模型上线

spark/featureEngSk.py

@@ -1053,16 +1053,18 @@ def get_and_save_card_feature(itemEsFeatureDF, predictClickStaticFeatures, predi
     itemFeatureDF = itemFeature.toPandas()
     conn = getRedisConn()
     BATCH = 5000
-    Key = 'strategy:model:rank:widedeep:service:feature'
+    Key_TMP = 'strategy:model:rank:widedeep:service:feature:tmp'
     def concat_service_feature(row):
         row = row[1]
         return '|'.join(row[columns_used].astype(str))
 
     columns_used = list(filter(lambda c: c.startswith('ITEM_'), list(itemFeatureDF.columns)))
     print('columns_used: ', columns_used)
+    conn.delete(Key_TMP)
     for start in range(0, len(itemFeatureDF), BATCH):
-        conn.hmset(Key, {row[1]['card_id']: concat_service_feature(row) for row in itemFeatureDF.iloc[start: start + BATCH].iterrows()})
-
+        conn.hmset(Key_TMP, {row[1]['card_id']: concat_service_feature(row) for row in itemFeatureDF.iloc[start: start + BATCH].iterrows()})
+    Key = 'strategy:model:rank:widedeep:service:feature'
+    conn.rename(Key_TMP, Key)
 
 
 def get_and_save_device_feature(spark, fields_na_value_dict, days = 180):
@@ -1121,10 +1123,12 @@ def get_and_save_device_feature(spark, fields_na_value_dict, days = 180):
     device_feature_df = device_feature_df.na.fill(fields_na_value_dict)
     conn = getRedisConn()
     BATCH = 5000
-    Key = 'strategy:model:rank:widedeep:device:feature'
+    Key_TMP = 'strategy:model:rank:widedeep:device:feature:tmp'
+    conn.delete(Key_TMP)
     for start in range(0, len(device_feature_df), BATCH):
-        conn.hmset(Key, {row[1]['device_id']: row[1]['os'] for row in device_feature_df.iloc[start: start + BATCH].iterrows()})
-
+        conn.hmset(Key_TMP, {row[1]['device_id']: row[1]['os'] for row in device_feature_df.iloc[start: start + BATCH].iterrows()})
+    Key = 'strategy:model:rank:widedeep:device:feature'
+    conn.rename(Key_TMP, Key)
 
 
 def get_click_exp_rating_df(trainDays, spark):