Update app.py
Browse files
app.py
CHANGED
|
@@ -203,20 +203,27 @@ def clean_answer(answer):
|
|
| 203 |
return clean[0].upper() if clean else None
|
| 204 |
|
| 205 |
def update_leaderboard(results):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
try:
|
| 207 |
-
new_entry = {
|
| 208 |
-
"Model Name": results['model_name'],
|
| 209 |
-
"Overall Accuracy": round(results['overall_accuracy'] * 100, 2),
|
| 210 |
-
"Valid Accuracy": round(results['valid_accuracy'] * 100, 2),
|
| 211 |
-
"Correct Predictions": results['correct_predictions'],
|
| 212 |
-
"Total Questions": results['total_questions'],
|
| 213 |
-
"Timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
|
| 214 |
-
}
|
| 215 |
new_entry_df = pd.DataFrame([new_entry])
|
| 216 |
-
new_entry_df.to_csv(
|
| 217 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 218 |
except Exception as e:
|
| 219 |
-
print(f"Error
|
|
|
|
| 220 |
|
| 221 |
|
| 222 |
def load_leaderboard():
|
|
@@ -274,6 +281,7 @@ def evaluate_predictions(prediction_file, model_name, add_to_leaderboard):
|
|
| 274 |
return "Evaluation completed and added to leaderboard.", load_leaderboard()
|
| 275 |
else:
|
| 276 |
return "Evaluation completed but not added to leaderboard.", load_leaderboard()
|
|
|
|
| 277 |
except Exception as e:
|
| 278 |
return f"Error during evaluation: {str(e)}", load_leaderboard()
|
| 279 |
|
|
|
|
| 203 |
return clean[0].upper() if clean else None
|
| 204 |
|
| 205 |
def update_leaderboard(results):
|
| 206 |
+
new_entry = {
|
| 207 |
+
"Model Name": results['model_name'],
|
| 208 |
+
"Overall Accuracy": round(results['overall_accuracy'] * 100, 2),
|
| 209 |
+
"Valid Accuracy": round(results['valid_accuracy'] * 100, 2),
|
| 210 |
+
"Correct Predictions": results['correct_predictions'],
|
| 211 |
+
"Total Questions": results['total_questions'],
|
| 212 |
+
"Timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
|
| 213 |
+
}
|
| 214 |
+
|
| 215 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 216 |
new_entry_df = pd.DataFrame([new_entry])
|
| 217 |
+
new_entry_df.to_csv(
|
| 218 |
+
LEADERBOARD_FILE,
|
| 219 |
+
mode='a', # Append mode
|
| 220 |
+
index=False,
|
| 221 |
+
header=not os.path.exists(LEADERBOARD_FILE) # Add headers if file doesn't exist
|
| 222 |
+
)
|
| 223 |
+
print("Leaderboard file updated successfully.")
|
| 224 |
except Exception as e:
|
| 225 |
+
print(f"Error writing to leaderboard file: {e}")
|
| 226 |
+
|
| 227 |
|
| 228 |
|
| 229 |
def load_leaderboard():
|
|
|
|
| 281 |
return "Evaluation completed and added to leaderboard.", load_leaderboard()
|
| 282 |
else:
|
| 283 |
return "Evaluation completed but not added to leaderboard.", load_leaderboard()
|
| 284 |
+
|
| 285 |
except Exception as e:
|
| 286 |
return f"Error during evaluation: {str(e)}", load_leaderboard()
|
| 287 |
|