Update app.py

app.py

@@ -17,7 +17,7 @@ class_names = ['Negative', 'Neutral', 'Positive']
 
 # ২. প্রেডিক্টর ফাংশন (Internal use for LIME and Real-time)
 def predictor(texts):
-    inputs = tokenizer(texts, return_tensors="pt", padding='max_length', truncation=True, max_length=64)
+    inputs = tokenizer(texts, return_tensors="pt", padding=True, truncation=True, max_length=64)
     with torch.no_grad():
         outputs = model(**inputs)
     probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
@@ -88,10 +88,9 @@ def get_explanation(text):
     label_idx = np.argmax(probs)
     label_name = class_names[label_idx]
 
-    # explainer = LimeTextExplainer(class_names=class_names, split_expression=r'\s+')
-    explainer = LimeTextExplainer(class_names=class_names)
+    explainer = LimeTextExplainer(class_names=class_names, split_expression=r'\s+')
     # লাইভ মোডে যাতে স্লো না হয়, তাই স্যাম্পল সংখ্যা ১০০০ রাখা হয়েছে
-    exp = explainer.explain_instance(text, predictor, num_features=10, labels=(label_idx,), num_samples=5000)
+    exp = explainer.explain_instance(text, predictor, num_features=10, labels=(label_idx,), num_samples=1000)
     weights = OrderedDict(exp.as_list(label=label_idx))
     
     lime_weights = pd.DataFrame({'words': list(weights.keys()), 'weights': list(weights.values())})