Nice to have toMap function in DataFrame

[alshan]

This code works for me:

val mpg_df = DataFrame.readCSV("https://jetbrains.bintray.com/lets-plot/mpg.csv")

val dat = (mpg_df.names.filter {it.isNotBlank()}.map { Pair(it, mpg_df.get(it).values())}).toMap()

but would be nice to just use:

mpg_df.toMap()

[holgerbrandl]

Thanks for the sharing. You could define an extension function to support this:

fun DataFrame.asMap() = names.filter { it.isNotBlank()}.map { Pair(it, get(it).values())}.toMap()

Clearly krangl's API needs to be extended to cover more use-cases, but unless you could elaborate more on the specific usecase, I'd think that such extension does not need to be part of the core API.

[alshan]

Here is my use-case: https://github.com/JetBrains/lets-plot-kotlin/blob/master/docs/examples/jupyter-notebooks/geom_smooth.ipynb

I read CSV file to krangl DataFrame and then have to convert it to a regular Map to pass to Lets-Plot as data parameter.
Lets-Plot is not dependent on Krangl but even if it were dependent creation of such an extension does look like a leaky abstraction to me.

The most reasonable place to put the toMap extension IMO would be Krangl itself.

[alshan]

Another use-case and the pain is right here

https://jaxenter.com/kotlin-jupyter-notebook-172202.html

[holgerbrandl]

Great examples. I took the liberty to disable the isNotBlank filter. Imho either krangl should not support empty column names (similar to how dplyr does it). Or, an empty column name (just one, because names must be unique by design) is legit, and should be preserved when doing toMap. But I guess this would be a different ticket.

[alshan]

My understanding was that no-name column was the index. As map doesn't have index I've dropped it.

[holgerbrandl]

There is no equivalent to pandas indices in krangl. I personally found the index concept always confusing compared to dplyr which works great (if not better) without confronting the user with a similar index-concept.

So in krangl, we enforce unqique columns already (following dplyr's lead), but we currently support an empty string as column name. However, I can't imagine any usecase for such a feature, so I'd be more in favor of enforcing not-null not-empty column names.

[holgerbrandl]

Btw, how did you write https://jetbrains.bintray.com/lets-plot/mpg.csv ? I've noticed that the latest version of the used csv parsing library commons-csv can not handle it because of https://issues.apache.org/jira/browse/CSV-257 and I'd like to comment on it by pointing to your example file.

[alshan]

I really couldn't recall now. Please feel free to point at it. I believe we've checked the license before using it.