Moving iterative parameter to a parameter to allow for a version of C…

…UR where you compute the pi score every n iterations, where if n=0 pi is never recomputed, and the default is n=1

docs/source/selection.rst

@@ -102,8 +102,8 @@ They are instantiated using
                         # int, number of eigenvectors to use in computing pi
                         k = 1,
 
-                        # boolean, whether to orthogonalize after each selection, defaults to true
-                        iterative = True,
+                        # int, number of steps after which to recompute pi
+                        recompute_every = 1,
 
                         # float, threshold below which scores will be considered 0, defaults to 1E-12
                         tolerance=1E-12,
@@ -149,8 +149,8 @@ and are instantiated using
                         # int, number of eigenvectors to use in computing pi
                         k = 1,
 
-                        # boolean, whether to orthogonalize after each selection, defaults to true
-                        iterative = True,
+                        # int, number of steps after which to recompute pi
+                        recompute_every = 1,
 
                         # float, threshold below which scores will be considered 0, defaults to 1E-12
                         tolerance=1E-12,

examples/FeatureSelection.ipynb

@@ -178,7 +178,7 @@
     "\n",
     "\n",
     "idx = PCovCUR(mixing=m, n_to_select=n).fit(X, y).selected_idx_\n",
-    "idx_non_it = PCovCUR(mixing=m, iterative=False, n_to_select=n).fit(X, y).selected_idx_\n",
+    "idx_non_it = PCovCUR(mixing=m, recompute_every=0, n_to_select=n).fit(X, y).selected_idx_\n",
     "\n",
     "plt.loglog(\n",
     "    range(1, n + 1),\n",

skcosmo/_selection.py

@@ -422,8 +422,9 @@ class _CUR(GreedySelector):
 
     Parameters
     ----------
-    iterative : bool
-                whether to orthogonalize after each selection, defaults to `true`
+    recompute_every : int
+                      number of steps after which to recompute the pi score
+                      defaults to 1, if 0 no re-computation is done
 
     k : int
         number of eigenvectors to compute the importance score with, defaults to 1
@@ -443,7 +444,7 @@ class _CUR(GreedySelector):
     def __init__(
         self,
         selection_type,
-        iterative=True,
+        recompute_every=1,
         k=1,
         tolerance=1e-12,
         n_to_select=None,
@@ -454,8 +455,8 @@ def __init__(
     ):
 
         self.k = k
-        self.iterative = iterative
         self.tolerance = tolerance
+        self.recompute_every = recompute_every
 
         super().__init__(
             selection_type=selection_type,
@@ -510,7 +511,7 @@ def _continue_greedy_search(self, X, y, n_to_select):
 
         for c in self.selected_idx_:
 
-            if (
+            if self.recompute_every != 0 and (
                 np.linalg.norm(np.take(self.X_current_, [c], axis=self._axis))
                 > self.tolerance
             ):
@@ -570,13 +571,15 @@ def _compute_pi(self, X, y=None):
     def _update_post_selection(self, X, y, last_selected):
         """
         Saves the most recently selected feature, increments the feature counter,
-        and, if the CUR is iterative, orthogonalizes the remaining features by
+        and, if the CUR is iterative (recompute_every>0), orthogonalizes the remaining features by
         the most recently selected.
         """
         super()._update_post_selection(X, y, last_selected)
 
-        if self.iterative:
+        if self.recompute_every != 0:
             self._orthogonalize(last_selected)
+
+        if len(self.selected_idx_) % self.recompute_every == 0:
             self.pi_ = self._compute_pi(self.X_current_)
 
         self.pi_[last_selected] = 0.0
@@ -605,8 +608,9 @@ class _PCovCUR(GreedySelector):
 
     Parameters
     ----------
-    iterative : bool
-                whether to orthogonalize after each selection, defaults to `true`
+    recompute_every : int
+                      number of steps after which to recompute the pi score
+                      defaults to 1, if 0 no re-computation is done
 
     k : int
         number of eigenvectors to compute the importance score with, defaults to 1
@@ -634,7 +638,7 @@ def __init__(
         self,
         selection_type,
         mixing=0.5,
-        iterative=True,
+        recompute_every=1,
         k=1,
         tolerance=1e-12,
         n_to_select=None,
@@ -646,7 +650,7 @@ def __init__(
         self.mixing = mixing
 
         self.k = k
-        self.iterative = iterative
+        self.recompute_every = recompute_every
         self.tolerance = tolerance
 
         super().__init__(
@@ -707,7 +711,7 @@ def _continue_greedy_search(self, X, y, n_to_select):
 
         for c in self.selected_idx_:
 
-            if (
+            if self.recompute_every != 0 and (
                 np.linalg.norm(np.take(self.X_current_, [c], axis=self._axis))
                 > self.tolerance
             ):
@@ -720,13 +724,15 @@ def _continue_greedy_search(self, X, y, n_to_select):
     def _update_post_selection(self, X, y, last_selected):
         """
         Saves the most recently selected feature, increments the feature counter,
-        and, if the CUR is iterative, orthogonalizes the remaining features by
+        and, if the CUR is iterative (recompute_every>0), orthogonalizes the remaining features by
         the most recently selected.
         """
         super()._update_post_selection(X, y, last_selected)
 
-        if self.iterative:
+        if self.recompute_every != 0:
             self._orthogonalize(last_selected)
+
+        if len(self.selected_idx_) % self.recompute_every == 0:
             self.pi_ = self._compute_pi(self.X_current_, self.y_current_)
 
         self.pi_[last_selected] = 0.0

skcosmo/feature_selection/_base.py

@@ -147,8 +147,9 @@ class CUR(_CUR):
 
     Parameters
     ----------
-    iterative : bool
-                whether to orthogonalize after each selection, defaults to `true`
+    recompute_every : int
+                      number of steps after which to recompute the pi score
+                      defaults to 1, if 0 no re-computation is done
 
     k : int
         number of eigenvectors to compute the importance score with, defaults to 1
@@ -193,7 +194,7 @@ class CUR(_CUR):
 
     def __init__(
         self,
-        iterative=True,
+        recompute_every=1,
         k=1,
         tolerance=1e-12,
         n_to_select=None,
@@ -204,7 +205,7 @@ def __init__(
     ):
         super().__init__(
             selection_type="feature",
-            iterative=iterative,
+            recompute_every=recompute_every,
             k=k,
             tolerance=tolerance,
             n_to_select=n_to_select,
@@ -223,8 +224,9 @@ class PCovCUR(_PCovCUR):
 
     Parameters
     ----------
-    iterative : bool
-                whether to orthogonalize after each selection, defaults to `true`
+    recompute_every : int
+                      number of steps after which to recompute the pi score
+                      defaults to 1, if 0 no re-computation is done
 
     k : int
         number of eigenvectors to compute the importance score with, defaults to 1
@@ -277,7 +279,7 @@ class PCovCUR(_PCovCUR):
     def __init__(
         self,
         mixing=0.5,
-        iterative=True,
+        recompute_every=1,
         k=1,
         tolerance=1e-12,
         n_to_select=None,
@@ -289,7 +291,7 @@ def __init__(
         super().__init__(
             selection_type="feature",
             mixing=mixing,
-            iterative=iterative,
+            recompute_every=recompute_every,
             k=k,
             tolerance=tolerance,
             n_to_select=n_to_select,

skcosmo/sample_selection/_base.py

@@ -152,8 +152,9 @@ class CUR(_CUR):
 
     Parameters
      ----------
-     iterative : bool
-                 whether to orthogonalize after each selection, defaults to `true`
+     recompute_every : int
+                       number of steps after which to recompute the pi score
+                       defaults to 1, if 0 no re-computation is done
 
      k : int
          number of eigenvectors to compute the importance score with, defaults to 1
@@ -199,7 +200,7 @@ class CUR(_CUR):
 
     def __init__(
         self,
-        iterative=True,
+        recompute_every=1,
         k=1,
         tolerance=1e-12,
         n_to_select=None,
@@ -210,7 +211,7 @@ def __init__(
     ):
         super().__init__(
             selection_type="sample",
-            iterative=iterative,
+            recompute_every=recompute_every,
             k=k,
             tolerance=tolerance,
             n_to_select=n_to_select,
@@ -234,8 +235,10 @@ class PCovCUR(_PCovCUR):
             The PCovR mixing parameter, as described in PCovR as
             :math:`{\\alpha}`. Stored in :py:attr:`self.mixing`.
 
-    iterative : bool
-                whether to orthogonalize after each selection, defaults to `true`
+    recompute_every : int
+                      number of steps after which to recompute the pi score
+                      defaults to 1, if 0 no re-computation is done
+
 
     k : int
         number of eigenvectors to compute the importance score with, defaults to 1
@@ -286,7 +289,7 @@ class PCovCUR(_PCovCUR):
     def __init__(
         self,
         mixing=0.5,
-        iterative=True,
+        recompute_every=1,
         k=1,
         tolerance=1e-12,
         n_to_select=None,
@@ -298,7 +301,7 @@ def __init__(
         super().__init__(
             selection_type="sample",
             mixing=mixing,
-            iterative=iterative,
+            recompute_every=recompute_every,
             k=k,
             tolerance=tolerance,
             n_to_select=n_to_select,

tests/test_feature_pcov_cur.py

@@ -39,7 +39,7 @@ def test_non_it(self):
         This test checks that the model can be run non-iteratively
         """
         self.idx = [2, 8, 3, 6, 7, 9, 1, 0, 5]
-        selector = PCovCUR(n_to_select=9, iterative=False)
+        selector = PCovCUR(n_to_select=9, recompute_every=0)
         selector.fit(self.X, self.y)
 
         self.assertTrue(np.allclose(selector.selected_idx_, self.idx))

tests/test_feature_simple_cur.py

@@ -40,7 +40,7 @@ def test_non_it(self):
         _, UC = np.linalg.eigh(C)
         ref_idx = np.argsort(-(UC[:, -1] ** 2.0))[:-1]
 
-        selector = CUR(n_to_select=self.X.shape[-1] - 1, iterative=False)
+        selector = CUR(n_to_select=self.X.shape[-1] - 1, recompute_every=0)
         selector.fit(self.X)
 
         self.assertTrue(np.allclose(selector.selected_idx_, ref_idx))

tests/test_sample_pcov_cur.py

@@ -18,7 +18,7 @@ def test_known(self):
         This test checks that the model returns a known set of indices
         """
 
-        selector = PCovCUR(n_to_select=10, mixing=0.5, iterative=True)
+        selector = PCovCUR(n_to_select=10, mixing=0.5)
         selector.fit(self.X, self.y)
 
         self.assertTrue(np.allclose(selector.selected_idx_, self.idx))
@@ -50,9 +50,8 @@ def test_non_it(self):
         """
         This test checks that the model can be run non-iteratively
         """
-        selector = PCovCUR(n_to_select=10, iterative=False)
         self.idx = [256, 32, 138, 290, 362, 141, 359, 254, 428, 9]
-        selector = PCovCUR(n_to_select=10, iterative=False)
+        selector = PCovCUR(n_to_select=10, recompute_every=0)
         selector.fit(self.X, self.y)
 
         self.assertTrue(np.allclose(selector.selected_idx_, self.idx))

tests/test_sample_simple_cur.py

@@ -42,7 +42,7 @@ def test_non_it(self):
         _, UK = np.linalg.eigh(K)
         ref_idx = np.argsort(-(UK[:, -1] ** 2.0))[: self.n_select]
 
-        selector = CUR(n_to_select=len(ref_idx), iterative=False)
+        selector = CUR(n_to_select=len(ref_idx), recompute_every=0)
         selector.fit(self.X)
 
         self.assertTrue(np.allclose(selector.selected_idx_, ref_idx))