DataFrames / LazyFrames have a very convenient method with_columns.
It would be nice for Structs to have a similar method, e.g. with_fields.
Motivation
When working with top-level columns, we can use with_columns to transform just one column, leaving everything else around intact.
However, when working with nested structures (structs inside lists, or lists inside structs), we have to manually keep the whole structure from falling apart.
from polars import DataFrame, col, struct, element
nested_df = DataFrame(
{
"coords": [
[{"x": 1, "y": 4}, {"x": 4, "y": 9}, {"x": 9, "y": 16}],
[{"x": 4, "y": 25}, {"x": 25, "y": 36}, {"x": 36, "y": 49}],
]
}
)
nested_df.with_columns(
col("coords").list.eval(
struct(
element().struct.field("x").sqrt(), # Here is the transformation I want.
element().struct.field("y"), # Here is the context I have to manually restore.
)
)
)
ritchie46
commented
1 month ago
Description
TL;DR
DataFrames / LazyFrames have a very convenient method
with_columns. It would be nice for Structs to have a similar method, e.g.with_fields.Motivation
When working with top-level columns, we can use
with_columnsto transform just one column, leaving everything else around intact.However, when working with nested structures (structs inside lists, or lists inside structs), we have to manually keep the whole structure from falling apart.