Merge branch 'dev'

Author: tlienart

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJLinearModels"
 uuid = "6ee0df7b-362f-4a72-a706-9e79364fb692"
 authors = ["Thibaut Lienart <tlienart@me.com>"]
-version = "0.8.1"
+version = "0.9.0"
 
 [deps]
 DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"
@@ -19,20 +19,4 @@ LinearMaps = "2.6, 3.2"
 MLJModelInterface = "0.3, 0.4, 1.0"
 Optim = "0.20, 0.21, 1"
 Parameters = "0.12"
-julia = "1.6"
-
-[extras]
-DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
-DelimitedFiles = "8bb1440f-4735-579b-a4ab-409b98df4dab"
-ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
-MLJ = "add582a8-e3ab-11e8-2d5e-e98b27df1bc7"
-MLJBase = "a7f614a8-145f-11e9-1d2a-a57a1082229d"
-PyCall = "438e738f-606a-5dbb-bf0a-cddfbfd45ab0"
-RCall = "6f49c342-dc21-5d91-9882-a32aef131414"
-RDatasets = "ce6b1742-4840-55fa-b093-852dadbb1d8b"
-Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
-StableRNGs = "860ef19b-820b-49d6-a774-d7a799459cd3"
-Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
-
-[targets]
-test = ["DelimitedFiles", "PyCall", "ForwardDiff", "Test", "Random", "RDatasets", "RCall", "MLJ", "MLJBase", "StableRNGs", "DataFrames"]
+julia = "1.6, 1"

docs/src/quickstart.md

@@ -150,3 +150,28 @@ ypred = predict_mode(mach, rows=1:2)
 ```
 
 Which, in my case, gives `setosa`, `setosa` (correct in both cases).
+
+### Customizing the solvers
+
+Depending on your data you may want to customize the default solver associated with your model. Since this package uses [Optim](https://julianlsolvers.github.io/Optim.jl/stable/) behind the scene, we can interact directly with this package.
+
+For instance, you may need to be more stringent about the convergence criterion of the LBFGS solver. This can be done by changing the general Optim `f_tol` parameter which defaults to ``10^{-4}``:
+
+```julia
+import Optim
+
+new_optim_options = Optim.Options(f_tol=1e-6)
+mdl = MultinomialClassifier(solver=LBFGS(optim_options=new_optim_options))
+mach = machine(mdl, X, y)
+fit!(mach)
+```
+
+You could also just try another solver:
+
+```julia
+mdl = MultinomialClassifier(solver=NewtonCG(optim_options=new_optim_options))
+mach = machine(mdl, X, y)
+fit!(mach)
+```
+
+For a full description of available solvers and API, see: [Solvers](@ref).

src/fit/solvers.jl

@@ -54,7 +54,7 @@ solver = MLJLinearModels.Newton(optim_options = Optim.Options(time_limit = 20),
 ```
 """
 @with_kw struct Newton{O,S} <: Solver
-    optim_options::O = Optim.Options()
+    optim_options::O = Optim.Options(f_tol=1e-4)
     newton_options::S = (; )
 end
 
@@ -79,7 +79,7 @@ solver = MLJLinearModels.Newton(optim_options = Optim.Options(time_limit = 20),
 
 """
 @with_kw struct NewtonCG{O,S} <: Solver
-    optim_options::O = Optim.Options()
+    optim_options::O = Optim.Options(f_tol=1e-4)
     newtoncg_options::S = (; )
 end
 
@@ -100,7 +100,7 @@ solver = MLJLinearModels.Newton(optim_options = Optim.Options(time_limit = 20),
 ```
 """
 @with_kw struct LBFGS{O,S} <: Solver
-    optim_options::O = Optim.Options()
+    optim_options::O = Optim.Options(f_tol=1e-4)
     lbfgs_options::S = (; )
 end
 

test/Project.toml

@@ -0,0 +1,24 @@
+[deps]
+DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
+DelimitedFiles = "8bb1440f-4735-579b-a4ab-409b98df4dab"
+ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
+LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+MLJ = "add582a8-e3ab-11e8-2d5e-e98b27df1bc7"
+MLJBase = "a7f614a8-145f-11e9-1d2a-a57a1082229d"
+Optim = "429524aa-4258-5aef-a3af-852621145aeb"
+PyCall = "438e738f-606a-5dbb-bf0a-cddfbfd45ab0"
+RCall = "6f49c342-dc21-5d91-9882-a32aef131414"
+RDatasets = "ce6b1742-4840-55fa-b093-852dadbb1d8b"
+Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+StableRNGs = "860ef19b-820b-49d6-a774-d7a799459cd3"
+Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+
+[compat]
+DataFrames = "1.4"
+ForwardDiff = "0.10"
+MLJ = "0.19"
+MLJBase = "0.21"
+PyCall = "1.95"
+RCall = "0.13"
+RDatasets = "0.7"
+StableRNGs = "1.0"

test/docs/Project.toml

@@ -78,7 +78,7 @@ end
     @test isapprox(J(θ1),         7.359476, rtol=1e-5)
     @test isapprox(J(θ_newton),   0.486388, rtol=1e-5)
     @test isapprox(J(θ_newtoncg), 0.486388, rtol=1e-5)
-    @test isapprox(J(θ_lbfgs),    0.486388, rtol=1e-5)
+    @test isapprox(J(θ_lbfgs),    0.486788, rtol=1e-5)
     @test isapprox(J(θ_iwls),     0.486388, rtol=1e-5)
 end
 

test/fit/robust.jl

@@ -1,5 +1,6 @@
 using MLJLinearModels, Test, LinearAlgebra
 using Random, StableRNGs, DataFrames, ForwardDiff
+import Optim
 import MLJ, MLJBase
 
 DO_COMPARISONS = false; include("testutils.jl")