乱序bert预测

bert/obtain_inst_vec.py

@@ -27,7 +27,8 @@ config_file = os.path.join('./bert/bert.json')
 def load_model():
     config = BertConfig.from_json_file(config_file)
     model = BertForPreTraining(config)
-    state_dict = torch.load(model_file)
+    # state_dict = torch.load(model_file)
+    state_dict = torch.load(model_file, map_location='cpu')
     model.load_state_dict(state_dict)
     model.eval()
 
@@ -103,9 +104,9 @@ def generate_inst_vec(inst, method="mean"):
         return result
 
 
-def bb2vec(inst):
+def bb2vec(item):
-    tmp = generate_inst_vec(inst, method="mean")
+    tmp = generate_inst_vec(item['opcode'], method="mean")
-    return list(np.mean(tmp.detach().numpy(), axis=0).astype(float))
+    return item['addr'], list(np.mean(tmp.detach().numpy(), axis=0).astype(float))
 
 
 if __name__ == "__main__":

json_feature2json.py

@@ -1,11 +1,13 @@
 import os
-from my_utils import save_json, load_json, setup_logger, multi_thread_order, THREAD_HALF,THREAD_FULL
+from my_utils import save_json, load_json, setup_logger, multi_thread_order, THREAD_HALF, THREAD_FULL
 from bert.obtain_inst_vec import bb2vec
 import multiprocessing
 from tqdm import tqdm
 import warnings
 from datetime import datetime
+import concurrent.futures
 
+warnings.filterwarnings("ignore")
 
 
 def addr2vec(base_file_path):
@@ -13,32 +15,56 @@ def addr2vec(base_file_path):
     file_name = str(os.path.basename(base_file_path))
     file_path = str(os.path.dirname(base_file_path))
 
-    # 忽略已生成的文件
-    if os.path.exists(os.path.join(file_path, 'final', file_name)):
-        return
-
     # 如果不是路径则开始转化
     if file_name:
+        # 忽略已生成的文件
+        if os.path.exists(os.path.join(file_path, 'final', file_name)):
+            process_bar.update(1)
+            return
         file_json = load_json(base_file_path)
         # 确保存在基础文件而不存在特征文件的情况
         feature_json = load_json(os.path.join(file_path, 'feature', file_name)) if os.path.exists(
             os.path.join(file_path, 'feature', file_name)) else None
         if feature_json is not None:
             feature_set = {}
-            for item in feature_json:
+
-                feature_set[item['addr']] = bb2vec(item['opcode'])
+            # 多线程预测bert
+            # with concurrent.futures.ThreadPoolExecutor(max_workers=os.cpu_count()) as executor:
+            #     try:
+            #         future_args = {
+            #             executor.submit(bb2vec, item): item for item in feature_json
+            #         }
+            #         for future in concurrent.futures.as_completed(future_args):
+            #             result = future.result()
+            #             feature_set[result[0]] = result[1]
+            #     except Exception as e:
+            #         logger.error(f"bert 解析出错 {file_name}，{e}")
+
+            with multiprocessing.Pool(processes=os.cpu_count()) as pool:
+                try:
+                    results = pool.imap_unordered(bb2vec, [item for item in feature_json])
+                    for result in results:
+                        feature_set[result[0]] = result[1]
+                except Exception as e:
+                    logger.error(f"bert 解析出错 {file_name}，{e}")
+
+            try:
                 for item in file_json['acfg_list']:
                     bb_feature_addr_list = item['block_features']
                     item['block_features'] = [feature_set[key] for key in bb_feature_addr_list]
+            except Exception as e:
+                logger.error(f"地址对应出错{file_name}, {e}")
             save_json(os.path.join(file_path, 'final', file_name), file_json)
+
         else:
             logger.error(f'文件{file_name}不存在特征文件')
+    process_bar.update(1)
     return
 
 
 if __name__ == '__main__':
     logger = setup_logger('feature2json', './log/feature2json.log')
-    sample_type = 'malware'
+    sample_type = 'benign'
     # json_path = os.path.join(f'./out/json/{sample_type}')
     json_path = os.path.join(f'./out/json/{sample_type}')
     json_files = os.listdir(json_path)
@@ -48,10 +74,9 @@ if __name__ == '__main__':
     # with multiprocessing.Pool(processes=os.cpu_count()) as pool:
     #     result = list(tqdm(pool.imap_unordered(addr2vec, [os.path.join(json_path, file) for file in json_files[:1] if os.path.isfile(os.path.join(json_path, file))]),
     #                        total=len(json_files)))
-    multi_thread_order(addr2vec, [os.path.join(json_path, file) for file in json_files[::-1] if os.path.isfile(os.path.join(json_path, file))], thread_num=THREAD_FULL)
+    # multi_thread_order(addr2vec, [os.path.join(json_path, file) for file in json_files if
-
+    #                               os.path.isfile(os.path.join(json_path, file))], thread_num=THREAD_FULL)
-
+    process_bar = tqdm(total=len(json_files))
-
+    for json_file in json_files:
-
+        if os.path.isfile(os.path.join(json_path, json_file)):
-    # for json_file in json_files:
+            addr2vec(os.path.join(json_path, json_file))
-    #     addr2vec(json_path, json_file)