[
https://issues.apache.org/jira/browse/SPARK-23011?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Li Jin updated SPARK-23011:
---------------------------
Description:
The current semantics of groupby apply is that the output schema of groupby
apply is the same as the output schema of the UDF. Because grouping column is
usually useful to users, users often need to output grouping columns in the
UDF. To further explain, consider the following example:
{code:java}
import statsmodels.api as sm
# df has four columns: id, y, x1, x2
group_column = 'id'
y_column = 'y'
x_columns = ['x1', 'x2']
schema = df.select(group_column, *x_columns).schema
@pandas_udf(schema, PandasUDFType.GROUP_MAP)
# Input/output are both a pandas.DataFrame
def ols(pdf):
group_key = pdf[group_column].iloc[0]
y = pdf[y_column]
X = pdf[x_columns]
X = sm.add_constant(X)
model = sm.OLS(y, X).fit()
return pd.DataFrame([[group_key] + [model.params[i] for i in x_columns]],
columns=[group_column] + x_columns)
beta = df.groupby(group_column).apply(ols)
{code}
Although the UDF (linear regression) has nothing to do with the grouping
column, the user needs to deal with grouping column in the UDF. In other words,
the UDF is tightly coupled with the grouping column.
After discussion in
was:
The current semantics of groupby apply is that the output schema of groupby
apply is the same as the output schema of the UDF. Because grouping column is
usually useful to users, users often need to output grouping columns in the
UDF. To further explain, consider the following example:
{code:java}
import statsmodels.api as sm
# df has four columns: id, y, x1, x2
group_column = 'id'
y_column = 'y'
x_columns = ['x1', 'x2']
schema = df.select(group_column, *x_columns).schema
@pandas_udf(schema, PandasUDFType.GROUP_MAP)
# Input/output are both a pandas.DataFrame
def ols(pdf):
group_key = pdf[group_column].iloc[0]
y = pdf[y_column]
X = pdf[x_columns]
X = sm.add_constant(X)
model = sm.OLS(y, X).fit()
return pd.DataFrame([[group_key] + [model.params[i] for i in x_columns]],
columns=[group_column] + x_columns)
beta = df.groupby(group_column).apply(ols)
{code}
Although the UDF (linear regression) has nothing to do with the grouping
column, the user needs to deal with grouping column in the UDF. In other words,
the UDF is tightly coupled with the grouping column.
Here I propose we prepend grouping columns that are not returned by the UDF to
the result of groupby apply. With this change, users can write UDFs that are
decoupled from the grouping column:
{code:java}
import statsmodels.api as sm
# df has four columns: id, y, x1, x2
group_column = 'id'
y_column = 'y'
x_columns = ['x1', 'x2']
schema = df.select(*x_columns).schema
@pandas_udf(schema, PandasUDFType.GROUP_MAP)
# Input/output are both a pandas.DataFrame
def ols(pdf):
y = pdf[y_column]
X = pdf[x_columns]
X = sm.add_constant(X)
model = sm.OLS(y, X).fit()
return pd.DataFrame([[model.params[i] for i in x_columns]],
columns=x_columns)
beta = df.groupby(group_column).apply(ols)
{code}
> Support alternative function form with group aggregate pandas UDF
> -----------------------------------------------------------------
>
> Key: SPARK-23011
> URL: https://issues.apache.org/jira/browse/SPARK-23011
> Project: Spark
> Issue Type: Sub-task
> Components: PySpark
> Affects Versions: 2.3.0
> Reporter: Li Jin
> Priority: Major
>
> The current semantics of groupby apply is that the output schema of groupby
> apply is the same as the output schema of the UDF. Because grouping column is
> usually useful to users, users often need to output grouping columns in the
> UDF. To further explain, consider the following example:
> {code:java}
> import statsmodels.api as sm
> # df has four columns: id, y, x1, x2
> group_column = 'id'
> y_column = 'y'
> x_columns = ['x1', 'x2']
> schema = df.select(group_column, *x_columns).schema
> @pandas_udf(schema, PandasUDFType.GROUP_MAP)
> # Input/output are both a pandas.DataFrame
> def ols(pdf):
> group_key = pdf[group_column].iloc[0]
> y = pdf[y_column]
> X = pdf[x_columns]
> X = sm.add_constant(X)
> model = sm.OLS(y, X).fit()
> return pd.DataFrame([[group_key] + [model.params[i] for i in
> x_columns]], columns=[group_column] + x_columns)
> beta = df.groupby(group_column).apply(ols)
> {code}
> Although the UDF (linear regression) has nothing to do with the grouping
> column, the user needs to deal with grouping column in the UDF. In other
> words, the UDF is tightly coupled with the grouping column.
>
> After discussion in
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
