预测前加载模型，预测后删除模型

src/_data_lib.py

@@ -266,7 +266,7 @@ class AutoMaintainer(object):
     '''
     def __init__(self):
         self.pm = PostManager(max_workers=1)  # max=1 即为同步
-        self.model = MODEL_DICT['decision_tree_model']
+        self.model = None
 
 
         # 用[数据库里的datasource_id] 查询对应的[config].
@@ -312,6 +312,11 @@ def daily_model_predict(self):
         每日更新模型全量预测结果
         """
         # 拆成24个小时的数据运行
+
+        # 加载模型
+        MODEL_DICT.load_model('decision_tree_model_v2', path_prefix='../')
+        self.model = MODEL_DICT['decision_tree_model']
+
         self._model_predicted_result_pool = []
         for i in range(24):
             try:
@@ -377,6 +382,10 @@ def limit_cpu(interval):
                 messager.send_to_bot_shortcut('出现报错，详细信息为:')
                 messager.send_to_bot_shortcut(str(e))
 
+        # 删除模型。
+        MODEL_DICT.model_dict.pop('decision_tree_model')
+        self.model = None
+
     def get_post_data_list(self, pending_datasources_id_list, maintainer:Maintainer):
         """
 

src/auto_sche/model_loader.py

@@ -18,15 +18,15 @@ def __init__(self):
 
     def load_all_model(self):
         for name in [
-            'decision_tree_model_v2',
+            # 'decision_tree_model_v2',
             'weekday_encoder_v2',
             'datasource_encoder_v2'
         ]:
             self.load_model(name)
 
-    def load_model(self, name, suffix='.joblib'):
+    def load_model(self, name, suffix='.joblib', path_prefix=''):
         # 加载模型、编码器等等
-        loaded_model = load('./ml_model/{}{}'.format(name, suffix))
+        loaded_model = load('{}./ml_model/{}{}'.format(path_prefix, name, suffix))
         self.model_dict[name.split('_v')[0]] = loaded_model
 
     def __getitem__(self, key):