Assuming the following DataFrame:
key.0 key.1 key.2 topic
1 abc def ghi 8
2 xab xcd xef 9
How can I combine the values of all the key.* columns into a single column 'key', that's associated with the topic value corresponding to the key.* columns? This is the result I want:
topic key
1 8 abc
2 8 def
3 8 ghi
4 9 xab
5 9 xcd
6 9 xef
Note that the number of key.N columns is variable on some external N.
You can melt your dataframe:
>>> keys = [c for c in df if c.startswith('key.')]
>>> pd.melt(df, id_vars='topic', value_vars=keys, value_name='key')
topic variable key
0 8 key.0 abc
1 9 key.0 xab
2 8 key.1 def
3 9 key.1 xcd
4 8 key.2 ghi
5 9 key.2 xef
It also gives you the source of the key.
From v0.20, melt is a first class function of the pd.DataFrame class:
>>> df.melt('topic', value_name='key').drop('variable', axis=1)
topic key
0 8 abc
1 9 xab
2 8 def
3 9 xcd
4 8 ghi
5 9 xef
After trying various ways, I find the following is more or less intuitive, provided stack's magic is understood:
# keep topic as index, stack other columns 'against' it
stacked = df.set_index('topic').stack()
# set the name of the new series created
df = stacked.reset_index(name='key')
# drop the 'source' level (key.*)
df.drop('level_1', axis=1, inplace=True)
The resulting dataframe is as required:
topic key
0 8 abc
1 8 def
2 8 ghi
3 9 xab
4 9 xcd
5 9 xef
You may want to print intermediary results to understand the process in full. If you don't mind having more columns than needed, the key steps are set_index('topic'), stack() and reset_index(name='key').
name argument for reset_index, could you explain how it works?
OK , cause one of the current answer is mark as duplicated of this question, I will answer here.
By Using wide_to_long
pd.wide_to_long(df, ['key'], 'topic', 'age').reset_index().drop('age',1)
Out[123]:
topic key
0 8 abc
1 9 xab
2 8 def
3 9 xcd
4 8 ghi
5 9 xef
