[FEA] Handle setting numpy array to GDF

[esnvidia]

Is your feature request related to a problem? Please describe.
Error message either should be clearer or handle setting values from numpy array to GDF

Describe the solution you'd like
sk_std_scaler is scikit-learn standard scaler.
sk_gdf.loc[:, numer_cols] = sk_std_scaler.fit_transform(sk_gdf[numer_cols].to_pandas())

or a more helpful error message:
Cannot assign numpy array to Device DataFrame. Please convert to cupy.array object first with cupy.array(numpy_array)

Describe alternatives you've considered
Convert to cupy array: sk_gdf.loc[:, numer_cols] = cp.array(sk_std_scaler.fit_transform(sk_gdf[numer_cols].to_pandas()))

Additional context
Untitled.zip
Also see issues here for additional context:
#8672
rapidsai/cuml#4034

[beckernick]

@esnvidia would you be able to provide a minimal reproducer in Python code (outside of a Jupyter notebook)? Thanks!

[esnvidia]

example.zip
@beckernick Cleaned up code to make a minimal py file that I ran with a conda install of rapids 21.06

[beckernick]

@esnvidia is this the same issue as #8672 ?

[esnvidia]

@beckernick Related but not the same. This one is related to setting a numpy matrix to a set of indices/columns, which is why this is a Feature Request. In #8672 the output is a cupy array IIRC. In that issue, the setting of the cupy array to the DataFrame was wrong.

[beckernick]

It looks like the Scaler issue was fixed in cuML based on rapidsai/cuml#4034 , and the setitem issue is covered by #8672

As a result, I'm going to close this issue. Please feel to re-open if I have misunderstood what the request is here.