pandas dataframe create a new dataframe by duplicating n times rows of the previous dataframe and change date

Question

I have a dataframe with around around 9k rows and 57 cols, this is 'df'.

I need to have a new dataframe: 'df_final' - for each row of 'df' i have to replicate each row 'x' times and increase the day in each row one by one, also 'x' times. While i can do this for a couple of iterations, when i do it for the full length of 'df' ' len(df)' the loop it takes so long (>3 hours) that i actually had to cancel it. I have never seen the end of it. Here's the current code:

df.shape
output: (9454, 57)

df_int = df[0:0]
df_final = df_int[0:0]
range_df = len(df)
for x in range(0,2):
    df_int = df.iloc[0+x:x+1]
    if abs(df_int.iat[-1,3]) > 0:
        df_int = pd.concat([df_int]*abs(df_int.iat[-1,3]), ignore_index=True)
        for i in range(1, abs(df_int.iat[-1,3])):
            df_int['Consumption Date'][i] = df_int['Consumption Date'][i-1] + datetime.timedelta(days = 1)
            i += 1
       df_final = df_final.append(df_int, ignore_index=True)
    x += 1

The result of the loops for the first two rows of ' df' are below.

First two rows of df:

Desired result:

Is there another way to get to the desired output. It seems pandas do not deal very well with loops. In VBA excel the same loop takes around 3/4 minutes...i am trying to change a process which is currently in excel to python, however, if there's no way to make this work i guess i will stick to the old ways...

Vexatious

Answer 1

Use repeat and cumcount

In [2972]: dff = df.loc[df.index.repeat(3)]

In [2973]: dff
Out[2973]:
        date   name
0 2017-05-03    bob
0 2017-05-03    bob
0 2017-05-03    bob
1 2017-06-13  sally
1 2017-06-13  sally
1 2017-06-13  sally

In [2974]: dff.loc[:, 'date'] += pd.to_timedelta(dff.groupby(level=0).cumcount(), 'D')

In [2975]: dff
Out[2975]:
        date   name
0 2017-05-03    bob
0 2017-05-04    bob
0 2017-05-05    bob
1 2017-06-13  sally
1 2017-06-14  sally
1 2017-06-15  sally

---

Details

In [2976]: df
Out[2976]:
        date   name
0 2017-05-03    bob
1 2017-06-13  sally

In [2977]: dff.groupby(level=0).cumcount()
Out[2977]:
0    0
0    1
0    2
1    0
1    1
1    2
dtype: int64

Kila

really nice sol – Nickola

Answer 2

Let's use this toy DataFrame:

df = pd.DataFrame({
    'date': pd.to_datetime(['2017-05-03', '2017-06-13']),
    'name': ['bob', 'sally'],
})

It looks like this:

        date   name
0 2017-05-03    bob
1 2017-06-13  sally

Then:

x = 3 # repeat count
ind = np.repeat(np.arange(len(df)), x) # 0,0,0,1,1,1
df_final = df.iloc[ind].copy()

That gives you the repeats:

        date   name
0 2017-05-03    bob
0 2017-05-03    bob
0 2017-05-03    bob
1 2017-06-13  sally
1 2017-06-13  sally
1 2017-06-13  sally

Now you just need to increment the dates:

inc = np.tile(np.arange(x), len(df)) # 0,1,2,0,1,2
df_final.date += pd.to_timedelta(inc, 'D')

And you get:

        date   name
0 2017-05-03    bob
0 2017-05-04    bob
0 2017-05-05    bob
1 2017-06-13  sally
1 2017-06-14  sally
1 2017-06-15  sally

Piano

Answer 3

Here is one solution

df1=df.reset_index().set_index('date').groupby('index').\
    apply(lambda x :x.reindex(pd.date_range(start=x.index[0],periods=3,freq='D'))).ffill()
df1
Out[202]: 
                  index   name
index                         
0     2017-05-03    0.0    bob
      2017-05-04    0.0    bob
      2017-05-05    0.0    bob
1     2017-06-13    1.0  sally
      2017-06-14    1.0  sally
      2017-06-15    1.0  sally

Then

df1.drop('index',1).reset_index().rename(columns={'level_1':'date'}).drop('index',1)

Out[212]: 
        date   name
0 2017-05-03    bob
1 2017-05-04    bob
2 2017-05-05    bob
3 2017-06-13  sally
4 2017-06-14  sally
5 2017-06-15  sally

Delvalle