ENH add Lasso.jl julia solver

.github/workflows/main.yml

@@ -62,9 +62,15 @@ jobs:
         branch=${BENCHOPT_BRANCH##*:}
         pip install -U git+https://github.com/$user/benchopt@$branch
 
+    - name: Disable Julia multi-threading on MacOS
+      if: matrix.os == 'macos-latest'
+      run: |
+        export JULIA_NUM_THREADS=1
+
     - name: Test
       run: |
         export OMP_NUM_THREADS=1
+        export JULIA_NUM_THREADS=1
         benchopt test . --env-name bench_test_env -vl
         benchopt test . --env-name bench_test_env -vl --skip-install
 

solvers/lasso_jl.jl

@@ -0,0 +1,51 @@
+using Lasso
+using PyCall
+using SparseArrays
+
+function scipyCSC_to_julia(A)
+    m, n = A.shape
+    colPtr = Int[i + 1 for i in PyArray(A."indptr")]
+    rowVal = Int[i + 1 for i in PyArray(A."indices")]
+    nzVal = Vector{Float64}(PyArray(A."data"))
+    B = SparseMatrixCSC{Float64,Int}(m, n, colPtr, rowVal, nzVal)
+
+    return B
+end
+
+function solve_lasso(
+    X,
+    y::Vector{Float64},
+    lambda::Vector{Float64},
+    fit_intercept::Bool,
+    tol::Float64,
+    cd_maxiter::Int,
+    get_null_solution::Bool,
+)
+    p = size(X, 2)
+
+    w = if fit_intercept zeros(Float64, p + 1) else zeros(Float64, p) end
+
+    converged = true
+
+    # TODO(jolars): once https://github.com/JuliaStats/Lasso.jl/issues/70 or
+    # maybe https://github.com/JuliaStats/Lasso.jl/issues/71 is/are resolved,
+    # we should not need the try-catch here
+    if !get_null_solution
+        try
+            lasso_fit = fit(
+                LassoPath,
+                X,
+                y;
+                λ=lambda, standardize=false, intercept=fit_intercept,
+                maxncoef=max(size(X, 1), size(X, 2)) * 100,
+                cd_maxiter=cd_maxiter,
+                cd_tol=tol,
+            )
+            w = coef(lasso_fit)
+        catch
+            converged = false
+        end
+    end
+
+    return w, converged
+end

solvers/lasso_jl.py

@@ -0,0 +1,89 @@
+from benchopt import safe_import_context
+from benchopt.helpers.julia import (
+    JuliaSolver,
+    assert_julia_installed,
+    get_jl_interpreter,
+)
+from benchopt.runner import INFINITY
+from benchopt.stopping_criterion import SufficientProgressCriterion
+
+with safe_import_context() as import_ctx:
+    assert_julia_installed()
+    import warnings
+    import numpy as np
+    from scipy import sparse
+
+    from pathlib import Path
+    # File containing the function to be called from julia
+    JULIA_SOLVER_FILE = str(Path(__file__).with_suffix(".jl"))
+
+
+class Solver(JuliaSolver):
+    name = "lasso_jl"
+    stopping_criterion = SufficientProgressCriterion(
+        patience=7, eps=1e-15, strategy="tolerance"
+    )
+    julia_requirements = [
+        "Lasso",
+        "PyCall",
+        "SparseArrays",
+    ]
+    references = [
+        'J. Friedman, T. J. Hastie and R. Tibshirani, "Regularization paths '
+        'for generalized linear models via coordinate descent", '
+        "J. Stat. Softw., vol. 33, no. 1, pp. 1-22, NIH Public Access (2010)"
+    ]
+
+    def set_objective(self, X, y, lmbd, fit_intercept):
+        self.n, self.p = X.shape
+        self.X = X
+        self.y = y
+        self.lmbd = np.array([lmbd])
+        self.fit_intercept = fit_intercept
+
+        jl = get_jl_interpreter()
+        jl.include(str(JULIA_SOLVER_FILE))
+        self.solve_lasso = jl.solve_lasso
+
+        if sparse.issparse(X):
+            scipyCSC_to_julia = jl.pyfunctionret(
+                jl.scipyCSC_to_julia, jl.Any, jl.PyObject
+            )
+            self.X = scipyCSC_to_julia(X)
+
+        # Trigger Julia JIT compilation
+        w_dim = self.p + 1 if fit_intercept else self.p
+        self.prev_solution = np.zeros(w_dim)
+
+        warnings.filterwarnings("ignore", category=FutureWarning)
+        self.run(1e-2)
+
+    def run(self, tol):
+        # remove possibly spurious warnings from pyjulia
+        # TODO: remove filter when
+        # https://github.com/JuliaPy/pyjulia/issues/497 is resolved or
+        # otherwise fix the warning
+        warnings.filterwarnings("ignore", category=FutureWarning)
+
+        coefs, converged = self.solve_lasso(
+            self.X,
+            self.y,
+            self.lmbd / len(self.y),
+            self.fit_intercept,
+            tol**1.8,
+            1_000_000,
+            tol == INFINITY,
+        )
+        if converged:
+            self.coefs = coefs
+            self.prev_solution = coefs
+        else:
+            self.coefs = self.prev_solution
+
+    def get_result(self):
+        coefs = np.ravel(self.coefs)
+
+        if self.fit_intercept:
+            coefs = np.hstack((coefs[1:], coefs[0]))
+
+        return coefs

test_config.py

@@ -15,6 +15,10 @@ def check_test_solver_install(solver_class):
     if 'julia' in solver_class.name.lower():
         pytest.xfail('Julia install from conda fails currently.')
 
+    # Julia segfauls on OSX
+    if solver_class.name.lower() == 'lasso_jl' and sys.platform == 'darwin':
+        pytest.xfail('Julia solvers segfauls on OSX.')
+
     # ModOpt install change numpy version, breaking celer install.
     # See CEA-COSMIC/ModOpt#144. Skipping for now
     if ('modopt' in solver_class.name.lower()):