英文:
How to fill up missing rows in a PySpark Dataframe with a combination of existing values and an input list of possible values?
问题
我有以下的PySpark数据框架：
Product Location Date Stock
Apple L1 2023-01-01 100
Apple L2 2023-01-15 30
Banana L1 2023-01-01 10
Banana L1 2023-01-08 20
Orange L2 2023-01-01 50
接下来，我有一个包含日期列所有可能值的列表：
dates = [2023-01-01, 2023-01-08, 2023-01-15]
现在，我希望数据框架包含每个现有的Product和Location组合以及dates列表中的所有日期的行，其中每个新添加行的Stock值应该为0。
输出数据框架将如下所示：
Product Location Date Stock
Apple L1 2023-01-01 100
Apple L1 2023-01-08 0
Apple L1 2023-01-15 0
Apple L2 2023-01-15 30
Apple L2 2023-01-01 0
Apple L2 2023-01-08 0
Banana L1 2023-01-01 10
Banana L1 2023-01-08 20
Banana L1 2023-01-15 0
Orange L2 2023-01-01 50
Orange L2 2023-01-08 0
Orange L2 2023-01-15 0
英文:
I have the following PySpark Dataframe:
Product Location Date Stock
Apple L1 2023-01-01 100
Apple L2 2023-01-15 30
Banana L1 2023-01-01 10
Banana L1 2023-01-08 20
Orange L2 2023-01-01 50
Next to this I have a list with all possibilities for the Date Column:
dates = [2023-01-01, 2023-01-08, 2023-01-15]
Now I want the Dataframe to contain a row for each existing combination of Product and Location and all dates from the dates list, where the Stock value for each newly added row should be equal to 0.
The output DataFrame would look like this:
Product Location Date Stock
Apple L1 2023-01-01 100
Apple L1 2023-01-08 0
Apple L1 2023-01-15 0
Apple L2 2023-01-15 30
Apple L2 2023-01-01 0
Apple L2 2023-01-08 0
Banana L1 2023-01-01 10
Banana L1 2023-01-08 20
Banana L1 2023-01-15 0
Orange L2 2023-01-01 50
Orange L2 2023-01-08 0
Orange L2 2023-01-15 0
Notes:
- Some combinations of
ProductandLocationare not already in the list and thus should not be added (i.e.(Banana, L2)or(Orange, L1)) - The possible values in
datesare all sundays, but that does not add any extra complexity to the algorithm since thedatesvalues are given in a fixed list
I think Window functions in PySpark could be useful here, but I could not find any example in which missing rows are filled up by a given list of values.
Thanks in advance!
答案1
得分: 2
最简单的方法是创建一个包含所有可能日期行的数据框，然后与“stock”字段进行连接。
以下是一个示例：
data_sdf. \
select('product', 'location'). \
dropDuplicates(). \
withColumn('dates', func.array(*[func.lit(k) for k in dates])). \
selectExpr('product', 'location', 'explode(dates) as date'). \
join(data_sdf, ['product', 'location', 'date'], 'left'). \
fillna(0, subset=['stock']). \
orderBy('product', 'location', 'date'). \
show()
+-------+--------+----------+-----+
|product|location| date|stock|
+-------+--------+----------+-----+
| Apple | L1|2023-01-01| 100|
| Apple | L1|2023-01-08| 0|
| Apple | L1|2023-01-15| 0|
| Apple | L2|2023-01-01| 0|
| Apple | L2|2023-01-08| 0|
| Apple | L2|2023-01-15| 30|
| Banana| L1|2023-01-01| 10|
| Banana| L1|2023-01-08| 20|
| Banana| L1|2023-01-15| 0|
| Orange| L2|2023-01-01| 50|
| Orange| L2|2023-01-08| 0|
| Orange| L2|2023-01-15| 0|
+-------+--------+----------+-----+
<details>
<summary>英文:</summary>
the easiest way to achieve it would be to create a dataframe with all possible date rows and then `join` the `stock` field.
here's an example
```py
data_sdf. \
select('product', 'location'). \
dropDuplicates(). \
withColumn('dates', func.array(*[func.lit(k) for k in dates])). \
selectExpr('product', 'location', 'explode(dates) as date'). \
join(data_sdf, ['product', 'location', 'date'], 'left'). \
fillna(0, subset=['stock']). \
orderBy('product', 'location', 'date'). \
show()
# +-------+--------+----------+-----+
# |product|location| date|stock|
# +-------+--------+----------+-----+
# | Apple | L1|2023-01-01| 100|
# | Apple | L1|2023-01-08| 0|
# | Apple | L1|2023-01-15| 0|
# | Apple | L2|2023-01-01| 0|
# | Apple | L2|2023-01-08| 0|
# | Apple | L2|2023-01-15| 30|
# | Banana| L1|2023-01-01| 10|
# | Banana| L1|2023-01-08| 20|
# | Banana| L1|2023-01-15| 0|
# | Orange| L2|2023-01-01| 50|
# | Orange| L2|2023-01-08| 0|
# | Orange| L2|2023-01-15| 0|
# +-------+--------+----------+-----+
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库，让每个人都能够通过互相帮助和分享经验来进步。
评论