Minor fixes

app.py

@@ -6,14 +6,14 @@ from transformers import AutoTokenizer
 
 
 tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased")
-normalized_input_embeddings = np.load("embeddings/bert-large-uncased/normalized.npy")
-unnormalized_input_embeddings = np.load("embeddings/bert-large-uncased/unnormalized.npy")
+normalized = np.load("embeddings/bert-large-uncased/normalized.npy")
+unnormalized = np.load("embeddings/bert-large-uncased/unnormalized.npy")
 
-index_L2 = IndexFlatL2(unnormalized_input_embeddings.shape[-1])  
-index_L2.add(unnormalized_input_embeddings)
+index_L2 = IndexFlatL2(unnormalized.shape[-1])  
+index_L2.add(unnormalized)
 
-index_IP = IndexFlatIP(normalized_input_embeddings.shape[-1])  
-index_IP.add(normalized_input_embeddings)
+index_IP = IndexFlatIP(normalized.shape[-1])  
+index_IP.add(normalized)
 
 
 vocab = {v:k for k,v in tokenizer.vocab.items()}
@@ -27,12 +27,12 @@ def get_first_subword(word):
 
 def search(token_to_lookup, num_neighbors=250):
     i = get_first_subword(token_to_lookup)
-    _ , I_IP = index_IP.search(normalized_input_embeddings[i:i+1], num_neighbors)
+    _ , I_IP = index_IP.search(normalized[i:i+1], num_neighbors)
     hits_IP = lookup_table.take(I_IP[0])
     results_IP = hits_IP.values[1:]
     results_IP = [r for r in results_IP if not "[unused" in r]
 
-    _ , I_L2 = index_L2.search(unnormalized_input_embeddings[i:i+1], num_neighbors)
+    _ , I_L2 = index_L2.search(unnormalized[i:i+1], num_neighbors)
     hits_L2 = lookup_table.take(I_L2[0])
     results_L2 = hits_L2.values[1:]
     results_L2 = [r for r in results_L2 if not "[unused" in r]