Update app.py

app.py

@@ -85,19 +85,23 @@ def encode_passages(texts):
 PASSAGE_EMBS=encode_passages(CORPUS)
 
 
-def retrieve_top_k(query,k=3):
+def retrieve_top_k(query, k=3):
 
-    qv=encode_queries([query])[0]
-    sims=np.dot(PASSAGE_EMBS,qv)
+    qv = encode_queries([query])[0]
+    sims = np.dot(PASSAGE_EMBS, qv)
 
-    idxs=np.argsort(-sims)[:k]
+    idxs = np.argsort(-sims)[:k]
 
-    results=[]
-    for rank,i in enumerate(idxs):
+    results = []
 
-        results.append(
-            {"rank":rank+1,"similarity":float(sims[i]),"context":CORPUS[i]}
-        )
+    for rank, i in enumerate(idxs):
+
+        results.append({
+            "rank": rank + 1,
+            "similarity": float(sims[i]),
+            "context": CORPUS[i],
+            "language": df.iloc[i]["language"]
+        })
 
     return results
 
@@ -117,21 +121,33 @@ def answer_with_context(question,context):
     return {"answer":out["answer"],"score":float(out["score"])}
 
 
-def no_context_flow(question,top_k=3):
+def no_context_flow(question, lang_choice, top_k=3):
 
-    cands=retrieve_top_k(question,k=top_k)
+    cands = retrieve_top_k(question, k=top_k)
 
-    best={"answer":"","score":-1,"used_context":""}
+    best = {"answer": "", "score": -1, "used_context": ""}
 
     for c in cands:
 
-        out=answer_with_context(question,c["context"])
+        # skip passages of other languages
+        if c["language"] != lang_choice[:2].lower():
+            continue
 
-        if out["score"]>best["score"]:
+        out = answer_with_context(question, c["context"])
 
-            best={"answer":out["answer"],"score":out["score"],"used_context":c["context"]}
+        if out["score"] > best["score"]:
+            best = {
+                "answer": out["answer"],
+                "score": out["score"],
+                "used_context": c["context"]
+            }
 
-    return {"answer":best["answer"],"score":best["score"],"used_context":best["used_context"],"retrieved":cands}
+    return {
+        "answer": best["answer"],
+        "score": best["score"],
+        "used_context": best["used_context"],
+        "retrieved": cands
+    }
 
 
 
@@ -209,7 +225,7 @@ def evaluate_answer(question):
 
     gold=row.iloc[0]["answer_text"]
 
-    result=no_context_flow(question)
+    result=no_context_flow(question,"kn",3)
 
     pred=result["answer"]