How to groupby multiple columns and aggregate data in pandas

Question

I have a pandas dataframe with several columns ( words, start time, stop time, speaker ):

• I want to combine all values in the word column while the values in the data column do not change OR the values in the column meta_data do not change.
• In addition, I want to keep the start value for the first word and the stop value for the last word in the combination.

I currently have:

      word  start  stop     data    meta_data  
0      but   2.72  2.85        2      9
1   that's   2.85  3.09        2      9
2  alright   3.09  3.47        2      1
3    we'll   8.43  8.69        1      4
4     have   8.69  8.97        1      4
5       to   8.97  9.07        1      4
6     okay   9.19 10.01        2      2
7     sure  10.02 11.01        2      1
8    what?  11.02 12.00        1      4

However, I would like to turn this into:

       word       start  start  data  meta_data
0  but that's      2.72   3.09     2    9
1  alright         3.09   3.47     2    1
2  we'll have to   8.43   9.07     1    4
3  okay            9.19  10.01     2    2
4  sure            10.02 11.01     2    1
5  what?           11.02 12.00     1    4

Answer 1

This need to create a help key, then we do shift + cumsum create the groupkey based on that

df['Key']=df[['data','meta_data']].apply(tuple,1)
d={'word':' '.join,'start':'min','stop':'max','data':'first','meta_data':'first'}
df.groupby(df.Key.ne(df.Key.shift()).cumsum()).agg(d).reset_index(drop=True)
Out[171]: 
            word  start   stop  data  meta_data
0     but that's   2.72   3.09     2          9
1        alright   3.09   3.47     2          1
2  we'll have to   8.43   9.07     1          4
3           okay   9.19  10.01     2          2
4           sure  10.02  11.01     2          1
5          what?  11.02  12.00     1          4

Answer 2

doing some math here + GroupBy.agg

s=df['data']+df['meta_data']
groups=s.ne(s.shift()).cumsum()
new_df=( df.groupby(groups)
           .agg({'word':' '.join,'start':'min',
                 'stop':'max','data':'first',
                 'meta_data':'first'}) )
print(new_df)

            word  start   stop  data  meta_data
1     but that's   2.72   3.09     2          9
2        alright   3.09   3.47     2          1
3  we'll have to   8.43   9.07     1          4
4           okay   9.19  10.01     2          2
5           sure  10.02  11.01     2          1
6          what?  11.02  12.00     1          4

---

if you think that the sum can correspond in two different and consecutive groups you can use a somewhat more complex function with decimals

p=(df['data']+0.1723).pow(df['meta_data']+2.017)
groups=p.ne(p.shift()).cumsum()