diff --git a/baselines/EMNLP2019/entitylinker.py b/baselines/EMNLP2019/entitylinker.py
index ce307b2..1eebaac 100644
--- a/baselines/EMNLP2019/entitylinker.py
+++ b/baselines/EMNLP2019/entitylinker.py
@@ -8,6 +8,7 @@
 import time
 import diffbot_nlapi
 import logging
+import pathlib
 
 from config import MODEL, NUMBER_URI_CANDIDATES, SOFT_COREF_CANDIDATES
 
@@ -15,6 +16,7 @@
 Span.set_extension("el_candidates", default=[])
 Span.set_extension("uri_candidates", default=[])
 
+pathlib.Path('tmp').mkdir(parents=True, exist_ok=True)
 db = SqliteDict(os.path.join('tmp','el.db'), autocommit=True)
 
 configuration = diffbot_nlapi.Configuration()
diff --git a/baselines/EMNLP2019/prepare_data.py b/baselines/EMNLP2019/prepare_data.py
index 80a2577..e7bb2fb 100644
--- a/baselines/EMNLP2019/prepare_data.py
+++ b/baselines/EMNLP2019/prepare_data.py
@@ -12,7 +12,7 @@
 import evaluator
 from instance import Instance
 import numpy as np
-from run import generate_candidates, classify_instances
+from run import generate_candidates
 from vocab import Vocab
 from config import NUMBER_URI_CANDIDATES_TO_CONSIDER, URI_THRESHOLD, MODEL, KNOWLEDGE_NET_DIR, MULTITASK
 
diff --git a/baselines/EMNLP2019/run.py b/baselines/EMNLP2019/run.py
index 6abfcf1..8333e0f 100644
--- a/baselines/EMNLP2019/run.py
+++ b/baselines/EMNLP2019/run.py
@@ -5,6 +5,7 @@
 import entitylinker
 import bert_wrapper
 import yaml
+import sys
 
 import spacy
 nlp = spacy.load('en_core_web_lg')
@@ -122,7 +123,7 @@ def generate_candidates(text):
 
   return instances
 
-def classify_instances(instances, predicate_ids_to_classify=predicate_thresholds.keys(), includeUri=True):
+def classify_instances(instances, predicate_ids_to_classify=None, includeUri=True):
   if len(instances) == 0:
     return
   max_length = max([ len(instance.get_words()) for instance in instances ])
@@ -148,6 +149,8 @@ def classify_instances(instances, predicate_ids_to_classify=predicate_thresholds
         global_features_batch.append(global_features)
 
   for sess, model_preds, predicate_thresholds in models:
+    if predicate_ids_to_classify is None:
+      predicate_ids_to_classify = predicate_thresholds.keys()
     model_preds = predicate_ids_to_classify if CANDIDATE_RECALL else model_preds
     if len(set(model_preds).intersection(predicate_ids_to_classify)) == 0:
       continue
@@ -195,6 +198,9 @@ def run_batch(texts):
   return ret
 
 if __name__ == "__main__":
+  if len(models) == 0:
+    print("No trained model")
+    sys.exit(0)
   import fileinput
   for line in fileinput.input():
     for fact in run_batch([line])[0]: