally, we’ve frozen such point cloud data flat, to view as dots scattered in the plane of a page or screen, perhaps varying by color or size to say a bit more. This convention can help spotlight one pattern...only to hide those lurking in other dimensions. And while adding plots to see other angles or metadata can help a bit, in the end, static plots can’t readily show which dots in one view of a point cloud correspond to which in another – a key question in probing complex data.
To help understand many kinds of such data, in genomics and other fields, we built the Point cloud analysis stereopticon (), an easy way to collaboratively explore planar views (e.g., PCA, MDS, comparative abundance, &c.) of data with many variables. lets you fluidly explore point cloud data posted by others – and easily make your own informatively responsive datascape to drop into a paper, talk, or tweet for others to explore.
Crucially, transcends key limitations of static plots, letting you
- informatively explore data in a responsive, shareable datascape
- smoothly switch views of the data, to track how each point shifts relative to others, among potentially many dimensions
- instantly recolor, highlight, and see more about points by clicking listed metadata, hovering or clicking on points, or text-searching all metadata
- smoothly zoom, to see patterns big and small
- caption a datascape to explain details, invite contact, &c.
What kinds of data does help with?
Basically, anything with up to 20 thousand or so datapoints, in several to many numeric dimensions, with (ideally) several kinds of categorical or other metadata. Could be principal components, of course – but lots of other data can benefit from smoothly switchable, metadata-responsive planar views. If you’ve got such data, try it out.
Does itself run PCA (or otherwise statistically analyze data)?
No. We aimed first to help explore and share many-dimension data that you (or others) have already generated, via a simple exchange format (.csv), knowing that many folks prefer to crunch rawer numbers with varied, question-tuned methods and software. Over time, we may build stats directly into – so stay tuned.
Will you mine, rent, or sell the data I upload?
How robust is ?
We’ve started with a beta hosted on a good standard web server, with basic features to spark collaborative insights – not (yet) to specially encrypt data or stop determined bad actors. Please use responsibly.
What about parallel-coordinate plots?
They too can help a lot, by showing many dimensions (especially qualitative or naturally ordered ones) at once. But following tangled curves, and spotting correlations between coordinates shown far apart, can be tough.
Why not show a 3d box instead of planes?
We started with planes that mean something real from underlying data, and that our brains can easily track when switching axes. (Plus, for data with many axes, a box shows little more of the whole than a plane does.)
But doesn’t a stereopticon show 3d images?
That’s a stereoscope ;-) In steampunk days, stereopticons just let people switch from one slide to another. We were also inspired by cubists (like Picasso, of course) who splintered complex real-world shapes into many-planar views, to convey them more deeply.
We hope to soon let you see simple regression or/and cluster analysis on plots, to better spot meaningful patterns -- along with other handy features, elaborated by us and others via CC0-licensed d3+ code, to help manage and interpret diverse complex, plane-projected datasets.