Spark DataFrame 和重命名多列（Java）

Question

Is there any nicer way to prefix or rename all or multiple columns at the same time of a given SparkSQL DataFrame than calling multiple times dataFrame.withColumnRenamed() ?

An example would be if I want to detect changes (using full outer join). Then I'm left with two DataFrame s with the same structure.

Answer 1

I suggest to use the select() method to perform this. In fact withColumnRenamed() method uses select() by itself. Here is example how to rename multiple columns:

import org.apache.spark.sql.functions._

val someDataframe: DataFrame = ...

val initialColumnNames = Seq("a", "b", "c")
val renamedColumns = initialColumnNames.map(name => col(name).as(s"renamed_$name"))
someDataframe.select(renamedColumns : _*)

Answer 2

I think this method can help you.

public static Dataset<Row> renameDataFrame(Dataset<Row> dataset) {
    for (String column : dataset.columns()) {
        dataset = dataset.withColumnRenamed(column, SystemUtils.underscoreToCamelCase(column));
    }
    return dataset;
}

    public static String underscoreToCamelCase(String underscoreName) {
        StringBuilder result = new StringBuilder();
        if (underscoreName != null && underscoreName.length() > 0) {
            boolean flag = false;
            for (int i = 0; i < underscoreName.length(); i++) {
                char ch = underscoreName.charAt(i);
                if ("_".charAt(0) == ch) {
                    flag = true;
                } else {
                    if (flag) {
                        result.append(Character.toUpperCase(ch));
                        flag = false;
                    } else {
                        result.append(ch);
                    }
                }
            }
        }
        return result.toString();
    }

Answer 3

I heve just found the answer

df1_r = df1.select(*(col(x).alias(x + '_df1') for x in df1.columns))

at stackoverflow here (see the end of the accepted answer)

Answer 4

or (a <- 0 to newsales.columns.length - 1) 
{ 
 var new_c = newsales.columns(a).replace('(','_').replace(')',' ').trim  
 newsales_var = newsales.withColumnRenamed(newsales.columns(a),new_c) 
}

Answer 5

Although it does not answer your question directly, but I always update column names one by one. Since it updates only DF metadata, there is no harm (no performance impact) on updating column names one by one, eg:

for c in DF.columns:
    new_c = c.strip().replace(' ','_')
    DF = DF.withColumnRenamed(c, new_c)