Monthly alignment of multiple daily time series

Question

So I have this dataset containing sales data for N items between day d1 and day d2. For each purchase, I have the time stamp, the customer ID and the item ID. My goal is to generate a dataframe of size (M x N), where df[ i, j ] is the total number of purchases for item j during month i.

Generate mockup data

import pandas as pd
import random

d1 = '2014-1-1'
d2 = '2014-3-31'
daily = pd.date_range( d1, d2, freq='D' )
npurchase = 1000
nitem = 20
olddf = pd.DataFrame( { 'dt': [ random.choice( daily ) for _ in xrange( npurchase) ], 'itemID': [ randint(nitem) for _ in xrange( npurchase ) ] } )
olddf.head()

Output:

                  dt    itemID
0   2014-02-24 00:00:00  19
1   2014-01-29 00:00:00  0
2   2014-01-27 00:00:00  7
3   2014-02-03 00:00:00  12
4   2014-01-24 00:00:00  3

Resample and align

rng = pd.date_range( d1, d2, freq='M')
newdf = pd.DataFrame( index=rng )

for name, group in olddf.groupby( 'itemID' ) :
    tmp = group.groupby( 'dt' ).size().resample( 'M', how='sum' )
    newdf[ name ] = tmp

newdf.fillna( 0, inplace=True )
newdf.ix[ :, :5 ]

Output

             0  1   2   3   4   5
2014-01-31   15  21  25  17  10  14
2014-02-28   10  13  16  20  15  8
2014-03-31   12  25  14  14  26  12

Is there a more efficient / elegant way to do it?

Answer 1

The following code is 2.5x faster and 4x more elegant than my initial try:

olddf[ 'ones' ] = 1
newdf2 = olddf.set_index( 'dt' ).groupby( 'itemID' )[ 'ones' ].resample( 'M', how='sum' ).unstack( 0 )
print newdf2[ range(5) ]

Output

itemID       0   1   2   3   4
dt                            
2014-01-31  10  21  19  23  15
2014-02-28   8  11  12  20  11
2014-03-31  21  18  13  12  18