IJulia display logic

[tbreloff]

I find it annoying that IJulia seems to call multiple display formats, which is a waste and sometimes does the wrong thing. I'll probably want to override this:

https://github.com/JuliaLang/IJulia.jl/blob/master/src/inline.jl#L31-L40

and I might want to wrap it in a @require IJulia.

[tbreloff]

Possible solution: https://github.com/spencerlyon2/PlotlyJS.jl/blob/5b896eb49f386fc21e8154997a24b26649df315b/src/display.jl#L102-L123

[tbreloff]

Fixed by ed1cce8

[jheinen]

Great - now it works as expected.
Will solve the GR.savestate / GR.restorestate problem (caused by exceptions) in GR - probably by sth like GR.rewind() => to unwind the attribute stack.

[diegozea]

Hi! The IJulia notebook breaks with huge plots. Will be great to use "png" or format smaller than html/svg with very heavy plots. Thanks!

[tbreloff]

There are a couple possible options... make the display output a keyword argument during plot creation (easy-ish), or add some conditional logic to change based on data size (tricky to do well). I'm open to suggestions... Hard data and benchmarks would be important to drive implementation choices. Also this will be backend-dependent, so even trickier to get right.

On Apr 26, 2016, at 8:36 PM, Diego Javier Zea notifications@github.com wrote:

Hi! The IJulia notebook breaks with huge plots. Will be great to use "png" or format smaller than html/svg with very heavy plots. Thanks!

—
You are receiving this because you modified the open/close state.
Reply to this email directly or view it on GitHub

[diegozea]

I prefer the keyword argument at the moment, since the conditional logic could be included in the future using a function as its default value.

[jheinen]

What about function backend(pkg::AbstractBackend, fmt="png"), overwriting the defaultOutputFormat in output.jl?

@diegozea : Do you have such a huge plot example? I'd like to find out what kind of output breaks IJulia (or the browser). I assume you were using the gr() backend, right?

[tbreloff]

What about function backend(pkg::AbstractBackend, fmt="png"), overwriting the defaultOutputFormat in output.jl?

@jheinen you may not realize this, but the convenience method for switching backends takes optional keyword arguments that will be passed through to the default(; kw...) method. This means that you can do something like:

using Plots
gr(fmt = "png")

and it will apply aliases and set default settings for keyword arguments.

The advantage of making it a "plot keyword" is that you can choose the format on a per-plot basis, if you want.

[diegozea]

I'm working with matrices of ~ 13000x200. I was mainly using PyPlot and PlotlyJS as backends.

It looks as a PyPlot problem. GR and PlotlyJS show fast the huge heatmap, but the PyPlot heatmap doesn't show up with 5000 or more rows (and uses a lot of RAM trying to do it). Even a PyPlot heatmap for a matrix of 1000x200 matrix takes ~ 1m to show up, and uses 0.4G of RAM to do it.

After the plot of the 13000x200 matrix, the RAM increases 0.1G with GR and 0.3G with PlotlyJS. I remember having troubles with PlotlyJS in some IJulia sessions using a smaller computer (not only with PyPlot). PlotlyJS times are a little misleading, since toc() was executed by Ijulia before the plot appears.

PyPlot

Small matrix

Large matrix

Takes forever and Ubuntu swaps...

GR

Awesome performance @jheinen !

Small matrix

Large matrix

PlotlyJS

Small matrix

Large matrix

Gadfly

Gadfly doesn't support heatmap yet

[jheinen]

@diegozea : Thanks for the info. So, there seems to be no need to fix the problem in GR.

@tbreloff 👍 Goot to know:

using Plots
gr(fmt = "png")

[diegozea]

@tbreloff The time and memory issue with pyplot can be solved using png. Since pyplot is fast when the display is omitted and the plot is saved into a png as show it using loadfrom Images.jl inside the IJulia notebook.