Question

我使用Spark的推荐系统。

在训练模型之后，我做了以下代码以获得推荐 model.recommendProductsForUsers（2）

[(10000, (Rating(user=10000, product=14780773, rating=7.35695469892999e-05), 
          Rating(user=10000, product=17229476, rating=5.648606256948921e-05))), 
 (0, (Rating(user=0, product=16750010, rating=0.04405213492474741), 
      Rating(user=0, product=17416511, rating=0.019491942665715176))), 
 (20000, (Rating(user=20000, product=17433348, rating=0.017938298063142653), 
          Rating(user=20000, product=17333969, rating=0.01505112418739887)))]

在这种情况下，Rec是RDD，请参见下文。

>>> type(Rec)
<class 'pyspark.rdd.RDD'>

如何将此信息放在像

这样的数据框中

 User | Product   | Rating 
1000  |  14780773 | 7.3e-05
1000  |  17229675 | 5.6e-05
(...)     (...)     (...) 
2000  |  17333969 | 0.015

谢谢你的时间

Answer 1

要验证，我使用以下pyspark代码重现您的RDD：

from pyspark.mllib.recommendation import Rating

Rec = sc.parallelize([(10000, (Rating(user=10000, product=14780773, rating=7.35695469892999e-05), 
                               Rating(user=10000, product=17229476, rating=5.648606256948921e-05))), 
                      (0, (Rating(user=0, product=16750010, rating=0.04405213492474741), 
                           Rating(user=0, product=17416511, rating=0.019491942665715176))), 
                      (20000, (Rating(user=20000, product=17433348, rating=0.017938298063142653), 
                               Rating(user=20000, product=17333969, rating=0.01505112418739887)))])

此RDD由键值对组成，每个值由具有Rating元组的记录组成。您需要映射RDD以仅保留记录，然后将结果分解为每个推荐都有单独的元组。 flatMap(f)函数会压缩这两个步骤：

flatRec = Rec.flatMap(lambda p: p[1])

导致形式为RDD：

[Rating(user=10000, product=14780773, rating=7.35695469892999e-05),
 Rating(user=10000, product=17229476, rating=5.648606256948921e-05),
 Rating(user=0, product=16750010, rating=0.04405213492474741),
 Rating(user=0, product=17416511, rating=0.019491942665715176),
 Rating(user=20000, product=17433348, rating=0.017938298063142653),
 Rating(user=20000, product=17333969, rating=0.01505112418739887)]

现在只需使用createDataFrame函数将其转换为DataFrame即可。每个评级元组都将转换为数据框架行，并且由于项目已标记，因此您无需指定架构。

recDF = sqlContext.createDataFrame(flatRec).show()

这将输出以下内容：

+-----+--------+--------------------+
| user| product|              rating|
+-----+--------+--------------------+
|10000|14780773| 7.35695469892999E-5|
|10000|17229476|5.648606256948921E-5|
|    0|16750010| 0.04405213492474741|
|    0|17416511|0.019491942665715176|
|20000|17433348|0.017938298063142653|
|20000|17333969| 0.01505112418739887|
+-----+--------+--------------------+

PySpark - ALS输出中的RDD到DataFrame

1 个答案: