Gadfly is definitely not efficient.
Your 2nd timing on Bokeh seems either extremely encouraging or a bit
suspicious. For reference, with Winston I get (after warmup)
julia> @time plot(x,y)
elapsed time: 0.003610142 seconds (278 kB allocated)
julia> @time display(plot(x,y))
elapsed time: 0.262181796 seconds (16 MB allocated)
""
The first one just creates the "task" (essentially, constructing the scene
graph), and the rendering happens after the `@time` macro ends. The second
version includes the time to render it. So, does your ~4ms timing include
rendering, or is it just the time needed to hand the task off to Bokeh?
--Tim
On Monday, February 23, 2015 06:43:49 PM Samuel Colvin wrote:
> Humm, gadfly does seem very slow even on subsequent calls, perhaps plot is
> generating the SVG even though it's not used?
>
> julia> @time draw(PNG("test.png", 6inch, 3inch), plot(x=x, y=y))
> elapsed time: 37.313627903 seconds (2044604824 bytes allocated, 3.81% gc
> time)
>
> julia> @time draw(PNG("test.png", 6inch, 3inch), plot(x=x, y=y))
> elapsed time: 21.228884246 seconds (1128013088 bytes allocated, 3.59% gc
> time)
>
> julia> @time draw(PNG("test.png", 6inch, 3inch), plot(x=x, y=y))
> elapsed time: 21.632743948 seconds (1128006012 bytes allocated, 6.86% gc
> time)
>
> julia> import Bokeh
>
> julia> Bokeh.autoopen(true)
>
> julia> @time Bokeh.plot(x, y)
> elapsed time: 1.815997064 seconds (125026480 bytes allocated, 4.95% gc time)
> Plot("Bokeh Plot" with 1 datacolumns)
>
> julia> @time Bokeh.plot(x, y)
> elapsed time: 0.003634852 seconds (24107784 bytes allocated)
> Plot("Bokeh Plot" with 1 datacolumns)
>
> julia> versioninfo()
> Julia Version 0.3.6
> Commit a05f87b* (2015-01-08 22:33 UTC)
> Platform Info:
> System: Linux (x86_64-linux-gnu)
> CPU: Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz
> WORD_SIZE: 64
> BLAS: libopenblas (NO_LAPACKE DYNAMIC_ARCH NO_AFFINITY Sandybridge)
> LAPACK: liblapack.so.3
> LIBM: libopenlibm
> LLVM: libLLVM-3.3
>
>
>
> --
>
> Samuel Colvin
> [email protected],
> 07801160713
>
> On 23 February 2015 at 18:35, Stefan Karpinski <[email protected]> wrote:
> > Yes, that 33 seconds for the first Gadfly plot is all code generation –
> > the second time you do it, it certainly doesn't take 33 seconds.
> >
> > On Mon, Feb 23, 2015 at 1:34 PM, Tim Holy <[email protected]> wrote:
> >> Do those timings include compilation? It's not really meaningful on the
> >> first
> >> run.
> >>
> >> For reference: on my laptop, Winston (on the second run):
> >>
> >> julia> @time (p = plot(x,y); display(p))
> >> elapsed time: 0.256627468 seconds (16 MB allocated, 1.77% gc time in 1
> >> pauses
> >> with 0 full sweep)
> >> ""
> >>
> >> Even this is slow by comparison to where I think we want to be.
> >>
> >> --Tim
> >>
> >> On Monday, February 23, 2015 06:27:24 PM Samuel Colvin wrote:
> >> > You're probably right about research publications, I guess plots these
> >> > don't need to be interactive which makes things easier from a cross
> >> > platform perspective.
> >> >
> >> > Performance wise I'm not sure you're right, with Julia 0.3.6 and latest
> >> > packages:
> >> >
> >> > julia> using Gadfly
> >> >
> >> > julia> x=1:1000000
> >> > 1:1000000
> >> >
> >> > julia> y=sqrt(x);
> >> >
> >> > julia> @time draw(PNG("test.png", 6inch, 3inch), plot(x=x, y=y))
> >> > elapsed time: 33.860814218 seconds (2043746808 bytes allocated, 3.83%
> >> > gc
> >> > time)
> >> >
> >> > julia> import Bokeh
> >> >
> >> > julia> Bokeh.autoopen(true)
> >> > true
> >> >
> >> > julia> @time Bokeh.plot(x, y)
> >> > elapsed time: 1.557460583 seconds (125617712 bytes allocated)
> >> > Plot("Bokeh Plot" with 1 datacolumns)
> >> >
> >> > Timing on my phone, the Bokeh plot had opened in chrome in 6 seconds.
> >> > It
> >> > was a little slow but still fine to zoom/pan etc.
> >> >
> >> > One of the nice things about Bokeh is that unlike d3, plotly or Gadfly
> >>
> >> it
> >>
> >> > uses canvas not SVG for it's plots which makes it way faster.
> >> >
> >> >
> >> > --
> >> >
> >> > Samuel Colvin
> >> > [email protected],
> >> > 07801160713
> >> >
> >> > On 23 February 2015 at 18:00, Stefan Karpinski <[email protected]>
> >>
> >> wrote:
> >> > > Bokeh and Bokeh.jl are both very cool – thanks so much for all the
> >>
> >> work on
> >>
> >> > > the package!
> >> > >
> >> > > There seem to still be visualization tasks that have scale and
> >>
> >> performance
> >>
> >> > > requirements such that HTML and JavaScript don't cut it. Web
> >>
> >> technologies
> >>
> >> > > are also generally not up to the task of producing
> >> > > publication-quality
> >> > > graphics, e.g. for research publications. The gaps are probably both
> >> > > diminishing, but I don't think we're quite there yet.
> >> > >
> >> > > On Mon, Feb 23, 2015 at 12:38 PM, Samuel Colvin <[email protected]>
> >> > >
> >> > > wrote:
> >> > >> To coincide (approximately) with the release of Bokeh v0.8.0 I've
> >> > >> released a significantly improved version of Bokeh.jl:
> >> > >>
> >> > >> http://bokeh.github.io/Bokeh.jl/
> >> > >>
> >> > >> This is the first plotting library I've built and the first proper
> >>
> >> Julia
> >>
> >> > >> package. I would therefore really appreciate any feedback on the
> >>
> >> plotting
> >>
> >> > >> interface and the structure of the package itself.
> >> > >>
> >> > >> Bokeh.jl is still a bit rough round the edges and missing some basic
> >> > >> features, but the examples above demonstrate what it can do.
> >> > >>
> >> > >> Bokeh <http://bokeh.pydata.org/en/latest/> is an interactive
> >>
> >> plotting
> >>
> >> > >> library originally developed for python which uses HTML & Javascript
> >>
> >> as
> >>
> >> > >> it's backend to display and manipulate plots.
> >> > >>
> >> > >> Whether by using Bokeh or other libraries, web technologies are the
> >> > >> obvious option for Julia to get great visualization/graphics/UI
> >>
> >> without
> >>
> >> > >> the
> >> > >> pain.
> >> > >>
> >> > >> I suggest (and I assume I'm about to get shot down) that the Julia
> >> > >> community stops messing around with any OS specific graphics code
> >> > >> and
> >> > >> adopts HTML for all future visualizations. Are there any cases where
> >>
> >> that
> >>
> >> > >> wouldn't work?
