MNT - split and organize utils.py (#123)

Badr-MOUFAD · web-flow · commit 5762352ac821 · 2022-11-18T15:10:33.000+01:00
diff --git a/examples/plot_lasso_vs_weighted.py b/examples/plot_lasso_vs_weighted.py
@@ -13,7 +13,7 @@
 import matplotlib.pyplot as plt
 
 from skglm import Lasso, WeightedLasso
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 
 n_features = 30
 X, _, _ = make_correlated_data(
diff --git a/examples/plot_logreg_various_penalties.py b/examples/plot_logreg_various_penalties.py
@@ -16,7 +16,7 @@
 from skglm import GeneralizedLinearEstimator
 from skglm.datafits import Logistic
 from skglm.penalties import L1_plus_L2, MCPenalty
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 
 
 n_samples, n_features = 50, 100
diff --git a/examples/plot_reweighted_l1.py b/examples/plot_reweighted_l1.py
@@ -14,7 +14,7 @@
 import matplotlib.pyplot as plt
 
 from skglm.penalties.separable import L0_5
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 from skglm.estimators import GeneralizedLinearEstimator
 from skglm.experimental import IterativeReweightedL1
 from skglm.solvers import AndersonCD
diff --git a/examples/plot_sparse_recovery.py b/examples/plot_sparse_recovery.py
@@ -15,10 +15,10 @@
 from sklearn.model_selection import train_test_split
 from sklearn.metrics import f1_score, mean_squared_error
 
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 from skglm.solvers import AndersonCD
 from skglm.datafits import Quadratic
-from skglm.utils import compiled_clone
+from skglm.utils.jit_compilation import compiled_clone
 from skglm.penalties import L1, MCPenalty, L0_5, L2_3, SCAD
 
 cmap = plt.get_cmap('tab10')
diff --git a/examples/plot_zero_weights_lasso.py b/examples/plot_zero_weights_lasso.py
@@ -12,7 +12,7 @@
 import matplotlib.pyplot as plt
 
 from skglm import WeightedLasso
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 
 n_features = 100
 w_true = np.zeros(n_features)
diff --git a/skglm/datafits/single_task.py b/skglm/datafits/single_task.py
@@ -4,7 +4,7 @@
 from numba import float64
 
 from skglm.datafits.base import BaseDatafit
-from skglm.utils import spectral_norm
+from skglm.utils.sparse_ops import spectral_norm
 
 
 class Quadratic(BaseDatafit):
diff --git a/skglm/estimators.py b/skglm/estimators.py
@@ -16,7 +16,7 @@
 from sklearn.preprocessing import LabelEncoder
 from sklearn.multiclass import OneVsRestClassifier, check_classification_targets
 
-from skglm.utils import compiled_clone
+from skglm.utils.jit_compilation import compiled_clone
 from skglm.solvers import AndersonCD, MultiTaskBCD
 from skglm.datafits import Quadratic, Logistic, QuadraticSVC, QuadraticMultiTask
 from skglm.penalties import L1, WeightedL1, L1_plus_L2, MCPenalty, IndicatorBox, L2_1
diff --git a/skglm/experimental/_plot_sqrt_lasso.py b/skglm/experimental/_plot_sqrt_lasso.py
@@ -2,7 +2,7 @@
 import numpy as np
 from numpy.linalg import norm
 import matplotlib.pyplot as plt
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 from skglm.experimental.sqrt_lasso import SqrtLasso, _chambolle_pock_sqrt
 
 X, y, _ = make_correlated_data(n_samples=200, n_features=100, random_state=24)
diff --git a/skglm/experimental/reweighted.py b/skglm/experimental/reweighted.py
@@ -2,7 +2,7 @@
 from skglm.datafits import Quadratic
 from skglm.estimators import GeneralizedLinearEstimator
 from skglm.penalties import WeightedL1, L0_5
-from skglm.utils import compiled_clone
+from skglm.utils.jit_compilation import compiled_clone
 
 
 class IterativeReweightedL1(GeneralizedLinearEstimator):
diff --git a/skglm/experimental/sqrt_lasso.py b/skglm/experimental/sqrt_lasso.py
@@ -5,7 +5,8 @@
 from sklearn.linear_model._base import LinearModel, RegressorMixin
 
 from skglm.penalties import L1
-from skglm.utils import compiled_clone, ST_vec, proj_L2ball
+from skglm.utils.prox_funcs import ST_vec, proj_L2ball
+from skglm.utils.jit_compilation import compiled_clone
 from skglm.datafits.base import BaseDatafit
 from skglm.solvers.prox_newton import ProxNewton
 
diff --git a/skglm/experimental/tests/test_reweighted.py b/skglm/experimental/tests/test_reweighted.py
@@ -2,7 +2,7 @@
 from numpy.linalg import norm
 
 from skglm.penalties.separable import L0_5
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 from skglm.experimental import IterativeReweightedL1
 from skglm.solvers import AndersonCD
 
diff --git a/skglm/experimental/tests/test_sqrt_lasso.py b/skglm/experimental/tests/test_sqrt_lasso.py
@@ -2,7 +2,7 @@
 import numpy as np
 from numpy.linalg import norm
 
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 from skglm.experimental.sqrt_lasso import SqrtLasso, _chambolle_pock_sqrt
 
 
diff --git a/skglm/penalties/block_separable.py b/skglm/penalties/block_separable.py
@@ -4,7 +4,7 @@
 from numba import float64, int32
 
 from skglm.penalties.base import BasePenalty
-from skglm.utils import (
+from skglm.utils.prox_funcs import (
     BST, prox_block_2_05, prox_SCAD, value_SCAD, prox_MCP, value_MCP)
 
 
diff --git a/skglm/penalties/non_separable.py b/skglm/penalties/non_separable.py
@@ -2,7 +2,7 @@
 from numba import float64
 
 from skglm.penalties.base import BasePenalty
-from skglm.utils import prox_SLOPE
+from skglm.utils.prox_funcs import prox_SLOPE
 
 
 class SLOPE(BasePenalty):
diff --git a/skglm/penalties/separable.py b/skglm/penalties/separable.py
@@ -3,7 +3,7 @@
 from numba.types import bool_
 
 from skglm.penalties.base import BasePenalty
-from skglm.utils import (
+from skglm.utils.prox_funcs import (
     ST, box_proj, prox_05, prox_2_3, prox_SCAD, value_SCAD, prox_MCP, value_MCP)
 
 
diff --git a/skglm/solvers/anderson_cd.py b/skglm/solvers/anderson_cd.py
@@ -4,7 +4,7 @@
 from sklearn.utils import check_array
 from skglm.solvers.common import construct_grad, construct_grad_sparse, dist_fix_point
 from skglm.solvers.base import BaseSolver
-from skglm.utils import AndersonAcceleration
+from skglm.utils.anderson import AndersonAcceleration
 
 
 class AndersonCD(BaseSolver):
diff --git a/skglm/solvers/fista.py b/skglm/solvers/fista.py
@@ -2,7 +2,7 @@
 from scipy.sparse import issparse
 from skglm.solvers.base import BaseSolver
 from skglm.solvers.common import construct_grad, construct_grad_sparse
-from skglm.utils import _prox_vec
+from skglm.utils.prox_funcs import _prox_vec
 
 
 class FISTA(BaseSolver):
diff --git a/skglm/solvers/gram_cd.py b/skglm/solvers/gram_cd.py
@@ -3,7 +3,7 @@
 from numba import njit
 from scipy.sparse import issparse
 from skglm.solvers.base import BaseSolver
-from skglm.utils import AndersonAcceleration
+from skglm.utils.anderson import AndersonAcceleration
 
 
 class GramCD(BaseSolver):
diff --git a/skglm/solvers/group_bcd.py b/skglm/solvers/group_bcd.py
@@ -1,7 +1,9 @@
 import numpy as np
 from numba import njit
+
 from skglm.solvers.base import BaseSolver
-from skglm.utils import AndersonAcceleration, check_group_compatible
+from skglm.utils.anderson import AndersonAcceleration
+from skglm.utils.validation import check_group_compatible
 
 
 class GroupBCD(BaseSolver):
diff --git a/skglm/solvers/group_prox_newton.py b/skglm/solvers/group_prox_newton.py
@@ -2,7 +2,7 @@
 from numba import njit
 from numpy.linalg import norm
 from skglm.solvers.base import BaseSolver
-from skglm.utils import check_group_compatible
+from skglm.utils.validation import check_group_compatible
 
 EPS_TOL = 0.3
 MAX_CD_ITER = 20
diff --git a/skglm/tests/test_datafits.py b/skglm/tests/test_datafits.py
@@ -8,7 +8,7 @@
 from skglm.penalties import L1, WeightedL1
 from skglm.solvers import AndersonCD, ProxNewton
 from skglm import GeneralizedLinearEstimator
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 
 
 @pytest.mark.parametrize('fit_intercept', [False, True])
diff --git a/skglm/tests/test_estimators.py b/skglm/tests/test_estimators.py
@@ -16,7 +16,7 @@
 
 from scipy.sparse import csc_matrix, issparse
 
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 from skglm.estimators import (
     GeneralizedLinearEstimator, Lasso, MultiTaskLasso, WeightedLasso, ElasticNet,
     MCPRegression, SparseLogisticRegression, LinearSVC)
diff --git a/skglm/tests/test_fista.py b/skglm/tests/test_fista.py
@@ -10,7 +10,10 @@
 from skglm.penalties import L1, IndicatorBox
 from skglm.solvers import FISTA, AndersonCD
 from skglm.datafits import Quadratic, Logistic, QuadraticSVC
-from skglm.utils import make_correlated_data, compiled_clone, spectral_norm
+
+from skglm.utils.sparse_ops import spectral_norm
+from skglm.utils.data import make_correlated_data
+from skglm.utils.jit_compilation import compiled_clone
 
 
 random_state = 113
diff --git a/skglm/tests/test_gram_solver.py b/skglm/tests/test_gram_solver.py
@@ -7,7 +7,9 @@
 
 from skglm.penalties import L1
 from skglm.solvers import GramCD
-from skglm.utils import make_correlated_data, compiled_clone
+
+from skglm.utils.data import make_correlated_data
+from skglm.utils.jit_compilation import compiled_clone
 
 
 @pytest.mark.parametrize("rho, X_density, greedy_cd",
diff --git a/skglm/tests/test_group.py b/skglm/tests/test_group.py
@@ -9,9 +9,12 @@
 from skglm.penalties.block_separable import WeightedGroupL2
 from skglm.datafits.group import QuadraticGroup, LogisticGroup
 from skglm.solvers import GroupBCD, GroupProxNewton
-from skglm.utils import (
-    _alpha_max_group_lasso, grp_converter, make_correlated_data, compiled_clone,
-    AndersonAcceleration)
+
+from skglm.utils.anderson import AndersonAcceleration
+from skglm.utils.jit_compilation import compiled_clone
+from skglm.utils.data import (make_correlated_data, grp_converter,
+                              _alpha_max_group_lasso)
+
 from celer import GroupLasso, Lasso
 from sklearn.linear_model import LogisticRegression
 
diff --git a/skglm/tests/test_penalties.py b/skglm/tests/test_penalties.py
@@ -10,7 +10,7 @@
     L2_1, L2_05, BlockMCPenalty, BlockSCAD)
 from skglm import GeneralizedLinearEstimator, Lasso
 from skglm.solvers import AndersonCD, MultiTaskBCD, FISTA
-from skglm.utils import make_correlated_data
+from skglm.utils.data import make_correlated_data
 
 
 n_samples = 20
diff --git a/skglm/tests/test_prox_newton.py b/skglm/tests/test_prox_newton.py
@@ -6,7 +6,9 @@
 from skglm.penalties import L1
 from skglm.datafits import Logistic
 from skglm.solvers.prox_newton import ProxNewton
-from skglm.utils import make_correlated_data, compiled_clone
+
+from skglm.utils.jit_compilation import compiled_clone
+from skglm.utils.data import make_correlated_data
 
 
 @pytest.mark.parametrize("X_density, fit_intercept", product([1., 0.5], [True, False]))
diff --git a/skglm/utils.py b/skglm/utils.py
diff --git a/skglm/utils/__init__.py b/skglm/utils/__init__.py
diff --git a/skglm/utils/anderson.py b/skglm/utils/anderson.py
diff --git a/skglm/utils/data.py b/skglm/utils/data.py
diff --git a/skglm/utils/jit_compilation.py b/skglm/utils/jit_compilation.py
diff --git a/skglm/utils/prox_funcs.py b/skglm/utils/prox_funcs.py
diff --git a/skglm/utils/sparse_ops.py b/skglm/utils/sparse_ops.py
diff --git a/skglm/utils/validation.py b/skglm/utils/validation.py
