CODEWITHSUNDEEP
pythonpython-3.xpandasdataframeseries
PedroA
PedroA

Reputation: 1925

Mask dataframe matching multiple conditions

I would like mask (or assign 'NA') the value of a column in a dataframe if two conditions are met. This would be relatively straightforward if the conditions were performed row-wise, with something like:

mask = ((df['A'] < x) & (df['B'] < y))
df.loc[mask, 'C'] = 'NA'

but I'm having some trouble figuring out of how to perform this task in my dataframe, which is structured more or less like:

df = pd.DataFrame({ 'A': (188, 750, 1330, 1385, 188, 750, 810, 1330, 1385),
                     'B': (2, 5, 7, 2, 5, 5, 3, 7, 2),
                     'C': ('foo', 'foo', 'foo', 'foo', 'bar', 'bar', 'bar', 'bar', 'bar') })

    A    B C
0   188  2 foo
1   750  5 foo
2   1330 7 foo
3   1385 2 foo
4   188  5 bar
5   750  5 bar
6   810  3 bar
7   1330 7 bar
8   1385 2 bar

The values in column 'A' when 'C' == 'foo' should also be found when 'C' == 'bar' (something like an index), although it can have missing data in both 'foo' and 'bar'. How can I mask (or assign 'NA') the rows of column 'B' if both 'foo' and 'bar' are lower than 5 or any of them is missing? In the example above the output would be something like:

    A    B C
0   188  2  foo
1   750  5  foo
2   1330 7  foo
3   1385 NA foo
4   188  5  bar
5   750  5  bar
6   810  NA bar
7   1330 7  bar
8   1385 NA bar

Upvotes: 4

Views: 4885

Answers (2)

PedroA
PedroA

Reputation: 1925

Another possible solution using groupby and some other ideas borrowed from jpp's answer:

# create a mapping test for each group from column 'A'
fmap = df.groupby(['A']).apply(lambda x: all(x['B'] < 5))
# and generate a new masking map from that
mask_map = df['A'].map(fmap)
# then just mask the values in the original DF
df['B'] = df['B'].mask(mask_map)

      A    B    C
0   188  2.0  foo
1   750  5.0  foo
2  1330  7.0  foo
3  1385  NaN  foo
4   188  5.0  bar
5   750  5.0  bar
6   810  NaN  bar
7  1330  7.0  bar
8  1385  NaN  bar

Upvotes: 1

jpp
jpp

Reputation: 164843

Here's one solution. The idea is to construct two Boolean masks, m1 and m2, from two mapping series, s1 and s2. Then use pd.Series.mask to mask series B.

# create separate mappings for foo and bar
s1 = df.loc[df['C'] == 'foo'].set_index('A')['B']
s2 = df.loc[df['C'] == 'bar'].set_index('A')['B']

# use -np.inf to cover missing mappings
m1 = df['A'].map(s1).fillna(-np.inf).lt(5)  
m2 = df['A'].map(s2).fillna(-np.inf).lt(5)

df['B'] = df['B'].mask(m1 & m2)

print(df)

      A    B    C
0   188  2.0  foo
1   750  5.0  foo
2  1330  7.0  foo
3  1385  NaN  foo
4   188  5.0  bar
5   750  5.0  bar
6   810  NaN  bar
7  1330  7.0  bar
8  1385  NaN  bar

Upvotes: 2

Related Questions