Spaces:
Sleeping
Sleeping
Enhance prediction function with validation checks and improved error handling
Browse files- utils/prediction.py +26 -26
utils/prediction.py
CHANGED
@@ -46,43 +46,43 @@ def predict_sentence(model, sentence, tokenizer, label_encoder):
|
|
46 |
"""
|
47 |
Make prediction for a single sentence with label validation.
|
48 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
model.eval()
|
50 |
|
51 |
# Tokenize
|
52 |
-
encoding = tokenizer(
|
53 |
-
sentence,
|
54 |
-
add_special_tokens=True,
|
55 |
-
max_length=512,
|
56 |
-
padding='max_length',
|
57 |
-
truncation=True,
|
58 |
-
return_tensors='pt'
|
59 |
-
)
|
60 |
-
|
61 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
with torch.no_grad():
|
63 |
-
# Get model outputs
|
64 |
outputs = model(encoding['input_ids'], encoding['attention_mask'])
|
65 |
probabilities = torch.softmax(outputs, dim=1)
|
66 |
-
|
67 |
-
# Get prediction and probability
|
68 |
prob, pred_idx = torch.max(probabilities, dim=1)
|
|
|
|
|
69 |
|
70 |
-
# Validate prediction index
|
71 |
-
if pred_idx.item() >= len(label_encoder.classes_):
|
72 |
-
print(f"Warning: Model predicted invalid label index {pred_idx.item()}")
|
73 |
-
return "Unknown", 0.0
|
74 |
-
|
75 |
-
# Convert to label
|
76 |
-
try:
|
77 |
-
predicted_class = label_encoder.classes_[pred_idx.item()]
|
78 |
-
return predicted_class, prob.item()
|
79 |
-
except IndexError:
|
80 |
-
print(f"Warning: Invalid label index {pred_idx.item()}")
|
81 |
-
return "Unknown", 0.0
|
82 |
-
|
83 |
except Exception as e:
|
84 |
print(f"Prediction error: {str(e)}")
|
85 |
-
return "Error", 0.0
|
86 |
|
87 |
def print_labels(label_encoder, show_counts=False):
|
88 |
"""Print all labels and their corresponding indices"""
|
|
|
46 |
"""
|
47 |
Make prediction for a single sentence with label validation.
|
48 |
"""
|
49 |
+
# Validation checks
|
50 |
+
if model is None:
|
51 |
+
print("Error: Model not loaded")
|
52 |
+
return "Error: Model not loaded", 0.0
|
53 |
+
if tokenizer is None:
|
54 |
+
print("Error: Tokenizer not loaded")
|
55 |
+
return "Error: Tokenizer not loaded", 0.0
|
56 |
+
if label_encoder is None:
|
57 |
+
print("Error: Label encoder not loaded")
|
58 |
+
return "Error: Label encoder not loaded", 0.0
|
59 |
+
|
60 |
+
# Force CPU device
|
61 |
+
device = torch.device('cpu')
|
62 |
+
model = model.to(device)
|
63 |
model.eval()
|
64 |
|
65 |
# Tokenize
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
try:
|
67 |
+
encoding = tokenizer(
|
68 |
+
sentence,
|
69 |
+
add_special_tokens=True,
|
70 |
+
max_length=512,
|
71 |
+
padding='max_length',
|
72 |
+
truncation=True,
|
73 |
+
return_tensors='pt'
|
74 |
+
).to(device)
|
75 |
+
|
76 |
with torch.no_grad():
|
|
|
77 |
outputs = model(encoding['input_ids'], encoding['attention_mask'])
|
78 |
probabilities = torch.softmax(outputs, dim=1)
|
|
|
|
|
79 |
prob, pred_idx = torch.max(probabilities, dim=1)
|
80 |
+
predicted_label = label_encoder.classes_[pred_idx.item()]
|
81 |
+
return predicted_label, prob.item()
|
82 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
83 |
except Exception as e:
|
84 |
print(f"Prediction error: {str(e)}")
|
85 |
+
return f"Error: {str(e)}", 0.0
|
86 |
|
87 |
def print_labels(label_encoder, show_counts=False):
|
88 |
"""Print all labels and their corresponding indices"""
|