Tried to do this with pandas.Series.apply function but it consider to be slow on big amount of data. Is there any quicker way to replace values?
Here is what I've tried, but it's slow on big Series (with million items for example)
s = pd.Series([1, 2, 3, 'str1', 'str2', 3])
s.apply(lambda x: x if type(x) == str else np.nan)
Use to_numeric with errors='coerce':
pd.to_numeric(s, errors='coerce')
If need also integers add Int64:
pd.to_numeric(s, errors='coerce').astype('Int64')
EDIT: You can use isinstance with map, and also Series.where:
#test 600k
N = 100000
s = pd.Series([1, 2, 3, 'str1', 'str2', 3] * N)
In [152]: %timeit s.apply(lambda x: x if type(x) == str else np.nan)
196 ms ± 2.81 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
In [153]: %timeit s.map(lambda x: x if isinstance(x, str) else np.nan)
174 ms ± 3.66 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
In [154]: %timeit s.where(s.map(lambda x: isinstance(x, str)))
168 ms ± 3.63 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
In [155]: %timeit s.where(pd.to_numeric(s, errors='coerce').isna())
366 ms ± 3.19 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
