def get_clusters(model_file,
word_vectors_file,
test_file,
get_gold_clusters=False):
# load model and vector mappings.
model = load_model(model_file)
embed_map = EmbedMap(word_vectors_file)
# calculate system clusters
gold_clusters, sys_clusters = {}, {}
test_file_reader = jsonlines.open(test_file)
for doc_data in test_file_reader.iter():
doc = Document(doc_data, embed_map, remove_multi_word_mention=False)
mention_pairs, ref_index = doc.generate_candidate_mention_pairs()
mention_pairs_array, _ = convert_train_data(mention_pairs)
# get predictions.
predicts = [score[0] for score in model.predict(mention_pairs_array)]
predicts = dict(zip(ref_index, predicts))
# generate clusters.
doc_clusters = cluster_mentions(list(doc.candidates.keys()), predicts)
gold_clusters[doc.doc_id] = doc.gold_clusters
sys_clusters[doc.doc_id] = doc_clusters
if get_gold_clusters:
return gold_clusters, sys_clusters
else:
return sys_clusters
