[FEA] cudf.DataFrame.from_dict

[mattf]

Is your feature request related to a problem? Please describe.
rewriting code from pandas into cudf, trying to use import cudf as pd

Describe the solution you'd like
cudf.DataFrame.from_dict matching https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.from_dict.html

[rjzamora]

+1 to this feature (and to it's dask_cudf counterpart).

[quasiben]

From a call with @shwina. We can already do something dict things like:

In [4]: cudf.DataFrame({'a': cudf.Series([1,2,3]), 'b': cudf.Series([4,5,6], index=[9,10,11])})
Out[4]:
       a     b
0      1  <NA>
1      2  <NA>
2      3  <NA>
9   <NA>     4
10  <NA>     5
11  <NA>     6

Hmm, actually the DataFrame constructor is pretty robust:

In [19]: cudf.DataFrame({'c': ['a', 'b', 'd'], 'b': cudf.Series([1, 2, 3]), 'd': pd.Series([10, 11, 12])})
Out[19]:
   c  b   d
0  a  1  10
1  b  2  11
2  d  3  12

In [23]: cudf.DataFrame({'a': cudf.Series([1,2,3]), 'b': cudf.Series([4,5,6], index=[9,10,11]),'c': pd.Series(['q', 'r', 's'])})
Out[23]:
       a     b     c
0      1  <NA>     q
1      2  <NA>     r
2      3  <NA>     s
9   <NA>     4  <NA>
10  <NA>     5  <NA>
11  <NA>     6  <NA>

@mattf can you, as a test, check if the standard constructor works for your usecase ? If so, maybe this is as easy as redirecting from_dict to DataFrame(...)

[rjzamora]

@quasiben - I also suspect that cudf.from_dict can simply wrap cudf.DataFrame - Probably raising a NotImplementedError for orient="index" and orient="tight". In the long run, it may be nice to support the same orient options as pandas. However, the most important thing is probably that the API exists.

More context: The dask.dataframe.DataFrame constructor does not accept a dict-like argument, and so dd.from_dict is now the recommended way to create a Dask-DataFrame from dict-formatted data. Now that dd.from_dict is backend-dispatchable, it would be nice to be able to call into an explicit cudf/dask_cudf.DataFrame method for the "cudf" backend.

[mattf]

@quasiben the cudf.DataFrame constructor works for the specific case i have, but the most important thing is the API matches what pandas.DataFrame.from_dict provides

[GregoryKimball]

@mattf Would you please provide a specific case where pandas and cudfdiverge infrom_dict` behavior?

[rjzamora]

Would you please provide a specific case where pandas and cudf diverge infrom_dict` behavior?

I think the problem is that there is no cudf.DataFrame.from_dict method at all.

[mattf]

@GregoryKimball pandas.DataFrame.from_dict exists and cudf.DataFrame.from_dict does not