add DataFrame.transpose/2

[lkarthee]

df = Explorer.DataFrame.new(a: [32.0, 33.0], b: [1, 2], c: ["a", "b"])
DF.print(df)
dft = DF.transpose(df, include_header: true, header_name: "names", column_names: ["x", "y"]) #, 

+--------------------------------------------+
| Explorer DataFrame: [rows: 3, columns: 3]  |
+--------------+--------------+--------------+
|    names     |      x       |      y       |
|   <string>   |   <string>   |   <string>   |
+==============+==============+==============+
| a            | 32.0         | 33.0         |
+--------------+--------------+--------------+
| b            | 1            | 2            |
+--------------+--------------+--------------+
| c            | a            | b            |
+--------------+--------------+--------------+

[josevalim]

We don't need both options. We can have a :headeroption. It may be false (default), true (default to column) or a string with the header name.

[lkarthee]

should :column_names be columns ? I see it as a common parameter for many DF functions ?

[lkarthee]

i have changed :column_names to column. I can revert it..

[josevalim]

Suggested change

nil -> nil

[lkarthee]

To avoid to_column_name converting nil to string, I added that clause -> nil -> nil ?

[josevalim]

When it would ever be nil? The accepted values should be Boolean or string, no?

[lkarthee]

Atoms are also accepted- they are converted to strings by to_column_name(). So when user passes nil, it will be converted to "nil". Just to guard it I put that condition. Is nil to "nil" conversion is acceptable?

[billylanchantin]

Handling atoms makes sense. Would an ArgumentError be more appropriate for header: nil?

Also I think guards on the to_column_name clause might be helpful.

[lkarthee]

If we disallow nil, user may pass "nil". df[nil] is valid, even though it looks funny.

iex> dft = DF.transpose(df, header: "nil", columns: ["x", "y"])
iex> dft[nil]
#Explorer.Series<
  Polars[3]
  string ["a", "b", "c"]
>

Not much we can do here. I have accepted Jose code suggestion and removed the statement.

[josevalim]

Ideally we want to validate what we can on Explorer side. We only let Polars raise when the error requires traversing the series itself. Can you validate this one in Elixir too? Thank you!

[lkarthee]

addressed in #823

[josevalim]

Looks great to me! @cigrainger should decide on the API (name and options). :)

[cigrainger]

Yeah this can be tough to do with pivot_wider. It's a bit niche, but all good. I'm cool with transpose, and I think you guys got to the ideal set of opts w/ header and columns. I'm happy with this!

[josevalim]

💚 💙 💜 💛 ❤️