Add support for LogisticRegression

README.md

@@ -24,25 +24,29 @@ This project is inspired by two existing projects:
 
 ## Supported models
 
-Use `add_predictor` to add a model.
+Use `add_predictor`:
 ```julia
-Omelette.add_predictor(model, model_ml, x, y)
-y = Omelette.add_predictor(model, model_ml, x)
+y = Omelette.add_predictor(model, predictor, x)
+```
+or:
+```julia
+Omelette.add_predictor!(model, predictor, x, y)
 ```
 
 ### LinearRegression
 
 ```julia
-num_features, num_observations = 2, 10
-X = rand(num_observations, num_features)
-θ = rand(num_features)
-Y = X * θ + randn(num_observations)
+using Omelette, GLM
+X, Y = rand(10, 2), rand(10)
 model_glm = GLM.lm(X, Y)
 predictor = Omelette.LinearRegression(model_glm)
-model = Model(HiGHS.Optimizer)
-set_silent(model)
-@variable(model, 0 <= x[1:num_features] <= 1)
-@constraint(model, sum(x) == 1.5)
-y = Omelette.add_predictor(model, predictor, x)
-@objective(model, Max, y[1])
+```
+
+### LogisticRegression
+
+```julia
+using Omelette, GLM
+X, Y = rand(10, 2), rand(Bool, 10)
+model_glm = GLM.glm(X, Y, GLM.Bernoulli())
+predictor = Omelette.LogisticRegression(model_glm)
 ```

ext/OmeletteGLMExt.jl

@@ -12,4 +12,12 @@ function Omelette.LinearRegression(model::GLM.LinearModel)
     return Omelette.LinearRegression(GLM.coef(model))
 end
 
+function Omelette.LogisticRegression(
+    model::GLM.GeneralizedLinearModel{
+        GLM.GlmResp{Vector{Float64},GLM.Bernoulli{Float64},GLM.LogitLink},
+    },
+)
+    return Omelette.LogisticRegression(GLM.coef(model))
+end
+
 end #module

src/models/LogisticRegression.jl

@@ -0,0 +1,24 @@
+# Copyright (c) 2024: Oscar Dowson and contributors
+#
+# Use of this source code is governed by an MIT-style license that can be found
+# in the LICENSE.md file or at https://opensource.org/licenses/MIT.
+
+struct LogisticRegression <: AbstractPredictor
+    parameters::Matrix{Float64}
+end
+
+function LogisticRegression(parameters::Vector{Float64})
+    return LogisticRegression(reshape(parameters, 1, length(parameters)))
+end
+
+Base.size(f::LogisticRegression) = size(f.parameters)
+
+function _add_predictor_inner(
+    model::JuMP.Model,
+    predictor::LogisticRegression,
+    x::Vector{JuMP.VariableRef},
+    y::Vector{JuMP.VariableRef},
+)
+    JuMP.@constraint(model, 1 ./ (1 .+ exp.(-predictor.parameters * x)) .== y)
+    return
+end

test/Project.toml

@@ -1,13 +1,15 @@
 [deps]
 GLM = "38e38edf-8417-5370-95a0-9cbb8c7f171a"
 HiGHS = "87dc4568-4c63-4d18-b0c0-bb2238e4078b"
+Ipopt = "b6b21f68-93f8-5de0-b562-5493be1d77c9"
 JuMP = "4076af6c-e467-56ae-b986-b466b2749572"
 Omelette = "e52c2cb8-508e-4e12-9dd2-9c4755b60e73"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [compat]
 GLM = "1"
 HiGHS = "1"
+Ipopt = "1"
 JuMP = "1"
 Test = "<0.0.1, 1.6"
 julia = "1.9"

test/test_LogisticRegression.jl

@@ -0,0 +1,75 @@
+# Copyright (c) 2024: Oscar Dowson and contributors
+#
+# Use of this source code is governed by an MIT-style license that can be found
+# in the LICENSE.md file or at https://opensource.org/licenses/MIT.
+
+module LogisticRegressionTests
+
+using JuMP
+using Test
+
+import GLM
+import Ipopt
+import Omelette
+
+is_test(x) = startswith(string(x), "test_")
+
+function runtests()
+    @testset "$name" for name in filter(is_test, names(@__MODULE__; all = true))
+        getfield(@__MODULE__, name)()
+    end
+    return
+end
+
+function test_LogisticRegression()
+    model = Model()
+    @variable(model, x[1:2])
+    @variable(model, y[1:1])
+    f = Omelette.LogisticRegression([2.0, 3.0])
+    Omelette.add_predictor!(model, f, x, y)
+    cons = all_constraints(model; include_variable_in_set_constraints = false)
+    obj = constraint_object(only(cons))
+    @test obj.set == MOI.EqualTo(0.0)
+    g = 1.0 / (1.0 + exp(-2.0 * x[1] - 3.0 * x[2])) - y[1]
+    @test isequal_canonical(obj.func, g)
+    return
+end
+
+function test_LogisticRegression_dimension_mismatch()
+    model = Model()
+    @variable(model, x[1:3])
+    @variable(model, y[1:2])
+    f = Omelette.LogisticRegression([2.0, 3.0])
+    @test size(f) == (1, 2)
+    @test_throws DimensionMismatch Omelette.add_predictor!(model, f, x, y[1:1])
+    @test_throws DimensionMismatch Omelette.add_predictor!(model, f, x[1:2], y)
+    g = Omelette.LogisticRegression([2.0 3.0; 4.0 5.0; 6.0 7.0])
+    @test size(g) == (3, 2)
+    @test_throws DimensionMismatch Omelette.add_predictor!(model, g, x, y)
+    return
+end
+
+function test_LogisticRegression_GLM()
+    num_features = 2
+    num_observations = 10
+    X = rand(num_observations, num_features)
+    θ = rand(num_features)
+    Y = X * θ + randn(num_observations) .>= 0
+    model_glm = GLM.glm(X, Y, GLM.Bernoulli())
+    model = Model(Ipopt.Optimizer)
+    set_silent(model)
+    model_ml = Omelette.LogisticRegression(model_glm)
+    @variable(model, 0 <= x[1:num_features] <= 1)
+    @constraint(model, sum(x) == 1.5)
+    y = Omelette.add_predictor(model, model_ml, x)
+    @objective(model, Max, only(y))
+    optimize!(model)
+    @assert is_solved_and_feasible(model)
+    y_star_glm = GLM.predict(model_glm, value.(x)')
+    @test isapprox(objective_value(model), y_star_glm; atol = 1e-6)
+    return
+end
+
+end
+
+LogisticRegressionTests.runtests()