fixes #531

CHANGELOG.md

@@ -17,6 +17,7 @@ Most recent releases are shown at the top. Each release shows:
 
 ### fixed:
 - Remove references to `paper-qa` (#530)
+- Reduce memory footprint of `TopicModel.filter` (#531)
 
 
 ## 0.41.3 (2024-04-05)

ktrain/text/eda.py

@@ -369,8 +369,10 @@ def filter(self, obj):
                 "Length of obj is not consistent with the number of documents "
                 + "supplied to get_topic_model"
             )
-        obj = np.array(obj) if isinstance(obj, list) else obj
-        return obj[self.bool_array]
+        #obj = np.array(obj) if isinstance(obj, list) else obj
+        #return obj[self.bool_array]
+        from itertools import compress
+        return list(compress(obj, self.bool_array))
 
     def get_docs(self, topic_ids=[], doc_ids=[], rank=False):
         """

tests/testenv.py

@@ -4,5 +4,6 @@
 
 os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
 os.environ["CUDA_VISIBLE_DEVICES"] = "0"
+os.environ["TF_USE_LEGACY_KERAS"] = "1"
 CURRDIR = os.path.dirname(__file__)
 sys.path.insert(0, os.path.join(CURRDIR, ".."))