pyspark fill missing date，填補缺失日期數據

運行結果：

Using Spark's default log4j profile: org/apache/spark/log4j-defaults.properties
Setting default log level to "WARN".
To adjust logging level use sc.setLogLevel(newLevel).
18/04/25 18:27:06 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Original Data
+----+----------+--------+
|col1|      date|quantity|
+----+----------+--------+
|   b|2016-09-10|       1|
|   a|2016-09-11|       2|
|   b|2016-09-14|       6|
|   a|2016-09-16|       1|
|   b|2016-09-17|       4|
|   a|2016-09-20|       2|
+----+----------+--------+

After mark interval date
+----+----------+--------+----+------------------------------------------------+
|col1|date      |quantity|diff|next_dates                                      |
+----+----------+--------+----+------------------------------------------------+
|a   |2016-09-11|2       |5   |[2016-09-12, 2016-09-13, 2016-09-14, 2016-09-15]|
|a   |2016-09-16|1       |4   |[2016-09-17, 2016-09-18, 2016-09-19]            |
|b   |2016-09-10|1       |4   |[2016-09-11, 2016-09-12, 2016-09-13]            |
|b   |2016-09-14|6       |3   |[2016-09-15, 2016-09-16]                        |
+----+----------+--------+----+------------------------------------------------+

convert every list to rows
+----+----------+--------+----+------------------------------------------------+
|col1|date      |quantity|diff|next_dates                                      |
+----+----------+--------+----+------------------------------------------------+
|a   |2016-09-12|0       |5   |[2016-09-12, 2016-09-13, 2016-09-14, 2016-09-15]|
|a   |2016-09-13|0       |5   |[2016-09-12, 2016-09-13, 2016-09-14, 2016-09-15]|
|a   |2016-09-14|0       |5   |[2016-09-12, 2016-09-13, 2016-09-14, 2016-09-15]|
|a   |2016-09-15|0       |5   |[2016-09-12, 2016-09-13, 2016-09-14, 2016-09-15]|
|a   |2016-09-17|0       |4   |[2016-09-17, 2016-09-18, 2016-09-19]            |
|a   |2016-09-18|0       |4   |[2016-09-17, 2016-09-18, 2016-09-19]            |
|a   |2016-09-19|0       |4   |[2016-09-17, 2016-09-18, 2016-09-19]            |
|b   |2016-09-11|0       |4   |[2016-09-11, 2016-09-12, 2016-09-13]            |
|b   |2016-09-12|0       |4   |[2016-09-11, 2016-09-12, 2016-09-13]            |
|b   |2016-09-13|0       |4   |[2016-09-11, 2016-09-12, 2016-09-13]            |
|b   |2016-09-15|0       |3   |[2016-09-15, 2016-09-16]                        |
|b   |2016-09-16|0       |3   |[2016-09-15, 2016-09-16]                        |
+----+----------+--------+----+------------------------------------------------+

union missing date into original data
+----+----------+--------+
|col1|      date|quantity|
+----+----------+--------+
|   a|2016-09-11|       2|
|   a|2016-09-12|       0|
|   a|2016-09-13|       0|
|   a|2016-09-14|       0|
|   a|2016-09-15|       0|
|   a|2016-09-16|       1|
|   a|2016-09-17|       0|
|   a|2016-09-18|       0|
|   a|2016-09-19|       0|
|   a|2016-09-20|       2|
|   b|2016-09-10|       1|
|   b|2016-09-11|       0|
|   b|2016-09-12|       0|
|   b|2016-09-13|       0|
|   b|2016-09-14|       6|
|   b|2016-09-15|       0|
|   b|2016-09-16|       0|
|   b|2016-09-17|       4|
+----+----------+--------+

Compare to original data
+----+----------+--------+
|col1|      date|quantity|
+----+----------+--------+
|   a|2016-09-11|       2|
|   a|2016-09-16|       1|
|   a|2016-09-20|       2|
|   b|2016-09-10|       1|
|   b|2016-09-14|       6|
|   b|2016-09-17|       4|
+----+----------+--------+

                                                                                
Process finished with exit code 0

代碼：

if __name__ == '__main__':

    conf = SparkConf()
    sparkSession = SparkSession.builder.appName("Test PredictionTool").config(conf=conf).getOrCreate()
    sc = sparkSession.sparkContext
    sc.setLogLevel("ERROR")

    dfp = pd.DataFrame({'date': ['2016-09-10 00:00:00',
                                 '2016-09-11 00:00:00',
                                 '2016-09-14 00:00:00',
                                 '2016-09-16 00:00:00',
                                 '2016-09-17 00:00:00',
                                 '2016-09-20 00:00:00'],
                        'quantity': [1, 2, 6, 1, 4, 2],
                        'col1': ['b', 'a', 'b', 'a', 'b', 'a']})
    df = sparkSession.createDataFrame(dfp)
    df = df.withColumn('date', to_date('date'))
    df = df.withColumn('quantity', df['quantity'].cast('Int'))
    print "Original Data"
    df.show()

    def udf_s_e(start, excludedDiff):
        # type: (datetime.datetime, int) -> list
        dtFormatter = start
        date_list = []
        for i in range(excludedDiff - 1):
            date_list.append(dtFormatter + datetime.timedelta(days=i+1))
        return date_list

    fill_dates = udf(udf_s_e, ArrayType(DateType()))

    w = Window.orderBy("col1", "date")
    # df = df.groupBy('col1')
    tempDf = df.withColumn("diff", datediff(lead("date", 1).over(w), "date"))\
        .filter("diff > 1")
    tempDf = tempDf.withColumn("next_dates", fill_dates("date", "diff"))
    print "After mark interval date"
    tempDf.show(truncate=False)
    tempDf = tempDf\
        .withColumn("quantity", functions.lit("0"))\
        .withColumn("date", functions.explode("next_dates"))
    print "convert every list to rows"
    tempDf.show(truncate=False)

    result = df.union(tempDf.select("col1", "date", "quantity")).orderBy("date")
    print "union missing date into original data"
    result.sort(['col1', 'date']).show()
    print "Compare to original data"
    df.sort(['col1', 'date']).show()
    exit(0)

參考：https://stackoverflow.com/questions/46709285/filling-missing-dates-in-spark-dataframe-column/46712692

其它tip：

函數定義中參數前的*表示的是將調用時的多個參數放入元組中,**則表示將調用函數時的關鍵字參數放入一個字典中