Write analysis file

Author: schmidek

baselines/EMNLP2019/evaluate.py

@@ -28,10 +28,13 @@
   instances = run.generate_candidates(document.documentText)
   for passage in document.passages:
     annotated_properties = set(map(lambda x: x.propertyId, passage.exhaustivelyAnnotatedProperties))
-    if len(annotated_properties) == 0:
+    if which == "dev" and len(annotated_properties) == 0:
       continue
     passage_instances = list(filter(lambda x: x.is_in_span(passage.passageStart, passage.passageEnd), instances))
-    run.classify_instances(passage_instances, annotated_properties)
+    if which == "dev":
+      run.classify_instances(passage_instances, annotated_properties)
+    else:
+      run.classify_instances(passage_instances)
     passage.facts = []
     for fact in passage_instances:
       for predicate_id, label in fact.labels.items():
@@ -59,6 +62,7 @@ def print_evaluation(eval_type):
   evals = evaluator.microEvaluation(confusionMatrix, True)
   evals.extend(evaluator.macroEvaluation(confusionMatrix))
 
+  evaluator.writeAnalysisFile(analysis, 'tmp', eval_type)
   evaluator.writeHtmlFile(analysis, 'tmp', eval_type, goldProperties)
 
 print_evaluation("span_overlap")

baselines/EMNLP2019/run.py

@@ -122,7 +122,7 @@ def generate_candidates(text):
 
   return instances
 
-def classify_instances(instances, predicate_ids_to_classify, includeUri=True):
+def classify_instances(instances, predicate_ids_to_classify=predicate_thresholds.keys(), includeUri=True):
   if len(instances) == 0:
     return
   max_length = max([ len(instance.get_words()) for instance in instances ])
@@ -190,7 +190,7 @@ def run_batch(texts):
   ret = []
   for text in texts:
     instances = generate_candidates(text)
-    classify_instances(instances, predicate_thresholds.keys())
+    classify_instances(instances)
     ret.append(instances)
   return ret