I would like to know if it is possible to have the number of unique items from a given column after a groupBy aggregation with Dask. I don't see anything like this in the documentation. It is available on pandas dataframe and really useful. I've seen some issue related to this, but I am not sure it is implemented.
Can someone give me some hints about this?
To expand on this comment you can use nunique on a SeriesGroupBy directly:
import pandas as pd
import dask.dataframe as dd
d = {'col1': [1, 2, 3, 4], 'col2': [5, 6, 7, 8]}
df = pd.DataFrame(data=d)
ddf = dd.from_pandas(df, npartitions=2)
ddf.groupby(['col1']).col2.nunique().to_frame().compute()
See https://github.com/dask/dask/issues/6280 for more discussion.
To implement nunique in dask groupby you have to use an aggregate function.
import pandas as pd
import dask.dataframe as dd
def chunk(s):
'''
The function applied to the
individual partition (map)
'''
return s.apply(lambda x: list(set(x)))
def agg(s):
'''
The function whic will aggrgate
the result from all the partitions(reduce)
'''
s = s._selected_obj
return s.groupby(level=list(range(s.index.nlevels))).sum()
def finalize(s):
'''
The optional functional that will be
applied to the result of the agg_tu functions
'''
return s.apply(lambda x: len(set(x)))
tunique = dd.Aggregation('tunique', chunk, agg,finalize)
df = pd.DataFrame({
'col': [0, 0, 1, 1, 2, 3, 3] * 10,
'g0': ['a', 'a', 'b', 'a', 'b', 'b', 'a'] * 10,
})
ddf = dd.from_pandas(df, npartitions=10)
res = ddf.groupby(['col']).agg({'g0': tunique}).compute()
print(res)
See:
From the source code it looks like you can maybe do nunique outside of agg.
