Adaptive Probability & Coordinate Importance

Adaptive Probability & Coordinate Importance#

Results are shown below. Again similar stuff to what was observed with just adaptive probability.

using DrWatson
quickactivate("../../AndersonStochastic")

  Activating 

project at `~/Documents/ClimateModelling/AndersonStochastic`

# Here you may include files from the source directory
include(srcdir("init.jl"))
include(srcdir("plotting.jl"))

println(
"""
Currently active project is: $(projectname())

Path of active project: $(projectdir())
"""
)

The WebIO Jupyter extension was not detected. See the WebIO Jupyter integration documentation for more information. 

Currently active project is: AndersonStochastic

Path of active project: /home/bretschneide/Documents/ClimateModelling/AndersonStochastic

using PlotlyJS  # For interactive HTML plotting

dff = collect_results(datadir("Oct1"));
dff[!, :method] .= ifelse.(((coalesce.(dff[!, :coordsample], 1.0) .== 1.0) .& (dff[!, :method] .== :ci)), :vanilla, dff[!, :method]);

[ Info: Scanning folder /home/bretschneide/Documents/ClimateModelling/AndersonStochastic/data/Oct1 for result files.
[ Info: Added 291 entries.

For 1e-5#

vanilladf = dff[(dff.scalardiffusivity .== 1e-5) .& (dff.method .== :vanilla), :];
sort!(vanilladf,:m);
apcidf = dff[(dff.scalardiffusivity .== 1e-5) .& (dff.method .== :apci), :];
sort!(apcidf,:probdecrease)
dataframesmall = vcat(vanilladf,apcidf);

plotmake(dataframesmall,[[i for i in 1:25],[25+i for i in 1:20],[1,2,3,25,28],[1,2,3,24,41],[28,33,38,43],[26,27,28,29,30]],"_static/plot.html")

"_static/plot.html"

html_content = """
<iframe src="_static/plot.html" width="100%" height="600"></iframe>
"""
display("text/html", html_content)

Again;

  • Slide 1; just how various vanilla methods perform. Best is again infinite history.

  • Slide 2; All the various APCI parameter combinations - bit of a mess.

  • Slide 3; Best APCI clearly better than best vanilla.

  • Slide 4; There are also some very bad choices of parameters for APCI.

  • Slide 5; Trends as ‘probdec’ is changed.

  • Slide 6; Trends as ‘%coords’ is changed. Apart from clearly bad choices all perform very similarly.

Conclusions#

APCI does work.

Contents