Speedup tests

utils.py

@@ -33,6 +33,8 @@ all_target_languages = list(test_dataset.unique("dialect"))
 supported_dialects = all_target_languages + ['All']
 languages_to_display_one_vs_all = all_target_languages # everything except All
 
+print(f'all_target_languages: {all_target_languages}')
+
 metrics = [
     'f1_score',
     'precision',
@@ -229,12 +231,8 @@ def make_binary(dialect, target):
         return 'Other'
     return target
 
-def run_eval_one_vs_all(model, data_test, TARGET_LANG='Morocco', language_mapping_dict=None, use_mapping=False):
+def run_eval_one_vs_all(data_test, TARGET_LANG='Morocco'):
     
-    # Predict labels using the model
-    print(f"[INFO] Running predictions...")
-    data_test['preds'] = data_test['text'].apply(lambda text: predict_label(text, model, language_mapping_dict, use_mapping=use_mapping))
-
     # map to binary
     df_test_preds = data_test.copy()
     df_test_preds.loc[df_test_preds['dialect'] == TARGET_LANG, 'dialect'] = TARGET_LANG
@@ -311,17 +309,20 @@ def handle_evaluation(model_path, model_path_bin, use_mapping=False):
     print(f"[INFO] Converting evaluation dataset to Pandas DataFrame...")
     df_eval = pd.DataFrame(eval_dataset)
     
+    # Predict labels using the model
+    print(f"[INFO] Running predictions...")
+    df_eval['preds'] = df_eval['text'].apply(lambda text: predict_label(text, model, language_mapping_dict, use_mapping=use_mapping))
+    
     # run the evaluation
-    result_df, _ = run_eval(model, df_eval, language_mapping_dict, use_mapping=use_mapping)
+    result_df, _ = run_eval(df_eval)
     # set the model name
     model_name = model_path + '/' + model_path_bin
     
     # update the multilingual leaderboard
     update_darija_multilingual_leaderboard(result_df, model_name, MULTILINGUAL_LEADERBOARD_FILE)
     
-    # # TODO
     for target_lang in all_target_languages:
-        result_df_one_vs_all =run_eval_one_vs_all(model, df_eval, TARGET_LANG=target_lang, language_mapping_dict=language_mapping_dict, use_mapping=use_mapping)
+        result_df_one_vs_all =run_eval_one_vs_all(df_eval, TARGET_LANG=target_lang)
         update_darija_one_vs_all_leaderboard(result_df_one_vs_all, model_name, target_lang, BINARY_LEADERBOARD_FILE)
     
     # load the updated leaderboard tables
@@ -332,7 +333,7 @@ def handle_evaluation(model_path, model_path_bin, use_mapping=False):
     
     return create_leaderboard_display_multilingual(df_multilingual, target_label, default_metrics), status_message
 
-def run_eval(model, df_eval, language_mapping_dict=None, use_mapping=False):
+def run_eval(df_eval):
     """Run evaluation on a dataset and compute metrics.
 
     Args:
@@ -346,20 +347,19 @@ def run_eval(model, df_eval, language_mapping_dict=None, use_mapping=False):
         pd.DataFrame: A DataFrame containing evaluation metrics.
     """
 
-    # Predict labels using the model
-    print(f"[INFO] Running predictions...")
-    df_eval['preds'] = df_eval['text'].apply(lambda text: predict_label(text, model, language_mapping_dict, use_mapping=use_mapping))
+    # map to binary
+    df_eval_multilingual = df_eval.copy()
     
     # now drop the columns that are not needed, i.e. 'text'
-    df_eval = df_eval.drop(columns=['text', 'metadata', 'dataset_source'])
+    df_eval_multilingual = df_eval_multilingual.drop(columns=['text', 'metadata', 'dataset_source'])
 
     # Compute evaluation metrics
     print(f"[INFO] Computing metrics...")
-    result_df, _ = compute_classification_metrics(df_eval)
+    result_df, _ = compute_classification_metrics(df_eval_multilingual)
     
     # update_darija_multilingual_leaderboard(result_df, model_path, MULTILINGUAL_LEADERBOARD_FILE)
     
-    return result_df, df_eval  
+    return result_df, df_eval_multilingual  
 
 def process_results_file(file, uploaded_model_name, base_path_save="./atlasia/submissions/", default_language='Morocco'):
     try: