plotting 5 million points in R

[Jon17]

I plotted 5 million points in R. Is there a way to clean this up? Make it easier to spot patterns?

[Jon17]

I'd like to reproduce this but don't know how:

[dpryan]

I would recommend using something like geom_tile() or geom_raster() in ggplot2 rather than plotting the actual points. That will make it vastly simpler to spot trends.

[Jon17]

geom_raster looks really nice!

Unfortunately I only have 2 data columns. Looks like geom_raster requires 3 columns? the faithfuld plot used in the example link below has 3 columns in it,

1) waiting
2) eruptions
3) density

all 3 are used to create the plot. Can you apply this to 2 column data?

http://docs.ggplot2.org/current/geom_tile.html

[dpryan]

Ah, right, try geom_density_2d() instead.

[Jon17]

Thanks for the tip, but now I'm having the opposite problem. The plots are too spars. Either the whole plot is colored or almost nothing at all. :-p

[Brian Bushnell]

Maybe this is too obvious, but have you tried randomly subsampling?

[Jon17]

That worked.... thanks! duh... :-)

I read threads of people plotting millions of datapoints so I just assumed R could handle it easily. Looks like 100k to 1 M is the perfect range for pch='.'

[gringer]

You can also use smoothscatter, which plots a smoothed contour plus points for any outliers

[thermophile]

If you want to plot the actual points, play with alpha (transparancy) using one of the solid shapes (15-20). R can certainly handle it just be aware that the file will be huge.

[gringer]

R can certainly handle it just be aware that the file will be huge.

If you output a PNG file rather than an SVG or PDF file, then the output file will not be huge. It just won't have as much information in it, and won't be infinitely scalable.