Merge branch 'master' of git.wanmeizhensuo.com:ML/ffm-baseline

add a py

predictDiary.py

@@ -23,8 +23,6 @@ def feature_en(user_profile):
     # 虽然预测y，但ffm转化需要y，并不影响预测结果
     data["y"] = 0
     data = data.drop("city_id", axis=1)
-    print(data.head(1))
-    print("特征工程处理结束")
     return data
 
 
@@ -89,6 +87,19 @@ def router(device_id):
         predict(user_profile)
 
 
+# 多进程预测
+def multi_predict(predict_list,processes=12):
+    pool = Pool(processes)
+    for device_id in predict_list:
+        start = time.time()
+        pool.apply_async(router, (device_id,))
+        end = time.time()
+        print("该用户{}预测耗时{}秒".format(device_id, (end - start)))
+
+    pool.close()
+    pool.join()
+
+
 if __name__ == "__main__":
     # TODO 如果耗时小于一分钟，下一次取到的device_id和上一次相同
     while True:
@@ -100,14 +111,13 @@ if __name__ == "__main__":
             time.sleep(58)
         else:
             old_device_id_list = pd.read_csv(DIRECTORY_PATH + "data_set_device_id.csv")["device_id"].values.tolist()
-            for device_id in device_id_list:
-                if device_id in old_device_id_list:
-                    start = time.time()
-                    router(device_id)
-                    end = time.time()
-                    print("该用户预测耗时{}秒".format(end - start))
-                else:
-                    print("该用户不是老用户，不能预测")
+            # 求活跃用户和老用户的交集，也就是只预测老用户
+            predict_list = list(set(device_id_list) & set(old_device_id_list))
+            multi_predict(predict_list)
+
+
+
+
 
 
 

processData.py

@@ -75,7 +75,7 @@ def ffm_transform(data, test_number, validation_number):
     print("Start ffm transform")
     start = time.time()
     ffm_train = multiFFMFormatPandas()
-    data = ffm_train.fit_transform(data, y='y',n=200000,processes=8)
+    data = ffm_train.fit_transform(data, y='y',n=100000,processes=6)
     with open(DIRECTORY_PATH+"ffm.pkl", "wb") as f:
         pickle.dump(ffm_train, f)