Manually converting LazyGroupBy to LazyFrame breaks printing

[etiennebacher]

I tried to manually convert a LazyGroupBy object to a LazyFrame using class() but it generates an error when I want to print the output. Is this expected?

library(polars)
#> Warning: le package 'polars' a été compilé avec la version R 4.3.1

test1 <- pl$LazyFrame(mtcars)
test1
#> [1] "polars LazyFrame naive plan: (run ldf$describe_optimized_plan() to see the optimized plan)"
#> DF ["mpg", "cyl", "disp", "hp"]; PROJECT */11 COLUMNS; SELECTION: "None"

test2 <- test1$groupby("am")
test2
#> polars LazyGroupBy: 
#> LazyGroupBy (internals are opaque)

class(test2) <- "LazyFrame"
test2
#> [1] "polars LazyFrame naive plan: (run ldf$describe_optimized_plan() to see the optimized plan)"
#> Error in .pr$LazyFrame$print(x): expected LazyFrame

---

To give some context, in tidypolars if the user has a group_by() then the data is either GroupBy or LazyGroupBy. This is fine to use summarize() (which uses $agg() under the hood) but to be able to use mutate() on grouped data I need to convert GroupBy to DataFrame and LazyGroupBy to LazyFrame. This is where I got the error. If you have a better way to ungroup data, I'm all for it

[etiennebacher]

Note however that there's no problem with DataFrame:

library(polars)

test1 <- pl$DataFrame(mtcars)
test1
#> shape: (32, 11)
#> ┌──────┬─────┬───────┬───────┬───┬─────┬─────┬──────┬──────┐
#> │ mpg  ┆ cyl ┆ disp  ┆ hp    ┆ … ┆ vs  ┆ am  ┆ gear ┆ carb │
#> │ ---  ┆ --- ┆ ---   ┆ ---   ┆   ┆ --- ┆ --- ┆ ---  ┆ ---  │
#> │ f64  ┆ f64 ┆ f64   ┆ f64   ┆   ┆ f64 ┆ f64 ┆ f64  ┆ f64  │
#> ╞══════╪═════╪═══════╪═══════╪═══╪═════╪═════╪══════╪══════╡
#> │ 21.0 ┆ 6.0 ┆ 160.0 ┆ 110.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 4.0  ┆ 4.0  │
#> │ 21.0 ┆ 6.0 ┆ 160.0 ┆ 110.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 4.0  ┆ 4.0  │
#> │ 22.8 ┆ 4.0 ┆ 108.0 ┆ 93.0  ┆ … ┆ 1.0 ┆ 1.0 ┆ 4.0  ┆ 1.0  │
#> │ 21.4 ┆ 6.0 ┆ 258.0 ┆ 110.0 ┆ … ┆ 1.0 ┆ 0.0 ┆ 3.0  ┆ 1.0  │
#> │ …    ┆ …   ┆ …     ┆ …     ┆ … ┆ …   ┆ …   ┆ …    ┆ …    │
#> │ 15.8 ┆ 8.0 ┆ 351.0 ┆ 264.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 4.0  │
#> │ 19.7 ┆ 6.0 ┆ 145.0 ┆ 175.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 6.0  │
#> │ 15.0 ┆ 8.0 ┆ 301.0 ┆ 335.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 8.0  │
#> │ 21.4 ┆ 4.0 ┆ 121.0 ┆ 109.0 ┆ … ┆ 1.0 ┆ 1.0 ┆ 4.0  ┆ 2.0  │
#> └──────┴─────┴───────┴───────┴───┴─────┴─────┴──────┴──────┘

test2 <- test1$groupby("am")
test2
#> shape: (32, 11)
#> ┌──────┬─────┬───────┬───────┬───┬─────┬─────┬──────┬──────┐
#> │ mpg  ┆ cyl ┆ disp  ┆ hp    ┆ … ┆ vs  ┆ am  ┆ gear ┆ carb │
#> │ ---  ┆ --- ┆ ---   ┆ ---   ┆   ┆ --- ┆ --- ┆ ---  ┆ ---  │
#> │ f64  ┆ f64 ┆ f64   ┆ f64   ┆   ┆ f64 ┆ f64 ┆ f64  ┆ f64  │
#> ╞══════╪═════╪═══════╪═══════╪═══╪═════╪═════╪══════╪══════╡
#> │ 21.0 ┆ 6.0 ┆ 160.0 ┆ 110.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 4.0  ┆ 4.0  │
#> │ 21.0 ┆ 6.0 ┆ 160.0 ┆ 110.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 4.0  ┆ 4.0  │
#> │ 22.8 ┆ 4.0 ┆ 108.0 ┆ 93.0  ┆ … ┆ 1.0 ┆ 1.0 ┆ 4.0  ┆ 1.0  │
#> │ 21.4 ┆ 6.0 ┆ 258.0 ┆ 110.0 ┆ … ┆ 1.0 ┆ 0.0 ┆ 3.0  ┆ 1.0  │
#> │ …    ┆ …   ┆ …     ┆ …     ┆ … ┆ …   ┆ …   ┆ …    ┆ …    │
#> │ 15.8 ┆ 8.0 ┆ 351.0 ┆ 264.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 4.0  │
#> │ 19.7 ┆ 6.0 ┆ 145.0 ┆ 175.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 6.0  │
#> │ 15.0 ┆ 8.0 ┆ 301.0 ┆ 335.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 8.0  │
#> │ 21.4 ┆ 4.0 ┆ 121.0 ┆ 109.0 ┆ … ┆ 1.0 ┆ 1.0 ┆ 4.0  ┆ 2.0  │
#> └──────┴─────┴───────┴───────┴───┴─────┴─────┴──────┴──────┘
#> groups: [[1]]
#> [1] "am"
#> 
#> maintain order:  FALSE

class(test2) <- "DataFrame"
test2
#> shape: (32, 11)
#> ┌──────┬─────┬───────┬───────┬───┬─────┬─────┬──────┬──────┐
#> │ mpg  ┆ cyl ┆ disp  ┆ hp    ┆ … ┆ vs  ┆ am  ┆ gear ┆ carb │
#> │ ---  ┆ --- ┆ ---   ┆ ---   ┆   ┆ --- ┆ --- ┆ ---  ┆ ---  │
#> │ f64  ┆ f64 ┆ f64   ┆ f64   ┆   ┆ f64 ┆ f64 ┆ f64  ┆ f64  │
#> ╞══════╪═════╪═══════╪═══════╪═══╪═════╪═════╪══════╪══════╡
#> │ 21.0 ┆ 6.0 ┆ 160.0 ┆ 110.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 4.0  ┆ 4.0  │
#> │ 21.0 ┆ 6.0 ┆ 160.0 ┆ 110.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 4.0  ┆ 4.0  │
#> │ 22.8 ┆ 4.0 ┆ 108.0 ┆ 93.0  ┆ … ┆ 1.0 ┆ 1.0 ┆ 4.0  ┆ 1.0  │
#> │ 21.4 ┆ 6.0 ┆ 258.0 ┆ 110.0 ┆ … ┆ 1.0 ┆ 0.0 ┆ 3.0  ┆ 1.0  │
#> │ …    ┆ …   ┆ …     ┆ …     ┆ … ┆ …   ┆ …   ┆ …    ┆ …    │
#> │ 15.8 ┆ 8.0 ┆ 351.0 ┆ 264.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 4.0  │
#> │ 19.7 ┆ 6.0 ┆ 145.0 ┆ 175.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 6.0  │
#> │ 15.0 ┆ 8.0 ┆ 301.0 ┆ 335.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 8.0  │
#> │ 21.4 ┆ 4.0 ┆ 121.0 ┆ 109.0 ┆ … ┆ 1.0 ┆ 1.0 ┆ 4.0  ┆ 2.0  │
#> └──────┴─────┴───────┴───────┴───┴─────┴─────┴──────┴──────┘

[vincentarelbundock]

Don't have any insight into this, but I wonder if there shouldn't be a better interface to this. Changing the class feels hackish. Perhaps a simple approach would be for pl$DataFrame() to convert the grouped object back to a standard polars data frame?

[etiennebacher]

The thing is there's no function that can be applied on a GroupBy object to ungroup it: https://pola-rs.github.io/polars/py-polars/html/reference/dataframe/groupby.html

So even a mechanism like this is implemented in r-polars it will still be a hack and not something Polars support. I also didn't find any issues mentioning "ungroup" on the rust-polars repo.

[etiennebacher]

FYI this is why Python's tidypolars doesn't have group_by() but uses a .by argument instead: markfairbanks/tidypolars#201

[eitsupi]

FYI this is why Python's tidypolars doesn't have group_by() but uses a .by argument instead: markfairbanks/tidypolars#201

Perhaps it would be best to give up reproducing group_by and ungroup in R tidypolars as well?

I believe the arrow package had numerous bugs due to reproducing these. (For example apache/arrow#14175)

[etiennebacher]

FYI I think I found a solution for tidypolars: etiennebacher/tidypolars#22
Just waiting for a PR in r-polars

[sorhawell]

@etiennebacher I think what you are trying will work.

For reference the rust-polars GroupBy is implemented with a lifetime chained to the DataFrame. However, extendr does not support lifetime annotated structs. A rust-polars GroupBy is just a reference to a DataFrame + some few options such maintain_order. In r-polars GroupBy is a "virtual R-side class abstraction" on top of DataFrame + some attributes on R side to remember options.

The r-polars LazyFrameGroupBy in rust-polars has no lifetime annotation and is just wrapped with extendr as is in r-polars .

I think you could get far by not using LazyFrameGroupBy and instead working with attributes as r-polars GroupBy.

I have approved your <LazyFrame>.clone()PR #347 which is needed as it is the only known way to have immutable attributes for the ExternalPtr.

[etiennebacher]

Thanks for approving the LazyFrame$clone() PR.

In r-polars GroupBy is a "virtual R-side class abstraction" on top of DataFrame + some attributes on R side to remember options.

I didn't notice that before, but using attributes only of the input Data/LazyFrame will make handling groups in tidypolars much much simpler and probably more robust.

It's probably not a good practice to manually change from GroupBy to DataFrame, as mentioned above, and I have another way to make this work, so I close this