Update app.py

app.py

@@ -12,14 +12,21 @@ model.eval()
 if torch.cuda.is_available():
     model.cuda()
 
+validation_results = json.load(open('validation_results.json'))
+scores, thresholds, precisions = validation_results['scores'], validation_results['thresholds'], validation_results['precisions']
+
+def get_threshold_precision(score_):
+    for score, threshold, precision in zip(all_scores, thresholds, precisions):
+        if score_ < score:
+            break
+        prev_score, prev_threshold, prev_precision = score, threshold, precision
+    if prev_threshold == prev_score:
+        prev_threshold = score_
+    return prev_threshold, prev_precision
 
 def normalize_spaces(text):
     return re.sub(r'\s+', ' ', text).strip()
 
-# Load your validation set
-#with open('validation_data.json', 'r') as file:
-#    validation_data = json.load(file)
-
 def fill_template(title, authors, abstract):
     title = normalize_spaces(title.replace('\n', ' '))
     authors = ', '.join([author.strip() for author in authors.split(',')])
@@ -50,9 +57,9 @@ def predict(title, authors, abstract):
     #selected = [d for d in validation_data if d['score'] >= score]
     #true_positives = sum(1 for d in selected if d['label'] == 1)
     #precision = true_positives / len(selected) if selected else 0
-    precision = 0.2
+    threshold, precision = get_threshold_precision(score)
     
-    result = f"Your score: {score:.2f}.\nFor papers with score >= {score:.2f}, {precision * 100:.2f}% are selected by AK."
+    result = f"Your score: {score:.2f}.\nFor papers with score >= {threshold:.2f}, {precision * 100:.2f}% are selected by AK."
     
     return score, result