Ndtyjky

I am having trouble figuring out how to work with xarray DataArrays and DataSets and perform algebra operations; especially when the dimensions have different levels and my cubes have different granularities. I would be very grateful if someone could suggest me some documentation or give me some advice.

In the example below I am trying to compute the contribution of each child (SKU) under a parent (PFS). I found that to get the right values I need to convert the cube slice into a pandas dataframe. Otherwise, Xarray duplicates the dimension I am working with.

import pandas as pd

import numpy as np

import xarray as xr

from itertools import product



# Create hierachies



usage_type_entities = (('Regular',), ('Sample',),

                       ('Tender',), ('Clinic Trial',))



usage_type_tree = pd.MultiIndex.from_tuples(

    usage_type_entities, names=('Usage_Type',))



product_tree_hierarchy = (("PF1", "PFS1", "SKU1"),

                          ("PF1", "PFS1", "SKU2"),

                          ("PF1", "PFS2", "SKU3"),

                          ("PF1", "PFS2", "SKU4"),

                          ("PF2", "PFS3", "SKU5"))



product_tree_entities = ("PF", "PFS", "SKU")



product_tree = pd.MultiIndex.from_tuples(product_tree_hierarchy,

                                         names=product_tree_entities)



market_tree_hierarchy = (("Group1", "Region1", "Market1"),

                         ("Group1", "Region1", "Market2"),

                         ("Group1", "Region2", "Market3"),

                         ("Group1", "Region2", "Market4"),

                         ("Group2", "Region3", "Market5"))



market_tree_entities = ("Groups", "Regions", "Markets")



market_tree = pd.MultiIndex.from_tuples(market_tree_hierarchy,

                                        names=market_tree_entities)



time_tree_hierarchy = [(y, y+q) for y, q in product([str(2013+x) for x in range(6)],

                                                    ["Q"+str(int(q)) for q in np.arange(1, 4.1, 1)])][0:22]



time_entities = ("Year", "Quarter")



time_tree = pd.MultiIndex.from_tuples(time_tree_hierarchy, names=time_entities)



# Create X-array Dataset



x1 = np.random.randint(100, size=(len(usage_type_tree), len(

    product_tree), len(market_tree), len(time_tree)))



xda = xr.DataArray(x1, coords=(usage_type_tree, product_tree, market_tree, time_tree),

                   dims=("Usage", "Product", "Market", "Time"))



# Operations - I need to convert my slice into a pandas df to get 

the right values. Converting to pandas df works ok. 



market = "Market1"

ut = "Regular"



(xda.sel(Markets=market, Usage_Type=ut)[:, 0].to_pandas() /

 xda.sel(Markets=market, Usage_Type=ut)[:, 0].to_pandas().groupby("PFS").sum(axis=0))

If I don't convert the slice to pandas df and I keep it as a xarray dataset, the dimension gets duplicated. For example, the line below produces a DatArray(Product: 5, Time: 22, PFS: 3), when it should be just (Product: 5, Time: 22)

(xda.sel(Markets=market, Usage_Type=ut)[:, 0] /

 xda.sel(Markets=market, Usage_Type=ut)[:, 0].groupby("PFS").sum(axis=0))