当Spark Scala数据帧中的多列数组包含彼此对齐的数组时，如何分解这些列？
当Spark Scala数据帧中的多列数组包含彼此对齐的数组时，如何分解这些列？
提问于 2020-02-05 04:28:33
我在阐述这个问题时有点问题，但我会试着解释一下。我知道如何分解数组的单列，但我有多个数组列，这些数组在索引值方面彼此对齐。在我的dataframe中，分解每一列基本上只是做一个无用的交叉连接，导致几十个无效的行。因此，我将从显示数据开始。
这显示了来自SparkNLP的一些结果，其中包含一些文本和文本的四组特征。从tr到nr的每一列都包含一个数组。这些数组中的每一个都与其他数组对齐。
+--+---------------------+---------------------+----------------------+--------------------+--------------------+
|ID| text| tr| lr| pr| nr|
+--+---------------------+---------------------+----------------------+--------------------+--------------------+
|10| thing: MacKay rolls|[thing, :, MacKay,...|[thing, :, MacKay, ...| [NN, :, NNP, NNS]| [O, O, I-PER, O]|
|11|thing: MacKay roll...|[thing, :, MacKay,...|[thing, :, MacKay, ...|[NN, :, NNP, NNS,...|[O, O, I-PER, O, ...|
|12| * I would like to...| [*, I, would, lik...| [*, I, would, lik...|[NN, PRP, MD, VB,...|[O, O, O, O, O, O...|
+--+---------------------+---------------------+----------------------+--------------------+--------------------+我想要的是一个新的dataframe，其中包含ID和文本，以及一行中所有数组中的每个第i项，如上面的dataframe所示：
+--+---------------------+---------------------+----------------------+--------------------+--------------------+------+-------+---+-----+
|ID| text| tr| lr| pr| nr| token| lemma|pos| ner|
+--+---------------------+---------------------+----------------------+--------------------+--------------------+------+-------+---+-----+
|10| thing: MacKay rolls|[thing, :, MacKay,...|[thing, :, MacKay, ...| [NN, :, NNP, NNS]| [O, O, I-PER, O]| thing| thing| NN| O|
|10| thing: MacKay rolls|[thing, :, MacKay,...|[thing, :, MacKay, ...| [NN, :, NNP, NNS]| [O, O, I-PER, O]| :| :| :| O|
|10| thing: MacKay rolls|[thing, :, MacKay,...|[thing, :, MacKay, ...| [NN, :, NNP, NNS]| [O, O, I-PER, O]|MacKay| MacKay|NNP|I-PER|
|10| thing: MacKay rolls|[thing, :, MacKay,...|[thing, :, MacKay, ...| [NN, :, NNP, NNS]| [O, O, I-PER, O]| rolls| roll|NNS| O|
|11|thing: MacKay roll...|[thing, :, MacKay,...|[thing, :, MacKay, ...|[NN, :, NNP, NNS,...|[O, O, I-PER, O, ...| thing| thing| NN| O|
|11|thing: MacKay roll...|[thing, :, MacKay,...|[thing, :, MacKay, ...|[NN, :, NNP, NNS,...|[O, O, I-PER, O, ...| :| :| :| O|
|11|thing: MacKay roll...|[thing, :, MacKay,...|[thing, :, MacKay, ...|[NN, :, NNP, NNS,...|[O, O, I-PER, O, ...|MacKay| MacKay|NNP|I-PER|
|11|thing: MacKay roll...|[thing, :, MacKay,...|[thing, :, MacKay, ...|[NN, :, NNP, NNS,...|[O, O, I-PER, O, ...| roll| roll|NNS| O|
|11|...
...
|12| * I would like to...| [*, I, would, lik...| [*, I, would, lik...|[NN, PRP, MD, VB,...|[O, O, O, O, O, O...| *| *| NN| O|
|12| * I would like to...| [*, I, would, lik...| [*, I, would, lik...|[NN, PRP, MD, VB,...|[O, O, O, O, O, O...| I| I|PRP| O|
|12| * I would like to...| [*, I, would, lik...| [*, I, would, lik...|[NN, PRP, MD, VB,...|[O, O, O, O, O, O...| would| would| MD| O|
|12| * I would like to...| [*, I, would, lik...| [*, I, would, lik...|[NN, PRP, MD, VB,...|[O, O, O, O, O, O...| like| like| VB| O|
|12| * I would like to...| [*, I, would, lik...| [*, I, would, lik...|[NN, PRP, MD, VB,...|[O, O, O, O, O, O...| to| ...|...| O|
|12|...
...
+--+---------------------+---------------------+----------------------+--------------------+--------------------+------+-------+---+-----+我不需要输出中的tr到nr列，但为了清楚起见，我把它们留了下来。
有没有办法做到这一点？
此外，是否还有一种方法可以同时提取数组索引(添加到输出行)？
回答 1
Stack Overflow用户
发布于 2020-11-11 11:07:33
在这个场景中，您想要做的是使用withColumn表达式分解单个列。假设您已将数据集加载为初始数据帧df。现在，您希望实现如下所示的功能。
val df = <load initial dataset>
val df1 = df.select($"id", $"text",$"tr", $"lr", $"pr", $"nr").withColumn("tr", explode($"tr")).withColumn("lr",explode($"lr")).withColumn("pr",explode($"pr")).withColumn("nr",explode($"nr"))这将导致将数组值添加到用ID & Text标记的记录中。这种方法的缺点是增加了记录计数和非数组列的重复。
页面原文内容由Stack Overflow提供。腾讯云小微IT领域专用引擎提供翻译支持
原文链接：
https://stackoverflow.com/questions/60065085
复制相关文章
相似问题
腾讯云开发者
