JSON轉換為Python中的RDD

Question

我對Spark非常陌生，並且已經嘗試了一段時間以讓Spark理解我的JSON輸入，但是我沒有進行管理。 總之，我正在使用Spark的ALS算法來提出建議。 當我提供一個csv文件作為輸入時，一切正常。 但是，我的輸入實際上是一個json，如下所示：

all_user_recipe_rating = [{'rating': 1, 'recipe_id': 8798, 'user_id': 2108}, {'rating': 4, 'recipe_id': 6985, 'user_id': 4236}, {'rating': 4, 'recipe_id': 13572, 'user_id': 2743}, {'rating': 4, 'recipe_id': 6312, 'user_id': 3156}, {'rating': 1, 'recipe_id': 12836, 'user_id': 768}, {'rating': 1, 'recipe_id': 9237, 'user_id': 1599}, {'rating': 2, 'recipe_id': 16946, 'user_id': 2687}, {'rating': 2, 'recipe_id': 20728, 'user_id': 58}, {'rating': 4, 'recipe_id': 12921, 'user_id': 2221}, {'rating': 2, 'recipe_id': 10693, 'user_id': 2114}, {'rating': 2, 'recipe_id': 18301, 'user_id': 4898}, {'rating': 2, 'recipe_id': 9967, 'user_id': 3010}, {'rating': 2, 'recipe_id': 16393, 'user_id': 4830}, {'rating': 4, 'recipe_id': 14838, 'user_id': 583}]

ratings_RDD = self.spark.parallelize(all_user_recipe_rating)

ratings = ratings_RDD.map(lambda row:
  (Rating(int(row['user_id']),
   int(row['recipe_id']),
   float(row['rating']))))

model = self.build_model(ratings)

這是我在看到一些示例后想到的，但這是我得到的：

MatrixFactorizationModel: User factor is not cached. Prediction could be slow.
16/12/21 03:54:53 WARN MatrixFactorizationModel: Product factor does not have a partitioner. Prediction on individual records could be slow.
16/12/21 03:54:53 WARN MatrixFactorizationModel: Product factor is not cached. Prediction could be slow.
16/12/21 03:54:53 WARN MatrixFactorizationModelWrapper: User factor does not have a partitioner. Prediction on individual records could be slow.

和

File "/usr/local/spark/python/pyspark/mllib/recommendation.py", line 147, in <lambda>
user_product = user_product.map(lambda u_p: (int(u_p[0]), int(u_p[1])))
TypeError: int() argument must be a string or a number, not 'Rating'

有人可以幫我嗎？ ：） 謝謝！

Answer 1

好，

您的錯誤是由於一件事而發生的。

您所遭受的此異常是關於ALS函數的 predictAll 函數 。

這里的問題是，您正在嘗試將Rating對象發送給需要接收RDD<int, int>的函數

我獲取了您的代碼，並構建了所需的代碼：

>>> from pyspark.mllib.recommendation import Rating
>>> ratings = ratings_RDD.map(lambda row:
...   (Rating(int(row['user_id']),
...    int(row['recipe_id']),
...    float(row['rating']))))
>>> model = ALS.trainImplicit(ratings, 1, seed=10)
>>> to_predict = spark.parallelize([[2108, 16393], [583, 20728]])
>>> model.predictAll(to_predict).take(2)
[Rating(user=583, product=20728, rating=0.0741161997082127), Rating(user=2108, product=16393, rating=0.05669039815320609)]

您的JSON沒錯，在調用predictAll ，您遇到的問題是您發送的是Rating對象而不是RDD<int, int>