Update app.py

app.py

@@ -10,9 +10,16 @@ def predict_label(text):
     ip_len = [len(ip)]
     
     span_scores = extract_spannet_scores(span_model,ip,ip_len, pos_col=1, task_col=2)
-    pooled_scores = pool_span_scores(span_scores, ip_len)
-    
-    ent_scores = extract_ent_scores(entity_model,ip,pooled_scores, pos_col=1, task_col=2)
+    span_pooled_scores = pool_span_scores(span_scores, ip_len)
+
+    msa_span_scores = extract_spannet_scores(msa_span_model,ip,ip_len, pos_col=1, task_col=2)
+    msa_pooled_scores = pool_span_scores(msa_span_scores, ip_len)
+
+    ensemble_span_scores = [score for scores in [span_scores, msa_span_scores] for score in scores]
+    ensemble_pooled_scores = pool_span_scores(ensemble_span_scores, ip_len)
+    ensemble_pred_tags = [entities_from_token_classes(sent_targs) for sent_targs in ensemble_pooled_scores]
+
+    ent_scores = extract_ent_scores(entity_model,ip,ensemble_pred_tags, pos_col=1, task_col=2)
     combined_sequences, ent_pred_tags = pool_ent_scores(ent_scores, ip_len)
     
     return combined_sequences
@@ -32,11 +39,14 @@ if __name__ == '__main__':
     
     from src.predict import extract_spannet_scores,extract_ent_scores,pool_span_scores,pool_ent_scores
     from src.network import SpanNet, EntNet
+    from src.validate import entities_from_token_classes
     
     span_path = 'models/span.model'
+    msa_span_path = 'models/entity.msa.model'
     entity_path= 'models/entity.msa.model'
-    # span_msa_path = 'models/sp'
+    
     span_model = SpanNet.load_model(span_path)
+    msa_span_model = SpanNet.load_model(msa_span_path)
     entity_model = EntNet.load_model(entity_path)
     
     iface = gr.Interface(fn=predict_label, inputs="text", outputs="text")