While using the coremltools I noticed this bug with the feature management and created a patch (below). It would be great if someone at Apple can accept this patch. I'd do it myself with a pull request if they would start using github...
--- _feature_management.py 017-06-14 17:51:15.000000000 -0400
+++ _feature_management_new.py 017-06-14 17:52:10.000000000 -0400
@@ -62,7 +62,7 @@
and len(output_features) == 2):
if supports_class_scores:
- out = [(output_features[0], output_class_types),
+ out = [(output_features[0], output_class_type),
(output_features[1], datatypes.Dictionary(output_class_type))]
else:
raise ValueError("Classifier model (as trained) does not support output scores for classes.")