问题
Let's say I have the following table:
+--------------------+--------------------+------+------------+--------------------+
| host| path|status|content_size| time|
+--------------------+--------------------+------+------------+--------------------+
|js002.cc.utsunomi...|/shuttle/resource...| 404| 0|1995-08-01 00:07:...|
| tia1.eskimo.com |/pub/winvn/releas...| 404| 0|1995-08-01 00:28:...|
|grimnet23.idirect...|/www/software/win...| 404| 0|1995-08-01 00:50:...|
|miriworld.its.uni...|/history/history.htm| 404| 0|1995-08-01 01:04:...|
| ras38.srv.net |/elv/DELTA/uncons...| 404| 0|1995-08-01 01:05:...|
| cs1-06.leh.ptd.net | | 404| 0|1995-08-01 01:17:...|
|dialip-24.athenet...|/history/apollo/a...| 404| 0|1995-08-01 01:33:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:35:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:36:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:36:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:36:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:36:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:36:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:36:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:37:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:37:...|
| h96-158.ccnet.com |/history/apollo/a...| 404| 0|1995-08-01 01:37:...|
|hsccs_gatorbox07....|/pub/winvn/releas...| 404| 0|1995-08-01 01:44:...|
|www-b2.proxy.aol....|/pub/winvn/readme...| 404| 0|1995-08-01 01:48:...|
|www-b2.proxy.aol....|/pub/winvn/releas...| 404| 0|1995-08-01 01:48:...|
+--------------------+--------------------+------+------------+--------------------+
How I would filter this table to have only distinct paths in PySpark? But the table should contains all columns.
回答1:
If you want to save rows where all values in specific column are distinct, you have to call dropDuplicates method on DataFrame.
Like this in my example:
dataFrame = ...
dataFrame.dropDuplicates(['path'])
where path is column name
回答2:
As for tuning which records are kept and discarded, if you can work your conditions into a Window expression, you can use something like this. This is in scala (more or less) but I imagine you can do it in PySpark, too.
val window = Window.parititionBy('columns,'to,'make,'unique).orderBy('conditionToPutRowToKeepFirst)
dataframe.withColumn("row_number",row_number().over(window)).where('row_number===1).drop('row_number)
来源：https://stackoverflow.com/questions/39287729/filter-rows-by-distinct-values-in-one-column-in-pyspark