fix: bug in recreating featurizers (#26)

src/helpers/predict_methods.py

@@ -12,10 +12,6 @@ def predict_onnx(model, dataset: JaqpotpyDataset, request):
         np_dtype = onnx.helper.tensor_dtype_to_np_dtype(
             independent_feature.type.tensor_type.elem_type
         )
-        # if np_dtype == "float64":
-        #     np_dtype = "float32"
-        # elif np_dtype == ["int64", "uint64"]:
-        #     np_dtype = "int32"
 
         if len(model.graph.input) == 1:
             input_feed[independent_feature.name] = dataset.X.values.astype(np_dtype)

src/helpers/recreate_featurizer.py

@@ -1,13 +1,15 @@
 import numpy as np
 
-def recreate_featurizer(featurizer_name,  featurizer_config):
-    featurizer_class = getattr(__import__(
-        'jaqpotpy.descriptors.molecular',
-          fromlist=[featurizer_name]), featurizer_name)
+
+def recreate_featurizer(featurizer_name, featurizer_config):
+    featurizer_class = getattr(
+        __import__("jaqpotpy.descriptors.molecular", fromlist=[featurizer_name]),
+        featurizer_name,
+    )
     featurizer = featurizer_class()
     for attr, value in featurizer_config.items():
-        if attr !='class': # skip the class attribute
+        if attr != "class":  # skip the class attribute
             if isinstance(value, list):
-                value = np.array(value['value'])
-            setattr(featurizer, attr, value['value'])
-    return featurizer
+                value = np.array(value["value"])
+            setattr(featurizer, attr, value)
+    return featurizer