New Gaussian class changes adapted to SVM models.

- Now the latest version of Gaussian class is compatible with SVM
  models.
- data/handler.py now has the `atoms_per_image` attribute available for
  both training and inference.
- ml4chem/utils.py get_chunks() function now should work for both SVM
  and non SVM models.

ml4chem/data/handler.py

@@ -58,10 +58,10 @@ def prepare_images(self, images, purpose=None):
         """
         logger.info("Preparing images for {}...".format(purpose))
         self.images = OrderedDict()
-
+        self.atoms_per_image = []
+
         if purpose == "training":
             self.targets = []
-            self.atoms_per_image = []
 
         duplicates = 0
 

ml4chem/features/gaussian.py

@@ -375,16 +375,17 @@ def calculate(self, images=None, purpose="training", data=None, svm=False):
             del stacked_features
 
             # Restack images
+            feature_space = []
+
             if svm:
-                computations = []
                 reference_space = []
 
                 for i, image in enumerate(images.items()):
-                    computations.append(
-                        self.restack_image(
-                            i, image, scaled_feature_space=scaled_feature_space, svm=svm
-                        )
+                    restacked = client.submit(
+                        self.restack_image,
+                        *(i, image, None, scaled_feature_space, svm)
                     )
+                    feature_space.append(restacked)
 
                     # image = (hash, ase_image) -> tuple
                     for atom in image[1]:
@@ -396,7 +397,6 @@ def calculate(self, images=None, purpose="training", data=None, svm=False):
                     *reference_space, scheduler=self.scheduler
                 )
             else:
-                feature_space = []
                 try:
                     for i, image in enumerate(images.items()):
                         restacked = client.submit(
@@ -408,16 +408,14 @@ def calculate(self, images=None, purpose="training", data=None, svm=False):
                 except UnboundLocalError:
                     # scaled_feature_space does not exist.
                     for i, image in enumerate(images.items()):
-                        computations.append(
-                            self.restack_image(
-                                i, image, feature_space=feature_space, svm=svm
-                            )
+                        restacked = client.submit(
+                            self.restack_image,
+                            *(i, image, feature_space, None, svm)
                         )
+                        feature_space.append(restacked)
 
             feature_space = client.gather(feature_space)
             feature_space = OrderedDict(feature_space)
-            # feature_space = dask.compute(*computations, scheduler=self.scheduler)
-            # del computations
 
             preprocessor.save_to_file(preprocessor, self.save_preprocessor)
 

ml4chem/utils.py

@@ -73,7 +73,8 @@ def get_chunks(sequence, chunk_size, svm=True):
     """
     res = []
 
-    if svm is False and isinstance(sequence, dict):
+    #if svm is False and isinstance(sequence, dict):
+    if isinstance(sequence, dict):
         sequence = sequence.items()
 
     for item in sequence: