Quick dataframe question, is it possible in a notebook to write a dataframe extension method after t...

# datascience

[Peter]

Quick dataframe question, is it possible in a notebook to write a dataframe extension method after the dataframe instance has been created based on a parsed CSV file

fun DataFrame<???>.myMethod() = filter { some_column == "Yes" }

So the question is how to get the required type since the dataframe schema is automatically generated based on the data found in the CSV file.

[Ilya Muradyan]

Hi, you can find it out from the generated code, see screenshots

[Ilya Muradyan]

Another way to find it out is completion

[Ilya Muradyan]

And.. One more way. Define the following function in you notebook:

@OptIn(kotlin.ExperimentalStdlibApi::class)
inline fun <reified T> DataFrame<T>.dtype(): kotlin.reflect.KType = kotlin.reflect.typeOf<T>()

And then just call it this way:

[Peter]

Thanks, works like a charm!!!

P.S What really would be great, if in the (far) future this could grow into something like "type providers", as exists in F# language

[Ilya Muradyan]

@Peter DataFrame has one more thing to suggest. You can define an interface, annotate it with @DataSchema, and all "compatible" dataframes will implement it. See example. Maybe it's even better than use implicitly generated types

[holgerbrandl]

In which repo is this generic magic implemented? I'd love to backport this to krangl if possible. 🙂

[Ilya Muradyan]

Code generation is in dataframe repository, but it extensively uses Jupyter kernel API. You can use it too like described here. Simple example is here.

[Nikita Klimenko [JB]]

@Peter Hi! We have a gradle plugin that does schema inference from data samples https://kotlin.github.io/dataframe/gradle.html#schema-inference. I think it kind of similar to FSharp.Data type provider in terms of functionality. I would love to hear any feedback so we can polish plugin UX 🙂

[holgerbrandl]

Thanks for the pointers!