如何在我的 spark 数据帧中添加来自 spark 数据帧的列（使用 Pyspark）？

Question

我有两个 spark 数据帧，我想将一列从一个 spark 数据帧添加到另一个。

我的代码是：

new = df.withColumn("prob", tr_df.prob)

在这里，我想将tr_df 中的列result2添加到名为prob 的数据帧df 中。 我搜索了这个，但没有任何效果对我有用，我收到了一个错误——

AnalysisException: u'resolved attribute(s) prob#579 missing from q1_n_words#388L,prediction#510,res1#390,q2_n_words#389L,tfidf_word_match#384,Average#379,prob#385,probability#485,Cosine#381,word_m#383,rawPrediction#461,features#438,res2#391,question1#373,Jaccard#382,test_id#372L,raw_pred#377,question2#374,q2len#376,Common#378L,result2#387,q1len#375,result1#386,Percentage#380 in operator !Project [test_id#372L, question1#373, question2#374, q1len#375, q2len#376, raw_pred#377, Common#378L, Average#379, Percentage#380, Cosine#381, Jaccard#382, word_m#383, tfidf_word_match#384, prob#579 AS prob#634, result1#386, result2#387, q1_n_words#388L, q2_n_words#389L, res1#390, res2#391, features#438, rawPrediction#461, probability#485, prediction#510];;\n!Project [test_id#372L, question1#373, question2#374, q1len#375, q2len#376, raw_pred#377, Common#378L, Average#379, Percentage#380, Cosine#381, Jaccard#382, word_m#383, tfidf_word_match#384, prob#579 AS prob#634, result1#386, result2#387, q1_n_words#388L, q2_n_words#389L, res1#390, res2#391, features#438, rawPrediction#461, probability#485, prediction#510]\n+- Project [test_id#372L, question1#373, question2#374, q1len#375, q2len#376, raw_pred#377, Common#378L, Average#379, Percentage#380, Cosine#381, Jaccard#382, word_m#383, tfidf_word_match#384, prob#385, result1#386, result2#387, q1_n_words#388L, q2_n_words#389L, res1#390, res2#391, features#438, rawPrediction#461, probability#485, UDF(rawPrediction#461) AS prediction#510]\n   +- Project [test_id#372L, question1#373, question2#374, q1len#375, q2len#376, raw_pred#377, Common#378L, Average#379, Percentage#380, Cosine#381, Jaccard#382, word_m#383, tfidf_word_match#384, prob#385, result1#386, result2#387, q1_n_words#388L, q2_n_words#389L, res1#390, res2#391, features#438, rawPrediction#461, UDF(rawPrediction#461) AS probability#485]\n      +- Project [test_id#372L, question1#373, question2#374, q1len#375, q2len#376, raw_pred#377, Common#378L, Average#379, Percentage#380, Cosine#381, Jaccard#382, word_m#383, tfidf_word_match#384, prob#385, result1#386, result2#387, q1_n_words#388L, q2_n_words#389L, res1#390, res2#391, features#438, UDF(features#438) AS rawPrediction#461]\n         +- Project [test_id#372L, question1#373, question2#374, q1len#375, q2len#376, raw_pred#377, Common#378L, Average#379, Percentage#380, Cosine#381, Jaccard#382, word_m#383, tfidf_word_match#384, prob#385, result1#386, result2#387, q1_n_words#388L, q2_n_words#389L, res1#390, res2#391, UDF(struct(q1len#375, q2len#376, cast(q1_n_words#388L as double) AS q1_n_words_double_VectorAssembler_4158baa8e5b4f3aced2b#435, cast(q2_n_words#389L as double) AS q2_n_words_double_VectorAssembler_4158baa8e5b4f3aced2b#436, cast(Common#378L as double) AS Common_double_VectorAssembler_4158baa8e5b4f3aced2b#437, Average#379, Percentage#380, Cosine#381, Jaccard#382, word_m#383, prob#385, raw_pred#377, res1#390, res2#391)) AS features#438]\n            +- LogicalRDD [test_id#372L, question1#373, question2#374, q1len#375, q2len#376, raw_pred#377, Common#378L, Average#379, Percentage#380, Cosine#381, Jaccard#382, word_m#383, tfidf_word_match#384, prob#385, result1#386, result2#387, q1_n_words#388L, q2_n_words#389L, res1#390, res2#391]\n'

tr_df 架构 --

tr_df.printSchema()
root
 |-- prob: float (nullable = true)

df 架构——

df.printSchema()
root
 |-- test_id: long (nullable = true)

请帮忙！ 提前致谢。

Answer 1

由于错误消息明确指出您需要将 spark.sql.crossJoin.enabled = true 设置为您的 spark 配置

您可以设置如下相同的内容：

val sparkConf = new SparkConf().setAppName("Test") 
sparkConf.set("spark.sql.crossJoin.enabled", "true")

然后通过传递这个 SparkConf 来获取或创建 SparkSession

val sparkSession = SparkSession.builder().config(sparkConf).getOrCreate()

然后加入...

来源： 如何在 Spark 2.0 中启用笛卡尔连接？

Answer 2

在 pyspark 中，您可以通过以下方式执行此操作。 希望它会很有用。

>>> spark.conf.set("spark.sql.crossJoin.enabled", True)
>>> df1.show()
+----+
|col1|
+----+
|  23|
|  56|
|  78|
|  31|
+----+

>>> df2.show()
+----+
|col2|
+----+
|  87|
|  45|
|  23|
|  11|
+----+

>>> final = df1.crossJoin(df2)
>>> final.withColumnRenamed('col2', 'result').show()
+----+------+                                                                   
|col1|result|
+----+------+
|  23|    87|
|  23|    45|
|  23|    23|
|  23|    11|
|  56|    87|
|  56|    45|
|  56|    23|
|  56|    11|
|  78|    87|
|  78|    45|
|  78|    23|
|  78|    11|
|  31|    87|
|  31|    45|
|  31|    23|
|  31|    11|
+----+------+