WIP + RFC. Add simple plotting recipe. Fix #85

[mkborregaard]

No description provided.

[mkborregaard]

A simple recipe for plotting. Right now it's the default plot for tuned models, that can be changed if that is not desired. It uses colored scatter points. There's an option for using heatmap also, but that currently behaves strangely for log scales. That seems to be a bug in Plots though, I think the recipe here is fine.

Using the tuned_ensemble object generated from here, you can do:

plot(tuned_ensemble)

The heatmap version (you can also use surface or contour):

heatmap(tuned_model)

[ablaom]

This looks super.

I'm having some problems though. Any ideas?

I'm using Plots.jl v0.23.1

julia> using Plots
julia> Plots.backend()
Plots.GRBackend()

julia> include("examples/two_parameter_tune.jl")
julia> tuned_ensemble isa MLJ.Machine{<:MLJ.EitherTunedModel}
true

julia> plot(tuned_ensemble)
ERROR: No user recipe defined for Machine{MLJ.DeterministicTunedModel{Grid,DeterministicEnsembleModel{Tuple{Array{Float64,2},Array{Float64,1}},KNNRegressor{Float64}}}}
Stacktrace:
 [1] error(::String) at ./error.jl:33
 [2] convertToAnyVector(::Machine{MLJ.DeterministicTunedModel{Grid,DeterministicEnsembleModel{Tuple{Array{Float64,2},Array{Float64,1}},KNNRegressor{Float64}}}}, ::Dict{Symbol,Any}) at /Users/anthony/.julia/packages/Plots/gylTh/src/series.jl:14
 [3] macro expansion at /Users/anthony/.julia/packages/Plots/gylTh/src/series.jl:139 [inlined]
 [4] apply_recipe(::Dict{Symbol,Any}, ::Type{Plots.SliceIt}, ::Nothing, ::Machine{MLJ.DeterministicTunedModel{Grid,DeterministicEnsembleModel{Tuple{Array{Float64,2},Array{Float64,1}},KNNRegressor{Float64}}}}, ::Nothing) at /Users/anthony/.julia/packages/RecipesBase/Uz5AO/src/RecipesBase.jl:275
 [5] _process_userrecipes(::Plots.Plot{Plots.GRBackend}, ::Dict{Symbol,Any}, ::Tuple{Machine{MLJ.DeterministicTunedModel{Grid,DeterministicEnsembleModel{Tuple{Array{Float64,2},Array{Float64,1}},KNNRegressor{Float64}}}}}) at /Users/anthony/.julia/packages/Plots/gylTh/src/pipeline.jl:83
 [6] _plot!(::Plots.Plot{Plots.GRBackend}, ::Dict{Symbol,Any}, ::Tuple{Machine{MLJ.DeterministicTunedModel{Grid,DeterministicEnsembleModel{Tuple{Array{Float64,2},Array{Float64,1}},KNNRegressor{Float64}}}}}) at /Users/anthony/.julia/packages/Plots/gylTh/src/plot.jl:178
 [7] #plot#136(::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::Function, ::Machine{MLJ.DeterministicTunedModel{Grid,DeterministicEnsembleModel{Tuple{Array{Float64,2},Array{Float64,1}},KNNRegressor{Float64}}}}) at /Users/anthony/.julia/packages/Plots/gylTh/src/plot.jl:57
 [8] plot(::Machine{MLJ.DeterministicTunedModel{Grid,DeterministicEnsembleModel{Tuple{Array{Float64,2},Array{Float64,1}},KNNRegressor{Float64}}}}) at /Users/anthony/.julia/packages/Plots/gylTh/src/plot.jl:51
 [9] top-level scope at none:0

julia> versioninfo()
Julia Version 1.0.3
Commit 099e826241 (2018-12-18 01:34 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin14.5.0)
  CPU: Intel(R) Core(TM) i5-3427U CPU @ 1.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.0 (ORCJIT, ivybridge)
Environment:
  JULIA_PATH = /Applications/Julia-1.0.app/Contents/Resources/julia/bin/julia

Edit Maybe add above code as test and see if Travis also dies?

[mkborregaard]

Sure I've added the suggested test, though this adds Plots as a testing dependency.
It looks like you may just not have the right branch checked out though.

[mkborregaard]

Seems to work for Travis.

[ablaom]

Yes, my mistake. In a recent rebuild of my working env I had done add instead of dev of the MLJ repos. It works great.

I'd like to pull this before adding a 1-parameter version (there was some discussion of adding uncertainty estimates in this case). @mkborregaard Could you please:

• add a check on the number of parameters and do error("Only two-parameter tuning plots supported. ") if the number is not two?

• remove the test and Plots dependency

@fkiraly Are you happy for me to pull this?

[mkborregaard]

I can - just to make sure, the test doesn't add Plots as a dependency, just as a test dependency. But you'd still like me to remove it again?

[ablaom]

Yes, let's leave it out for now thanks.