remove redundant "minibatch" algorithm

Author: AdrianSosic

examples/minibatch_algorithms.py

@@ -98,7 +98,6 @@ def progress_callback(nmf_instance: TransformInvariantNMF, iteration: int) -> bo
 for params in (
     dict(),
     #
-    dict(algorithm=MiniBatchAlgorithm.Basic_MU, batch_size=10),
     dict(algorithm=MiniBatchAlgorithm.Cyclic_MU, batch_size=10),
     dict(algorithm=MiniBatchAlgorithm.ASG_MU, batch_size=10),
     dict(algorithm=MiniBatchAlgorithm.GSG_MU, batch_size=10),

tnmf/TransformInvariantNMF.py

@@ -48,7 +48,6 @@ class MiniBatchAlgorithm(Enum):
     r"""
     MiniBatch algorithms that can be used with :meth:`.TransformInvariantNMF.fit_minibatch`.
     """
-    Basic_MU = 3   # Algorithm 3 Basic alternating scheme for MU rules
     Cyclic_MU = 4  # Algorithm 4 Cyclic mini-batch for MU rules
     ASG_MU = 5     # Algorithm 5 Asymmetric SG mini-batch MU rules (ASG-MU)
     GSG_MU = 6     # Algorithm 6 Greedy SG mini-batch MU rules (GSG-MU)
@@ -352,9 +351,9 @@ def fit_batch(
     def fit_minibatches(
             self,
             V: np.ndarray,
-            algorithm: MiniBatchAlgorithm = MiniBatchAlgorithm.Basic_MU,
+            algorithm: MiniBatchAlgorithm = MiniBatchAlgorithm.ASG_MU,
             batch_size: int = 3,
-            n_epochs: int = 1000,   # corresponds to max_iter if algorithm == MiniBatchAlgorithm.Basic_MU
+            n_epochs: int = 1000,
             sag_lambda: float = 0.2,
             keep_W: bool = False,
             sparsity_H: float = 0.,
@@ -375,10 +374,9 @@ def fit_minibatches(
         algorithm: MiniBatchAlgorithm
             MiniBatch update scheme to be used. See :class:`MiniBatchAlgorithm` and [3]_ for the different choices.
         batch_size: int, default = 3
-            Number of samples per mini batch. Ignored if algorithm==MiniBatchAlgorithm.Basic_MU
+            Number of samples per mini batch.
         n_epochs: int, default = 1000
-            Maximum number of epochs (iterations if algorithm==MiniBatchAlgorithm.Basic_MU) across the full
-            sample set to be performed.
+            Maximum number of epochs across the full sample set to be performed.
         sag_lambda: float, default = 0.2
             Exponential forgetting factor for for the stochastic _average_ gradient updates, i.e.
             MiniBatchAlgorithm.ASAG_MU and MiniBatchAlgorithm.GSAG_MU
@@ -416,7 +414,6 @@ def fit_minibatches(
         batches = list(_compute_sequential_minibatches(len(self._V), batch_size))
 
         epoch_update = {
-            MiniBatchAlgorithm.Basic_MU: self._epoch_update_algorithm_3,
             MiniBatchAlgorithm.Cyclic_MU: self._epoch_update_algorithm_4,
             MiniBatchAlgorithm.ASG_MU: self._epoch_update_algorithm_5,
             MiniBatchAlgorithm.GSG_MU: self._epoch_update_algorithm_6,
@@ -441,8 +438,7 @@ def fit_minibatches(
                 if not progress_callback(self, epoch):
                     break
             else:
-                self._logger.info(f"{'Iteration' if algorithm == MiniBatchAlgorithm.Basic_MU else 'Epoch'}: {epoch}\t"
-                                  f"Energy function: {self._energy_function()}")
+                self._logger.info(f"{'Epoch'}: {epoch}\tEnergy function: {self._energy_function()}")
 
         self._logger.info("MiniBatch TNMF finished.")
 
@@ -459,12 +455,6 @@ def _accumulate_gradient_W(self, gradW_neg, gradW_pos, sag_lambda: float, s: sli
 
         return gradW_neg, gradW_pos
 
-    def _epoch_update_algorithm_3(self, _, ___, args_update_H, __):
-        # update H for all samples
-        self._update_H(**args_update_H)
-        # update W after processing all batches using all samples
-        self._update_W()
-
     def _epoch_update_algorithm_4(self, _, batches, args_update_H, __):
         gradW_neg, gradW_pos = 0, 0
         for batch in batches:

tnmf/tests/test_minibatch.py

@@ -16,7 +16,7 @@
 
 # hard-coded expected energy levels for the different algorithms
 expected_energies = {
-    MiniBatchAlgorithm.Basic_MU: 14434.02658,
+    'full_batch': 14434.02658,
     MiniBatchAlgorithm.Cyclic_MU: 14434.02658,
     MiniBatchAlgorithm.ASG_MU: 4558.86695,
     MiniBatchAlgorithm.GSG_MU: 14223.14454,
@@ -57,13 +57,20 @@ def fit_nmf(backend, algorithm):
         verbose=3,
         reconstruction_mode='valid',
     )
-    nmf.fit_minibatches(
-        V,
-        sparsity_H=0.1,
-        algorithm=algorithm,
-        batch_size=3,
-        n_epochs=5,
-        sag_lambda=0.8)
+    if isinstance(algorithm, MiniBatchAlgorithm):
+        nmf.fit_minibatches(
+            V,
+            sparsity_H=0.1,
+            algorithm=algorithm,
+            batch_size=3,
+            n_epochs=5,
+            sag_lambda=0.8)
+    else:
+        nmf.fit_batch(
+            V,
+            sparsity_H=0.1,
+            n_iterations=5,
+        )
 
     return nmf
 

tnmf/tests/test_stream.py

@@ -19,7 +19,6 @@
 # hard-coded expected energy levels for the different algorithms
 expected_energies = {
     # no need to test all algorithms here
-    MiniBatchAlgorithm.Basic_MU: 136.84096550,
     # MiniBatchAlgorithm.Cyclic_MU: 136.8409655,
     # MiniBatchAlgorithm.ASG_MU: 97.0072791,
     # MiniBatchAlgorithm.GSG_MU: 136.43285833,
@@ -98,7 +97,7 @@ def test_with_generator_limited():
     nmf.fit(
         V,
         sparsity_H=0.1,
-        algorithm=MiniBatchAlgorithm.Basic_MU,
+        algorithm=MiniBatchAlgorithm.Cyclic_MU,
         subsample_size=50,
         max_subsamples=5,
         batch_size=3,