如何在 Databricks 中的事实表的数据透视查询 sql 中用零替换空值

Question

我看到了很多关于如何在有列被查询的情况下执行此操作的解决方案，包括以下内容...

how to Replace null with zero in pivot SQL query

Oracle 11g SQL - Replacing NULLS with zero where query has PIVOT

Replacing null values in dynamic pivot sql query

等等，等等，等等，

但是，当您为条件的存在创建事实表时，如何替换数据透视查询中的空值。

例如，在 Databricks 中： 如何替换以下内容的空值

设置

drop table if exists patient_dx;

create table patient_dx (patient_id string, dx string);

insert into patient_dx values
  ('Bob', 'cough'),
  ('Donna', 'cough'),
  ('Jerry', 'cough'),
  ('Bob', 'feaver'),
  ('Donna', 'head ache')
;

查询：

select * from (
  select
    patient_id,
    dx,
    cast (1 as int) cnt
  from
    patient_dx
)
pivot (
  max(cnt)
  for dx in ('cough','feaver','head ache')
)
;

结果

我尝试了以下几种排列方式：

cast(0 + cast(coalesce(sum(coalesce(cnt,0)),0) as int) as int) as cnt

无济于事

Answer 1

您必须使用 coalesce 或 NOT NULL 来替换选择查询中的空值。

如果有帮助，请检查以下内容：

---

试试这个：

spark.sql("""
select
 patient_id,
 CASE 
 when cough is NOT NULL THEN cough
 else 0
 END as cough,
 CASE 
 when feaver is NOT NULL THEN feaver
 else 0
 END as feaver,
 CASE 
 when `head ache` is NOT NULL THEN `head ache`
 else 0
 END as `head ache`
 from ( 
select * from patient
)
PIVOT(
  Count(dx)
  for dx in ('cough','feaver','head ache')
)
;
""").show()

输出将是：

patient_id	cough	feaver	head ache
Donna	1	0	1
Jerry	1	0	0
Bob	1	1	0

如果你希望它是动态的

dist=spark.sql("select collect_set(dx) from patient;").toPandas()
val=spark.sql("""
select
 patient_id,
 coalesce(cough,0) as `cough`,
 coalesce(feaver,0) as `feaver`,
 coalesce(`head ache`,0) as `head ache`
 from ( 
select * from patient
)
PIVOT(
  Count(dx)
  for dx in """
+
str(tuple(map(tuple, *dist.values))[0])
+
"""
)
;
""")