Merge pull request #363 from jakirkham/add_gen_dict_n_comps

Provide optional number of components argument for dictionary learning

nanshe/imp/segment.py

@@ -757,7 +757,10 @@ def preprocess_data(new_data, out=None, **parameters):
 
 @prof.log_call(trace_logger)
 @hdf5.record.static_array_debug_recorder
-def generate_dictionary(new_data, initial_dictionary=None, **parameters):
+def generate_dictionary(new_data,
+                        initial_dictionary=None,
+                        n_components=None,
+                        **parameters):
     """
         Generates a dictionary using the data and parameters given for trainDL.
 
@@ -769,6 +772,9 @@ def generate_dictionary(new_data, initial_dictionary=None, **parameters):
             initial_dictionary(numpy.ndarray):  dictionary to start the
                                                 algorithm with.
 
+            n_components(int):                  number of components for the
+                                                dictionary to use.
+
             **parameters(dict):                 passed directly to
                                                 spams.trainDL.
 
@@ -778,6 +784,29 @@ def generate_dictionary(new_data, initial_dictionary=None, **parameters):
 
     import nanshe.box
 
+    # Sync the number of components with the method.
+    if n_components is None:
+        if "spams.trainDL" in parameters:
+            n_components = parameters["spams.trainDL"]["K"]
+        elif "sklearn.decomposition.dict_learning_online" in parameters:
+            n_components = parameters["sklearn.decomposition.dict_learning_online"]["n_components"]
+        else:
+            assert False, "Unknown algorithm must define `n_components`."
+    else:
+        if "spams.trainDL" in parameters:
+            assert parameters["spams.trainDL"].get("K", n_components) == n_components,\
+                "If `n_components` and `spams.trainDL[\"K\"]` are defined," \
+                " they should be defined the same."
+            parameters["spams.trainDL"]["K"] = n_components
+        elif "sklearn.decomposition.dict_learning_online" in parameters:
+            assert parameters["sklearn.decomposition.dict_learning_online"].get("n_components", n_components) == n_components,\
+                "If `n_components` and " \
+                "`sklearn.decomposition.dict_learning_online[\"n_components\"]`" \
+                " are defined, they should be defined the same."
+            parameters["sklearn.decomposition.dict_learning_online"]["n_components"] = n_components
+        else:
+            assert False, "Unknown algorithm cannot use `n_components`."
+
     # Needs to be floating point.
     # However, it need not be double precision as there is single precision
     # function signature.

tests/test_nanshe/test_imp/test_segment.py

@@ -1042,12 +1042,12 @@ def test_generate_dictionary_04(self):
         d = nanshe.imp.segment.generate_dictionary(
             g.astype(numpy.float32),
             g.astype(numpy.float32),
+            len(g),
             **{
                 "spams.trainDL" : {
                     "gamma2" : 0,
                     "gamma1" : 0,
                     "numThreads" : 1,
-                    "K" : len(g),
                     "iter" : 10,
                     "modeD" : 0,
                     "posAlpha" : True,
@@ -1098,12 +1098,12 @@ def test_generate_dictionary_05(self):
         d = nanshe.imp.segment.generate_dictionary(
             g.astype(float),
             g.astype(float),
+            len(g),
             **{
                 "spams.trainDL" : {
                     "gamma2" : 0,
                     "gamma1" : 0,
                     "numThreads" : 1,
-                    "K" : len(g),
                     "iter" : 10,
                     "modeD" : 0,
                     "posAlpha" : True,
@@ -1155,12 +1155,12 @@ def test_generate_dictionary_06(self):
         d = nanshe.imp.segment.generate_dictionary(
             g.astype(numpy.float32),
             g.astype(numpy.float32),
+            len(g),
             **{
                 "spams.trainDL" : {
                     "gamma2" : 0,
                     "gamma1" : 0,
                     "numThreads" : 1,
-                    "K" : len(g),
                     "iter" : 10,
                     "modeD" : 0,
                     "posAlpha" : True,
@@ -1212,12 +1212,12 @@ def test_generate_dictionary_07(self):
         d = nanshe.imp.segment.generate_dictionary(
             g.astype(float),
             g.astype(float),
+            len(g),
             **{
                 "spams.trainDL" : {
                     "gamma2" : 0,
                     "gamma1" : 0,
                     "numThreads" : 1,
-                    "K" : len(g),
                     "iter" : 10,
                     "modeD" : 0,
                     "posAlpha" : True,
@@ -1359,10 +1359,10 @@ def test_generate_dictionary_10(self):
         d = nanshe.imp.segment.generate_dictionary(
             g.astype(float),
             g.astype(float),
+            len(g),
             **{
                 "sklearn.decomposition.dict_learning_online" : {
                     "n_jobs" : 1,
-                    "n_components" : len(g),
                     "n_iter" : 20,
                     "batch_size" : 256,
                     "alpha" : 0.2
@@ -1408,10 +1408,10 @@ def test_generate_dictionary_11(self):
         d = nanshe.imp.segment.generate_dictionary(
             g.astype(float),
             g.astype(float),
+            len(g),
             **{
                 "sklearn.decomposition.dict_learning_online" : {
                     "n_jobs" : 1,
-                    "n_components" : len(g),
                     "n_iter" : 20,
                     "batch_size" : 256,
                     "alpha" : 0.2