在 PySpark 中查找通过 window.partitionBy 从中提取最小值的行值

Question

我有一个像这样的PySpark数据帧：

+--------+-------------+--------------+-----------------------+
|material|purchase_date|mkt_prc_usd_lb|min_mkt_prc_over_1month|
+--------+-------------+--------------+-----------------------+
|  Copper|   2019-01-09|        2.6945|                 2.6838|
|  Copper|   2019-01-23|        2.6838|                 2.6838|
|    Zinc|   2019-01-23|        1.1829|                 1.1829|
|    Zinc|   2019-06-26|        1.1918|                 1.1918|
|Aluminum|   2019-01-02|        0.8363|                 0.8342|
|Aluminum|   2019-01-09|        0.8342|                 0.8342|
|Aluminum|   2019-01-23|        0.8555|                 0.8342|
|Aluminum|   2019-04-03|        0.8461|                 0.8461|
+--------+-------------+--------------+-----------------------+

最后一列"min_mkt_prc_over_1month"计算为材料在一个月内的最小"mkt_prc_usd_lb"(第 3 列(，即(-15 天，至 +15 天(在材料上，purchase_date窗口

：代码为：

w2 = (Window()
.partitionBy("material")
.orderBy(col("purchase_date").cast("timestamp").cast("long"))
.rangeBetween(-days(15), days(15)))

现在，我想看看当金额是/将是最小金额时，"purchase_date"是什么？

预期输出：(从前两行开始(

+--------+-------------+--------------+-----------------------+------------------+
|material|purchase_date|mkt_prc_usd_lb|min_mkt_prc_over_1month|date_of_min_price |
+--------+-------------+--------------+-----------------------+------------------+
|  Copper|   2019-01-09|        2.6945|                 2.6838|        2019-01-23|
|  Copper|   2019-01-23|        2.6838|                 2.6838|        2019-01-23|
+--------+-------------+--------------+-----------------------+------------------+

Answer 1

试试这个。我们可以在two prc are the same to populate it with purchase dateotherwise to put Null的地方创建一个列，然后我们可以在newly created column using our window w2.上使用First with ignoreNulls=True。

from pyspark.sql.functions import *
from pyspark.sql.window import Window
days= lambda i: i * 86400
w2 = (Window()
.partitionBy("material")
.orderBy(col("purchase_date").cast("timestamp").cast("long"))
.rangeBetween(-days(15), days(15)))

df.withColumn("first",
expr("""IF(mkt_prc_usd_lb=min_mkt_prc_over_1month,purchase_date,null)"""))
.withColumn("date_of_min_price", first("first", True).over(w2)).drop("first")
.show()
#+--------+-------------+--------------+-----------------------+-----------------+
#|material|purchase_date|mkt_prc_usd_lb|min_mkt_prc_over_1month|date_of_min_price|
#+--------+-------------+--------------+-----------------------+-----------------+
#|  Copper|   2019-01-09|        2.6945|                 2.6838|       2019-01-23|
#|  Copper|   2019-01-23|        2.6838|                 2.6838|       2019-01-23|
#|    Zinc|   2019-01-23|        1.1829|                 1.1829|       2019-01-23|
#|    Zinc|   2019-06-26|        1.1918|                 1.1918|       2019-06-26|
#|Aluminum|   2019-01-02|        0.8363|                 0.8342|       2019-01-09|
#|Aluminum|   2019-01-09|        0.8342|                 0.8342|       2019-01-09|
#|Aluminum|   2019-01-23|        0.8555|                 0.8342|       2019-01-09|
#|Aluminum|   2019-04-03|        0.8461|                 0.8461|       2019-04-03|
#+--------+-------------+--------------+-----------------------+-----------------+