[MRG+1] Make partial_fit ignore n_iter in MiniBatchDictionaryLearning (…

…scikit-learn#17433)

* Make partial_fit ignore n_iter.

* Revert whiteline.

* Fix self.iter_offset.

* Add test on iter_offset_

* Add what's new entry

* Apply thomasjpfan suggestions.

doc/whats_new/v0.24.rst

@@ -44,6 +44,14 @@ Changelog
     :pr:`123456` by :user:`Joe Bloggs <joeongithub>`.
     where 123456 is the *pull request* number, not the issue number.
 
+:mod:`sklearn.decomposition`
+............................
+
+- |Fix| Fixed a bug in
+  :func:`decomposition.MiniBatchDictionaryLearning.partial_fit` which should
+  update the dictionary by iterating only once over a mini-batch.
+  :pr:`17433` by :user:`Chiara Marmo <cmarmo>`
+
 :mod:`sklearn.ensemble`
 .......................
 

sklearn/decomposition/_dict_learning.py

@@ -1490,7 +1490,7 @@ def partial_fit(self, X, y=None, iter_offset=None):
             iter_offset = getattr(self, 'iter_offset_', 0)
         U, (A, B) = dict_learning_online(
             X, self.n_components, alpha=self.alpha,
-            n_iter=self.n_iter, method=self.fit_algorithm,
+            n_iter=1, method=self.fit_algorithm,
             method_max_iter=self.transform_max_iter,
             n_jobs=self.n_jobs, dict_init=dict_init,
             batch_size=len(X), shuffle=False,
@@ -1504,5 +1504,5 @@ def partial_fit(self, X, y=None, iter_offset=None):
         # Keep track of the state of the algorithm to be able to do
         # some online fitting (partial_fit)
         self.inner_stats_ = (A, B)
-        self.iter_offset_ = iter_offset + self.n_iter
+        self.iter_offset_ = iter_offset + 1
         return self

sklearn/decomposition/tests/test_dict_learning.py

@@ -391,6 +391,23 @@ def test_dict_learning_online_partial_fit():
                               decimal=2)
 
 
+def test_dict_learning_iter_offset():
+    n_components = 12
+    rng = np.random.RandomState(0)
+    V = rng.randn(n_components, n_features)
+    dict1 = MiniBatchDictionaryLearning(n_components, n_iter=10,
+                                        dict_init=V, random_state=0,
+                                        shuffle=False)
+    dict2 = MiniBatchDictionaryLearning(n_components, n_iter=10,
+                                        dict_init=V, random_state=0,
+                                        shuffle=False)
+    dict1.fit(X)
+    for sample in X:
+        dict2.partial_fit(sample[np.newaxis, :])
+
+    assert dict1.iter_offset_ == dict2.iter_offset_
+
+
 def test_sparse_encode_shapes():
     n_components = 12
     rng = np.random.RandomState(0)