TypeError during training should yield more helpful error message

[rlvoyer]

My hunch is that this a result of not having any positive examples in my training set (will confirm), but this should have a more helpful error message:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-49-9a80de5295c7> in <module>()
----> 1 deduper.train()

/Users/robertvoyer/anaconda3/envs/entity-search/lib/python3.6/site-packages/dedupe/api.py in train(self, recall, index_predicates)
    669         self.classifier.fit(self.data_model.distances(examples), y)
    670
--> 671         self._trainBlocker(recall, index_predicates)
    672
    673     def _trainBlocker(self, recall, index_predicates):  # pragma: no cover

/Users/robertvoyer/anaconda3/envs/entity-search/lib/python3.6/site-packages/dedupe/api.py in _trainBlocker(self, recall, index_predicates)
    680
    681         self.predicates = block_learner.learn(matches,
--> 682                                               recall)
    683
    684         self.blocker = blocking.Blocker(self.predicates)

/Users/robertvoyer/anaconda3/envs/entity-search/lib/python3.6/site-packages/dedupe/training.py in learn(self, matches, recall)
     31         comparison_count = self.comparisons(self.total_cover, compound_length)
     32
---> 33         coverable_dupes = set.union(*viewvalues(dupe_cover))
     34         uncoverable_dupes = [pair for i, pair in enumerate(matches)
     35                              if i not in coverable_dupes]

TypeError: descriptor 'union' of 'set' object needs an argument

[rlvoyer]

Confirmed: the error went away after adding some positive examples to my training set.

[alexing]

Is this getting fixed? It's pretty annoying

[lokhande-vishnu]

you can fix it by making a change like this
if len(viewvalues(dupe_cover)) != 0:
coverable_dupes = set.union(*viewvalues(dupe_cover))
else:
coverable_dupes = set()