Update app.py

app.py

@@ -139,16 +139,26 @@ def process_upload(api_key, files, lang):
 
 # Define the calculate_similarity function
 def calculate_similarity(response, ground_truth):
+    # Encode the response and ground truth
     response_embedding = sentence_model.encode(response, convert_to_tensor=True)
     truth_embedding = sentence_model.encode(ground_truth, convert_to_tensor=True)
-    
-    # Normalize the embeddings
-    response_embedding = response_embedding / response_embedding.norm(p=2)
-    truth_embedding = truth_embedding / truth_embedding.norm(p=2)
-    
-    # Calculate cosine similarity using sklearn's cosine_similarity function
-    similarity = cosine_similarity(response_embedding.reshape(1, -1), truth_embedding.reshape(1, -1))[0][0]
-    return similarity * 100  # Convert to percentage
+
+    # Convert embeddings to numpy arrays for easier manipulation
+    response_embedding = response_embedding.cpu().numpy()
+    truth_embedding = truth_embedding.cpu().numpy()
+
+    # Normalize the embeddings to unit vectors (magnitude of 1)
+    response_embedding = response_embedding / np.linalg.norm(response_embedding)
+    truth_embedding = truth_embedding / np.linalg.norm(truth_embedding)
+
+    # Calculate cosine similarity using numpy's dot product
+    similarity = np.dot(response_embedding, truth_embedding)
+
+    # Return similarity as a percentage (between 0 and 100)
+    similarity_percentage = (similarity + 1) / 2 * 100  # Normalize from [-1, 1] to [0, 100]
+
+    return similarity_percentage
+
 
 # This is the missing query_app function that needs to be defined
 def query_app(query, model_name, use_similarity_check, openai_api_key):