MNT - Refactor dist_fix_point function (scikit-learn-contrib#194)

Author: Badr-MOUFAD

skglm/solvers/anderson_cd.py

@@ -2,7 +2,9 @@
 from numba import njit
 from scipy import sparse
 from sklearn.utils import check_array
-from skglm.solvers.common import construct_grad, construct_grad_sparse, dist_fix_point
+from skglm.solvers.common import (
+    construct_grad, construct_grad_sparse, dist_fix_point_cd
+)
 from skglm.solvers.base import BaseSolver
 from skglm.utils.anderson import AndersonAcceleration
 
@@ -104,7 +106,7 @@ def solve(self, X, y, datafit, penalty, w_init=None, Xw_init=None):
             if self.ws_strategy == "subdiff":
                 opt = penalty.subdiff_distance(w[:n_features], grad, all_feats)
             elif self.ws_strategy == "fixpoint":
-                opt = dist_fix_point(
+                opt = dist_fix_point_cd(
                     w[:n_features], grad, lipschitz, datafit, penalty, all_feats
                 )
 
@@ -181,7 +183,7 @@ def solve(self, X, y, datafit, penalty, w_init=None, Xw_init=None):
                     if self.ws_strategy == "subdiff":
                         opt_ws = penalty.subdiff_distance(w[:n_features], grad_ws, ws)
                     elif self.ws_strategy == "fixpoint":
-                        opt_ws = dist_fix_point(
+                        opt_ws = dist_fix_point_cd(
                             w[:n_features], grad_ws, lipschitz, datafit, penalty, ws
                         )
 

skglm/solvers/common.py

@@ -3,7 +3,7 @@
 
 
 @njit
-def dist_fix_point(w, grad_ws, lipschitz, datafit, penalty, ws):
+def dist_fix_point_cd(w, grad_ws, lipschitz, datafit, penalty, ws):
     """Compute the violation of the fixed point iterate scheme.
 
     Parameters
@@ -28,16 +28,20 @@ def dist_fix_point(w, grad_ws, lipschitz, datafit, penalty, ws):
 
     Returns
     -------
-    dist_fix_point : array, shape (n_features,)
+    dist : array, shape (n_features,)
         Violation score for every feature.
     """
-    dist_fix_point = np.zeros(ws.shape[0])
+    dist = np.zeros(ws.shape[0])
+
     for idx, j in enumerate(ws):
-        lcj = lipschitz[j]
-        if lcj != 0:
-            dist_fix_point[idx] = np.abs(
-                w[j] - penalty.prox_1d(w[j] - grad_ws[idx] / lcj, 1. / lcj, j))
-    return dist_fix_point
+        if lipschitz[j] == 0.:
+            continue
+
+        step_j = 1 / lipschitz[j]
+        dist[idx] = np.abs(
+            w[j] - penalty.prox_1d(w[j] - step_j * grad_ws[idx], step_j, j)
+        )
+    return dist
 
 
 @njit

skglm/solvers/multitask_bcd.py

@@ -66,7 +66,7 @@ def solve(self, X, Y, datafit, penalty, W_init=None, XW_init=None):
             if self.ws_strategy == "subdiff":
                 opt = penalty.subdiff_distance(W, grad, all_feats)
             elif self.ws_strategy == "fixpoint":
-                opt = dist_fix_point(W, grad, datafit, penalty, all_feats)
+                opt = dist_fix_point_bcd(W, grad, datafit, penalty, all_feats)
             stop_crit = np.max(opt)
             if self.verbose:
                 print(f"Stopping criterion max violation: {stop_crit:.2e}")
@@ -150,7 +150,7 @@ def solve(self, X, Y, datafit, penalty, W_init=None, XW_init=None):
                     if self.ws_strategy == "subdiff":
                         opt_ws = penalty.subdiff_distance(W, grad_ws, ws)
                     elif self.ws_strategy == "fixpoint":
-                        opt_ws = dist_fix_point(
+                        opt_ws = dist_fix_point_bcd(
                             W, grad_ws, lipschitz, datafit, penalty, ws
                         )
 
@@ -231,7 +231,7 @@ def path(self, X, Y, datafit, penalty, alphas, W_init=None, return_n_iter=False)
 
 
 @njit
-def dist_fix_point(W, grad_ws, lipschitz, datafit, penalty, ws):
+def dist_fix_point_bcd(W, grad_ws, lipschitz, datafit, penalty, ws):
     """Compute the violation of the fixed point iterate schema.
 
     Parameters
@@ -256,17 +256,20 @@ def dist_fix_point(W, grad_ws, lipschitz, datafit, penalty, ws):
 
     Returns
     -------
-    dist_fix_point : array, shape (ws_size,)
+    dist : array, shape (ws_size,)
         Contain the violation score for every feature.
     """
-    dist_fix_point = np.zeros(ws.shape[0])
+    dist = np.zeros(ws.shape[0])
+
     for idx, j in enumerate(ws):
-        lcj = lipschitz[j]
-        if lcj:
-            dist_fix_point[idx] = norm(
-                W[j] - penalty.prox_1feat(W[j] - grad_ws[idx] / lcj, 1. / lcj, j)
-            )
-    return dist_fix_point
+        if lipschitz[j] == 0.:
+            continue
+
+        step_j = 1 / lipschitz[j]
+        dist[idx] = norm(
+            W[j] - penalty.prox_1feat(W[j] - step_j * grad_ws[idx], step_j, j)
+        )
+    return dist
 
 
 @njit