Polars literal DuplicateError using when/then with selectors

DuplicateError

The issue is that the output name of when/then comes from the first .then() branch.

In this case 1 is parsed as pl.lit(1) which has the default name of literal.

pl.when(cs.numeric() > 50).then(1)

You can think of it as there being an implicit alias() call with the name from the then branch.

pl.when(cs.numeric() > 50).then(pl.lit(1)).alias("literal")

Expression expansion then turns this into 3 separate calls, so you get a DuplicateError.

pl.when(pl.col.j > 50).then(pl.lit(1)).alias("literal"),
pl.when(pl.col.k > 50).then(pl.lit(1)).alias("literal"),
pl.when(pl.col.l > 50).then(pl.lit(1)).alias("literal")

Keep name with a 'literal' then

.name.keep() can be added to use the column name as the output name instead.

df.select(
   pl.when(cs.numeric() < 50)
     .then(1)
     .otherwise(2)
     .name.keep()
)

shape: (10, 3)
┌─────┬─────┬─────┐
│ j   ┆ k   ┆ l   │
│ --- ┆ --- ┆ --- │
│ i32 ┆ i32 ┆ i32 │
╞═════╪═════╪═════╡
│ 2   ┆ 2   ┆ 2   │
│ 1   ┆ 1   ┆ 2   │
│ 1   ┆ 1   ┆ 2   │
│ 2   ┆ 2   ┆ 1   │
│ 2   ┆ 1   ┆ 2   │
│ 2   ┆ 1   ┆ 2   │
│ 2   ┆ 1   ┆ 2   │
│ 2   ┆ 2   ┆ 2   │
│ 2   ┆ 1   ┆ 1   │
│ 2   ┆ 2   ┆ 1   │
└─────┴─────┴─────┘

Keep name with a 'column' then

As per the comments, if you want to use a col() inside then() you will get an error.

df.select(
    pl.when(cs.numeric() < 50).then(pl.col.j).otherwise(2).name.keep()
)
# DuplicateError: projections contained duplicate output name 'j'.

What you can do nest another when/then inside your then() branch.

df.select(
    pl.when(cs.numeric() < 50)
      .then(pl.when(False).then(cs.numeric() < 50).otherwise(pl.col.j))
      .otherwise(2)
)

shape: (10, 3)
┌─────┬─────┬─────┐
│ j   ┆ k   ┆ l   │
│ --- ┆ --- ┆ --- │
│ i64 ┆ i64 ┆ i64 │
╞═════╪═════╪═════╡
│ 2   ┆ 2   ┆ 2   │
│ 26  ┆ 26  ┆ 2   │
│ 12  ┆ 12  ┆ 2   │
│ 2   ┆ 2   ┆ 92  │
│ 2   ┆ 95  ┆ 2   │
│ 2   ┆ 75  ┆ 2   │
│ 2   ┆ 61  ┆ 2   │
│ 2   ┆ 2   ┆ 2   │
│ 2   ┆ 73  ┆ 73  │
│ 2   ┆ 2   ┆ 66  │
└─────┴─────┴─────┘

You need the "column selector" inside the then() branch in order to retain the name.

pl.when(False) can be used to "broadcast" column j values into the other columns while keeping their name.

df.select(pl.when(False).then(cs.numeric() < 50).otherwise(pl.col.j))

shape: (10, 3)
┌─────┬─────┬─────┐
│ j   ┆ k   ┆ l   │
│ --- ┆ --- ┆ --- │
│ i64 ┆ i64 ┆ i64 │
╞═════╪═════╪═════╡
│ 71  ┆ 71  ┆ 71  │
│ 26  ┆ 26  ┆ 26  │
│ 12  ┆ 12  ┆ 12  │
│ 92  ┆ 92  ┆ 92  │
│ 95  ┆ 95  ┆ 95  │
│ 75  ┆ 75  ┆ 75  │
│ 61  ┆ 61  ┆ 61  │
│ 74  ┆ 74  ┆ 74  │
│ 73  ┆ 73  ┆ 73  │
│ 66  ┆ 66  ┆ 66  │
└─────┴─────┴─────┘

Technically you only .then(cs.numeric()) for the "inner" when, but I've just repeated the predicate.