Question

我有一个spark聚合，我想将结果输出到csv，但是我发现spark总是以科学计数法输出大量的小数。我已经尝试过this question中提到的解决方案，但这也没有用。

预期输出：

foo,avg(bar)
a,0.0000002
b,0.0000001

实际输出：

foo,avg(bar)
a,2.0E-7
b,1.0E-7

请参见以下示例：

from os import path
import shutil
import glob
from pyspark.sql import SQLContext, functions as F, types

def test(sc):
    sq = SQLContext(sc)
    data = [("a", 1e-7), ("b", 1e-7), ("a", 3e-7)]
    df = sq.createDataFrame(data, ['foo', 'bar'])

    # 12 digits with 9 decimal places
    decType = types.DecimalType(precision=12, scale=9)

    # Cast both the column input and column output to Decimal
    aggs = [F.mean(F.col("bar").cast(decType)).cast(decType)]

    groups = [F.col("foo")]
    result = df.groupBy(*groups).agg(*aggs)
    write(result)
    return df, aggs, groups, result

def write(result):
    tmpDir = path.join("res", "tmp")
    config = {"sep": ","}
    result.write.format("csv")\
        .options(**config)\
        .save(tmpDir)

    # Once the distributed portion is done, write out to a single a file
    allFiles = glob.glob(path.join(tmpDir,"*.csv"))

    fullOut = path.join("res", "final.csv")
    with open(fullOut, 'wb') as wfd:
        # First write out the header row
        header = config.get("sep", ',').join(result.columns)
        wfd.write(header + "\n")
        for f in allFiles:
            with open(f, 'rb') as fd:
                shutil.copyfileobj(fd, wfd)
                pass
            pass
    shutil.rmtree(tmpDir)
    return

在pyspark外壳程序中：

import spark_test as t
t.test(sc)

Answer 1

>>> df1 = spark.createDataFrame([('a','2.0e-7'),('b','1e-5'),('c','1.0e-7')],['foo','avg'])
>>> df1.show()
+---+------+
|foo|   avg|
+---+------+
|  a|2.0e-7|
|  b|  1e-5|
|  c|1.0e-7|
+---+------+

>>> df1.select('foo','avg',format_string('%.7f',df1.avg.cast('float')).alias('converted')).show()
+---+------+---------+
|foo|   avg|converted|
+---+------+---------+
|  a|2.0e-7|0.0000002|
|  b|  1e-5|0.0000100|
|  c|1.0e-7|0.0000001|
+---+------+---------+

Answer 2

您是否尝试将汇总结果转换为String？这样，excel不会将值识别为十进制，因此不会给出科学的符号表示法。

PySpark-如何在csv输出中删除科学计数法

2 个答案: