I have a Pandas Series of lists of strings:
0 [slim, waist, man]
1 [slim, waistline]
2 [santa]
As you can see, the lists vary by length. I want an efficient way to collapse this into one series
0 slim
1 waist
2 man
3 slim
4 waistline
5 santa
I know I can break up the lists using
series_name.split(' ')
But I am having a hard time putting those strings back into one list.
Thanks!
Here's a simple method using only pandas functions:
import pandas as pd
s = pd.Series([
['slim', 'waist', 'man'],
['slim', 'waistline'],
['santa']])
Then
s.apply(pd.Series).stack().reset_index(drop=True)
gives the desired output. In some cases you might want to save the original index and add a second level to index the nested elements, e.g.
0 0 slim
1 waist
2 man
1 0 slim
1 waistline
2 0 santa
If this is what you want, just omit .reset_index(drop=True) from the chain.
In pandas version 0.25.0 appeared a new method 'explode' for series and dataframes. Older versions do not have such method.
It helps to build the result you need.
For example you have such series:
import pandas as pd
s = pd.Series([
['slim', 'waist', 'man'],
['slim', 'waistline'],
['santa']])
Then you can use
s.explode()
To get such result:
0 slim
0 waist
0 man
1 slim
1 waistline
2 santa
In case of dataframe:
df = pd.DataFrame({
's': pd.Series([
['slim', 'waist', 'man'],
['slim', 'waistline'],
['santa']
]),
'a': 1
})
You will have such DataFrame:
s a
0 [slim, waist, man] 1
1 [slim, waistline] 1
2 [santa] 1
Applying explode on s column:
df.explode('s')
Will give you such result:
s a
0 slim 1
0 waist 1
0 man 1
1 slim 1
1 waistline 1
2 santa 1
If your series, contain empty lists
import pandas as pd
s = pd.Series([
['slim', 'waist', 'man'],
['slim', 'waistline'],
['santa'],
[]
])
Then running explode will introduce NaN values for empty lists, like this:
0 slim
0 waist
0 man
1 slim
1 waistline
2 santa
3 NaN
If this is not desired, you can dropna method call:
s.explode().dropna()
To get this result:
0 slim
0 waist
0 man
1 slim
1 waistline
2 santa
Dataframes also have dropna method:
df = pd.DataFrame({
's': pd.Series([
['slim', 'waist', 'man'],
['slim', 'waistline'],
['santa'],
[]
]),
'a': 1
})
Running explode without dropna:
df.explode('s')
Will result into:
s a
0 slim 1
0 waist 1
0 man 1
1 slim 1
1 waistline 1
2 santa 1
3 NaN 1
with dropna:
df.explode('s').dropna(subset=['s'])
Result:
s a
0 slim 1
0 waist 1
0 man 1
1 slim 1
1 waistline 1
2 santa 1
You are basically just trying to flatten a nested list here.
You should just be able to iterate over the elements of the series:
slist =[]
for x in series:
slist.extend(x)
or a slicker (but harder to understand) list comprehension:
slist = [st for row in s for st in row]
series_name.sum()
does exactly what you need. Do make sure it's a series of lists otherwise your values will be concatenated (if string) or added (if int)
You can try using itertools.chain to simply flatten the lists:
In [70]: from itertools import chain
In [71]: import pandas as pnd
In [72]: s = pnd.Series([['slim', 'waist', 'man'], ['slim', 'waistline'], ['santa']])
In [73]: s
Out[73]:
0 [slim, waist, man]
1 [slim, waistline]
2 [santa]
dtype: object
In [74]: new_s = pnd.Series(list(chain(*s.values)))
In [75]: new_s
Out[75]:
0 slim
1 waist
2 man
3 slim
4 waistline
5 santa
dtype: object
If your pandas version is too old to use series_name.explode(), this should work too:
from itertools import chain
pd.Series(
chain.from_iterable(
value
for i, value
in series_name.iteritems()
)
)
You can use the list concatenation operator like below -
lst1 = ['hello','world']
lst2 = ['bye','world']
newlst = lst1 + lst2
print(newlst)
>> ['hello','world','bye','world']
Or you can use list.extend() function as below -
lst1 = ['hello','world']
lst2 = ['bye','world']
lst1.extend(lst2)
print(lst1)
>> ['hello', 'world', 'bye', 'world']
Benefits of using extend function is that it can work on multiple types, where as concatenation operator will only work if both LHS and RHS are lists.
Other examples of extend function -
lst1.extend(('Bye','Bye'))
>> ['hello', 'world', 'Bye', 'Bye']
Flattening and unflattening can be done using this function
def flatten(df, col):
col_flat = pd.DataFrame([[i, x] for i, y in df[col].apply(list).iteritems() for x in y], columns=['I', col])
col_flat = col_flat.set_index('I')
df = df.drop(col, 1)
df = df.merge(col_flat, left_index=True, right_index=True)
return df
Unflattening:
def unflatten(flat_df, col):
flat_df.groupby(level=0).agg({**{c:'first' for c in flat_df.columns}, col: list})
After unflattening we get the same dataframe except column order:
(df.sort_index(axis=1) == unflatten(flatten(df)).sort_index(axis=1)).all().all()
>> True
You may also try:
combined = []
for i in s.index:
combined = combined + s.iloc[i]
print(combined)
s = pd.Series(combined)
print(s)
output:
['slim', 'waist', 'man', 'slim', 'waistline', 'santa']
0 slim
1 waist
2 man
3 slim
4 waistline
5 santa
dtype: object
The accepted answer (by @mcwitt) looks nicely pandas-ish, but is awfully slow, is extremely memory hungry if there are outliers in the size of lists, and buggy (see comments to that answer).
+1 for @Tadej Magajna for his answer, taking the sum() over the series. Since it is adding lists together, a more efficient way is using numpy's flatten() in case the series elements are nparrays:
series_name.values.flatten().
