How to transform variable over time, with time_delta as a dimension

[ConstantinVasilev]

Hi, I have a DataArray which variable represents exchange rates, defined over dimensions:

• currency
• time

I want to:

1. Apply transformation to the variable across time dimension (in that case difference but may be other function)
2. Consider only positive time deltas representing only future periods.
3. Resulting transformed variable should have dimensions:

• currency
• time
• horizon_delta

I got to point 2. The transformation is equivalent to taking upper triangular matrix and aligning all non-null values to the left.

import numpy as np
import xarray as xr
import pandas as pd
np.random.seed(1)

currency = ['GBP', 'EUR', 'USD']
n_currency = len(currency)

date = pd.date_range(start='2020-01-01', periods=100)
n_date = len(date)

variable = xr.DataArray(
    data=np.random.uniform(0, 1, (n_currency,n_date)),
    coords={
        'currency': currency,
        'date': date
    }
)

# Calc variable delta over time, including all possible time_deltas, even negative
variable_delta = variable.rename({'date': 'date_future'}) - variable

# Calculate time_delta and filter out negative -> horizon_delta
horizon_delta = variable_delta.date_future - variable_delta.date
horizon_delta = horizon_delta.where(horizon_delta > pd.Timedelta('0D'))