CODEWITHSUNDEEP
pythonpandasdaskseries
Matt Elgazar
Matt Elgazar

Reputation: 735

Dask item assignment. Cannot use loc for item assignment

I have a folder of parquet files that I can't fit in memory so I am using dask to perform the data cleansing operations. I have a function where I want to perform item assignment but I can't seem to find any solutions online that qualify as solutions to this particular function. Below is the function that works in pandas. How do I get the same results in a dask dataframe? I thought delayed might help but all of the solutions I've tried to write haven't been working.

def item_assignment(df):

    new_col = np.bitwise_and(df['OtherCol'], 0b110)

    df['NewCol'] = 0
    df.loc[new_col == 0b010, 'NewCol'] = 1
    df.loc[new_col == 0b100, 'NewCol'] = -1 

    return df

TypeError: '_LocIndexer' object does not support item assignment

Upvotes: 13

Views: 7279

Answers (2)

Sven Harris
Sven Harris

Reputation: 2939

You can use map_partitions in this case where you can use raw pandas functionality. I.e.

ddf.map_partitions(item_assignment)

this operates on the individual pandas constituent dataframes of the dask dataframe

df = pd.DataFrame({"OtherCol":[0b010, 0b110, 0b100, 0b110, 0b100, 0b010]})
ddf = dd.from_pandas(df, npartitions=2)
ddf.map_partitions(item_assignment).compute()

And we see the result as expected:

   OtherCol  NewCol
0         2       1
1         6       0
2         4      -1
3         6       0
4         4      -1
5         2       1

Upvotes: 6

jpp
jpp

Reputation: 164843

You can replace your loc assignments with dask.dataframe.Series.mask:

df['NewCol'] = 0
df['NewCol'] = df['NewCol'].mask(new_col == 0b010, 1)
df['NewCol'] = df['NewCol'].mask(new_col == 0b100, -1)

Upvotes: 21

Related Questions