Pandas resampling from months to weeks

Question

I am attempting to downsample monthly data to weekly data and have a time series dataframe of months that looks like this:

             qty
PERIOD_NAME 
2017-09-01  49842.0
2017-10-01  27275.0
2017-11-01  29159.0
2017-12-01  51344.0
2018-01-01  19103.0
2018-02-01  23570.0
2018-03-01  45139.0
2018-04-01  25722.0
2018-05-01  22644.0

I've attempted using a resample to weeks like this:

tgt_item_by_445_wk = tgt_item_by_445_wk.resample('W').sum()

which yields:

             qty
PERIOD_NAME 
2017-09-03  49842.0
2017-09-10  0.0
2017-09-17  0.0
2017-09-24  0.0
2017-10-01  27275.0
2017-10-08  0.0
2017-10-15  0.0
2017-10-22  0.0
2017-10-29  0.0

I've tried interpolation, but I can't get what I am looking for, which is a fill of the unsampled (0's) with an even split of the first value like this:

              qty
PERIOD_NAME 
2017-09-03  12460.5
2017-09-10  12460.5
2017-09-17  12460.5
2017-09-24  12460.5
2017-10-01  5455.0
2017-10-08  5455.0
2017-10-15  5455.0
2017-10-22  5455.0
2017-10-29  5455.0

Is there some method using resample, fills and interpolation that allows this?

Answer 1

Let's try asfreq and groupby.

v = df.asfreq('W', method='ffill')
v /= v.groupby(v.index.strftime('%Y-%m')).transform('count')

                  qty
PERIOD_NAME          
2017-09-03   12460.50
2017-09-10   12460.50
2017-09-17   12460.50
2017-09-24   12460.50
2017-10-01    5455.00
2017-10-08    5455.00
2017-10-15    5455.00
2017-10-22    5455.00
2017-10-29    5455.00
2017-11-05    7289.75
2017-11-12    7289.75
2017-11-19    7289.75
2017-11-26    7289.75
2017-12-03   10268.80
2017-12-10   10268.80
2017-12-17   10268.80
2017-12-24   10268.80
2017-12-31   10268.80
2018-01-07    4775.75
2018-01-14    4775.75
2018-01-21    4775.75
2018-01-28    4775.75
2018-02-04    5892.50
2018-02-11    5892.50
2018-02-18    5892.50
2018-02-25    5892.50
2018-03-04   11284.75
2018-03-11   11284.75
2018-03-18   11284.75
2018-03-25   11284.75
2018-04-01    5144.40
2018-04-08    5144.40
2018-04-15    5144.40
2018-04-22    5144.40
2018-04-29    5144.40

This works well since your values are always on the first of each month. Alternatively, you may use

v /= v.groupby(v.qty).transform('count').values 

for the second step.