Correctly process empty documents in AuthorTopicModel

gensim/models/atmodel.py

@@ -461,10 +461,11 @@ def inference(self, chunk, author2doc, doc2author, rhot, collect_sstats=False, c
                 ids = [int(idx) for idx, _ in doc]
             else:
                 ids = [idx for idx, _ in doc]
-            cts = np.array([cnt for _, cnt in doc])
+            ids = np.array(ids, dtype=np.integer)
+            cts = np.array([cnt for _, cnt in doc], dtype=np.integer)
 
             # Get all authors in current document, and convert the author names to integer IDs.
-            authors_d = [self.author2id[a] for a in self.doc2author[doc_no]]
+            authors_d = np.array([self.author2id[a] for a in self.doc2author[doc_no]], dtype=np.integer)
 
             gammad = self.state.gamma[authors_d, :]  # gamma of document d before update.
             tilde_gamma = gammad.copy()  # gamma that will be updated.
@@ -972,9 +973,9 @@ def bound(self, chunk, chunk_doc_idx=None, subsample_ratio=1.0, author2doc=None,
             else:
                 doc_no = d
             # Get all authors in current document, and convert the author names to integer IDs.
-            authors_d = [self.author2id[a] for a in self.doc2author[doc_no]]
-            ids = np.array([id for id, _ in doc])  # Word IDs in doc.
-            cts = np.array([cnt for _, cnt in doc])  # Word counts.
+            authors_d = np.array([self.author2id[a] for a in self.doc2author[doc_no]], dtype=np.integer)
+            ids = np.array([id for id, _ in doc], dtype=np.integer)  # Word IDs in doc.
+            cts = np.array([cnt for _, cnt in doc], dtype=np.integer)  # Word counts.
 
             if d % self.chunksize == 0:
                 logger.debug("bound: at document #%i in chunk", d)

gensim/test/test_atmodel.py

@@ -35,7 +35,6 @@
 # increases the bound.
 # Test that models are compatiple across versions, as done in LdaModel.
 
-
 # Assign some authors randomly to the documents above.
 author2doc = {
     'john': [0, 1, 2, 3, 4, 5, 6],
@@ -110,6 +109,17 @@ def testBasic(self):
         jill_topics = matutils.sparse2full(jill_topics, model.num_topics)
         self.assertTrue(all(jill_topics > 0))
 
+    def testEmptyDocument(self):
+        local_texts = common_texts + [['only_occurs_once_in_corpus_and_alone_in_doc']]
+        dictionary = Dictionary(local_texts)
+        dictionary.filter_extremes(no_below=2)
+        corpus = [dictionary.doc2bow(text) for text in local_texts]
+        a2d = author2doc.copy()
+        a2d['joaquin'] = [len(local_texts) - 1]
+
+        _ = self.class_(corpus, author2doc=a2d, id2word=dictionary, num_topics=2)
+        assert(_)
+
     def testAuthor2docMissing(self):
         # Check that the results are the same if author2doc is constructed automatically from doc2author.
         model = self.class_(