[FEA] Standardize applymap support with pandas to enable Dask applymap

[beckernick]

Today, we support the applymap interface on Series but not DataFrames. Pandas supports applymap on DataFrames but not Series. In pandas, the interface provides applies a scalar function/UDF to every element in the dataframe (elementwise UDF).

The reason for our Series.applymap implementation may be that, until recently, we did not support an elementwise apply interface directly. Now that we do, it's possible Series.applymap is redundant, as both interfaces explicitly provide users access to elementwise UDFs run via our udf pipeline. (Please feel free to correct me if I'm off base here).

For compatibility with Dask, we should explore aligning our interfaces with pandas. Right now, it's not possible to use applymap with Dask-cuDF, as the Dask.Series object does not have our Series interface and we don't implement the DataFrame interface.

We might consider:

• Deprecating (and eventually removing) Series.applymap
• Implementing DataFrame.applymap with our existing udf pipeline machinery (independently and sequentially processing each column)
• Adding unit tests for dask_cudf Dataframes
• Update the Guide to UDFs as appropriate

import dask_cudf
import cudf
import pandas as pd
import dask.dataframe as dd

df = pd.DataFrame({"a":[0,1,2,3,4]})
gdf = cudf.from_pandas(df)

def func(x):
    return x + 10

# Dask CPU
ddf = dd.from_pandas(df, 2)
ddf.applymap(func).compute() # succeeds
# ddf.a.applymap(func).compute() # doesn't exist

# Dask GPU
ddf = dask_cudf.from_cudf(gdf, 2)
# ddf.applymap(func).compute() # fails
# ddf.a.applymap(func).compute() # fails

[brandon-b-miller]

I can look into this, @beckernick .

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[github-actions]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[brandon-b-miller]

This is in progress, applymap is deprecated and we're moving to supercede it with apply. In addition we now have DataFrame.applymap instead that truly mirrors pandas. This will be closed when Series.applymap is removed.

[brandon-b-miller]

with #11031 merged I think this is done :)