[FEA] Convert two columns into a new column as a dictionary

[MikeChenfu]

Is your feature request related to a problem? Please describe.
Hello Guys, I see dict in cudf is supported when I instantiate the df. df = cudf.DataFrame({'a': [{'a':1},{'b':1}]}). Is there a method to convert two columns into a new column as a dictionary?

Describe the solution you'd like

df = cudf.DataFrame({'a': [1, 1, 1, 2, 2], 'b': [1, 1, 2, 2, 3], 'c': [1, 2, 3, 4, 5]})
df['d'] = {'a':df.a, 'b':df.b}

[kkraus14]

@MikeChenfu we're looking into this, but just to clarify, we don't support map types yet, which would be the equivalent of dict which allow arbitrary key-value pairs. We support struct types, which requires the same keys for every row of data.

[MikeChenfu]

Thanks @kkraus14 for the information. Currently I am implementing hive named_struct using cudf. I am very interested in the struct type. Is cudf able to write it as orc file? Appreciate it if some docs or examples are provided.

[kkraus14]

Thanks @kkraus14 for the information. Currently I am implementing hive named_struct using cudf. I am very interested in the struct type. Is cudf able to write it as orc file? Appreciate it if some docs or examples are provided.

We don't supporting writing struct columns to ORC quite yet, but it's being worked on currently. Hopefully it should land in the next release or two. Parquet writing is supported though.

[MikeChenfu]

Good to know that. Parquet is also good to me.:)

[beckernick]

The core request here has been implemented in #8728 (df[["a","b"]].to_struct()) and struct writing to ORC is covered by #7830 . As a result, I'm going to close this issue to consolidate discussion in 7830.