熊猫df差异数据必须为一维

Question

I am trying to calculate the difference in item price for combinations of shop and item like this:

index_cols = ['shop_id', 'item_id', 'date_block_num']

#get aggregated values for (shop_id, item_id, month)
gb = sales.groupby(index_cols).agg({'item_cnt_day':[np.sum], 'item_price':[np.mean]}).reset_index()\
.rename(columns={'sum': 'item_cnt_month','mean':'item_price'})

gb['diff'] = gb.groupby(['shop_id','item_id'])['item_price'].transform(np.diff).fillna(0)

gb

As you can see I am trying to use np.diff (from numpy) for faster computation buy I am getting the following error:

Exception: Data must be 1-dimensional

EDIT:

Data Sample:

            shop_id   item_id date_block_num item_cnt_day item_price
            0         30      1              31.0         265.0
            0         31      1              11.0         434.0
            0         32      0              6.0          221.0
            0         32      1              10.0         221.0
            0         33      0              3.0          347.0
            59        22164   27             2.0          699.0
            59        22164   30             1.0          699.0
            59        22167   9              1.0          299.0
            59        22167   11             2.0          299.0
            59        22167   17             1.0          299.0

Any idea to avoid this error while using numpy or a faster way to do it? Thanks.

Answer 1

Remove one element lists for [np.mean] and [np.sum] to np.mean and np.sum for prevent MultiIndex in columns:

gb = (sales.groupby(index_cols)
           .agg({'item_cnt_day':np.sum, 'item_price':np.mean})
           .reset_index()
           .rename(columns={'sum': 'item_cnt_month','mean':'item_price'}))

Then is possible use (but not 100% sure if better performance):

gb['diff'] = gb.groupby(['shop_id','item_id'])['item_price'].diff()

EDIT:

Data sample:

index_cols = ['shop_id', 'item_id', 'date_block_num']

sales = pd.DataFrame({
        'item_id':list('aaaaaa'),
         'shop_id':list('aaabbb'),
         'date_block_num':[4,5,4,5,5,4],
         'item_cnt_day':[7,8,9,4,2,3],
         'item_price':[1,3,5,7,1,0]

})

---

gb = (sales.groupby(index_cols)
           .agg({'item_cnt_day':[np.sum], 'item_price':[np.mean]})
           .reset_index()
           .rename(columns={'sum': 'item_cnt_month','mean':'item_price'}))
print (gb)
  shop_id item_id date_block_num   item_cnt_day item_price
                                 item_cnt_month item_price
0       a       a              4             16          3
1       a       a              5              8          3
2       b       a              4              3          0
3       b       a              5              6          4

---

gb = (sales.groupby(index_cols)
           .agg({'item_cnt_day':np.sum, 'item_price':np.mean})
           .reset_index()
           .rename(columns={'sum': 'item_cnt_month','mean':'item_price'}))
print (gb)

  shop_id item_id  date_block_num  item_cnt_day  item_price
0       a       a               4            16           3
1       a       a               5             8           3
2       b       a               4             3           0
3       b       a               5             6           4