Add log file

experiments/scripts/attack.py

@@ -12,6 +12,7 @@ from textfooler import Attack, TextFooler, Similarity, BaseLine, \
     process, run_queue, filter_similarity_queue, spoil_queue, \
     AttackMethod, get_xai_importance_diff
 from time import sleep, time
+from date import datetime
 from multiprocessing import Process
 from multiprocessing import Queue, Manager
 from threading import Thread
@@ -67,7 +68,8 @@ def join_punct(words):
     return "".join(w if set(w) <= punc else " " + w for w in words).lstrip()
 
 
-def data_producer(queue_out, dataset_df, queue_recurse, queue_log):
+def data_producer(queue_out, dataset_df, queue_recurse, queue_log, log_file):
+    try:
         for i, cols in tqdm(
                 dataset_df[[TEXT, ID, LEMMAS, TAGS, ORTHS, PRED, NER]].iterrows(), total=len(dataset_df)
         ):
@@ -92,10 +94,16 @@ def data_producer(queue_out, dataset_df, queue_recurse, queue_log):
                 for n in ner:
                     ners.extend(n[TEXT].split(" "))
                 queue_out.put([sentence, orths, ners, lemmas, tags, sent_id, y_pred, sub])
+    except Exception as e:
+        queue_log.put(f"Error in data producer: {e}")
+        with open(log_file, "a") as f:
+            f.write("Producer failed with {e}\n")
+        queue_out.put(None)
 
 
 def data_saver(queue_in, queue_log, queue_recurse, output_file,
-               output_dir, cases_nbr, queues_kill, to_kill_nbr, max_sub):
+               output_dir, cases_nbr, queues_kill, to_kill_nbr, max_sub, log_file):
+    try:
         processed_nbr, start = 0, time()
         item = 1
         test_y, pred_y = [], []
@@ -180,9 +188,15 @@ def data_saver(queue_in, queue_log, queue_recurse, output_file,
         }
         with open(f"{output_dir}/metrics.json", mode="w") as fd:
             json.dump(metrics, fd)
+    except Exception as e:
+        queue_log.put(f"Error in data saver: {e}")
+        with open(log_file, "a") as f:
+            f.write("Saver failed with {e}\n")
+        queue_in.put(None)
 
 
-def classify_queue(queue_in, queue_out, queue_log, dataset_name):
+def classify_queue(queue_in, queue_out, queue_log, dataset_name, log_file):
+    try:
         fun = getattr(
             importlib.import_module(f"text_attacks.models.{dataset_name}"),
             "get_classify_function",
@@ -207,6 +221,11 @@ def classify_queue(queue_in, queue_out, queue_log, dataset_name):
                 classified = classify_fun(sentences) if sentences else []
                 queue_out.put((sent_id, org_sentence, changed, y_pred, classified, synonyms_nbr, sent_words, subs))
                 queue_log.put(f"Classified sentences {sent_id}")
+    except Exception as e:
+        queue_log.put(f"Error in classifier: {e}")
+        with open(log_file, "a") as f:
+            f.write("Classifier failed with {e}\n")
+        queue_in.put(None)
 
 
 def log_queues(queues):
@@ -347,22 +366,23 @@ def main(dataset_name: str, attack_type: str):
     else:
         ft_model_name = None
 
-
+    log_file = f"{attack_type}_{datetime.now().strftime('%Y-%m-%d_%H-%M')}.log"
     processes_nbr = 12
     sim = Similarity(queues[5], similarity_bound, sent_model, menli_model, lang)
-    processes = [Process(target=data_producer, args=(queues[0], dataset_df, queues[4], queues[5]))]  # loading data file_in -> 0
+    processes = [Process(target=data_producer, args=(queues[0], dataset_df, queues[4], queues[5], log_file))]  # loading data file_in -> 0
 
     processes.extend([Process(target=spoil_queue, args=(queues[0], queues[1], queues[5],
-                                                        attack_type, params, ft_model_name))
+                                                        attack_type, params, ft_model_name, log_file))
                      for _ in range(processes_nbr)])  # spoiling 0 -> 1
 
-    processes.extend([Process(target=filter_similarity_queue, args=(queues[1], queues[2], queues[5], sim)),
-                      Process(target=filter_similarity_queue, args=(queues[1], queues[2], queues[5], sim)),  # cosim 1 -> 2
-                      Process(target=classify_queue, args=(queues[2], queues[3], queues[5], dataset_name)),
-                      Process(target=classify_queue, args=(queues[2], queues[3], queues[5], dataset_name)),
+    processes.extend([Process(target=filter_similarity_queue, args=(queues[1], queues[2], queues[5], sim, log_file)),
+                      Process(target=filter_similarity_queue, args=(queues[1], queues[2], queues[5], sim, log_file)),  # cosim 1 -> 2
+                      Process(target=classify_queue, args=(queues[2], queues[3], queues[5], dataset_name, log_file)),
+                      Process(target=classify_queue, args=(queues[2], queues[3], queues[5], dataset_name, log_file)),
                       # classify changed 2 -> 3
                       Process(target=data_saver, args=(queues[3], queues[5], queues[4], output_path,
-                                                       output_dir, len(dataset_df), queues, processes_nbr+6, max_sub)) # saving 3 -> file_out
+                                                       output_dir, len(dataset_df), queues, processes_nbr+6, max_sub,
+                                                       log_file)) # saving 3 -> file_out
                      ])
 
     [p.start() for p in processes]