# KeywordAnalyzer
# Initialize jvm
-JVM = initVM(CLASSPATH)
+JVM = initVM(CLASSPATH, maxheap=settings.JVM_MAXHEAP)
import sys
import os
else:
tag = catalogue.models.Tag.objects.get(id=doc.get("tag_id"))
# don't add the pdcounter tag if same tag already exists
- if not (is_pdcounter and filter(lambda t: tag.slug == t.slug, tags)):
- tags.append(tag)
+
+ tags.append(tag)
+
except catalogue.models.Tag.DoesNotExist: pass
except PDCounterAuthor.DoesNotExist: pass
except PDCounterBook.DoesNotExist: pass