pyxarr

This is a package which provides a minimal and light-weight class to work with multi-dimensional labeled arrays.

Description

The software of pyxarr is written from scratch. I have tried to mimic the classes DataArray, Dataset and Coords from xarray. My main goal is to keep the code light and fast.

Installation

Pyxarr is available as pyxarr on PyPI. To install it use pip:

$ pip install pyxarr

The module pyxarr requires Python3.10+ and Python modules: h5py, numpy.

Usage

Working with pyxarr DataArray:

import numpy as np
from pyxarr import DataArray

xda = DataArray()
bool(xda) # will return False, thus if xda: ... will work
len(xda)  # will return 0

rng = np.random.default_rng()
xda = DataArray(
   rng.random((120, 11, 17)),
   dims=("time", "y", "x"),
   coords=(
      np.arange("2025-02-24T14:32:00", "2025-02-24T15:32:00", 30, dtype="datetime64[s]"),
      list(range(11)),
      np.arange(17),
   ),
   attrs={
      "long_name": "noisy signal",
      "units": "1",
   },
)
bool(xda)  # returns True
len(xda)   # returns 120
xda.shape  # returns (120, 11, 17)
xda.size   # returns 22440
"x" in xda.coords  # returns True
xda[4, :, :]  # slicing works
xda.coords += ("orbit", list(range(500, 620))  # will add an auxiliary coordinate
xda.swap_dims("orbit", "time")  # will make the auxiliary coordinate the dimension coordinate and visa versa
xda.mean("time")  # will return a new DataArray averaged over the time axis
xda + xda2  # will return a new DataArray with the sum of the data of both arrays, other supported operators are sub, div and mul.

Working with pyxarr Dataset:

from pyxarr import Dataset

xds = Dataset()
bool(xds)  # will return False, thus if xds: ... will work
len(xds)   # will return 0

xds["foo"] = xda  # adding data to a dataset all dimensions will be added as coordinates
xds.attrs["title"] = "test example"  # adding attributes to a dataset

Working with pyxarr Coords:

import numpy as np
from pyxarr import Coords

coords = Coords()
bool(coords)  # will return False, thus if xds: ... will work
len(coords)   # will return 0

coords += ("time", np.arange("2025-07-01", "2025-08-01", dtype="datetime64[D]))
coords += ("column", np.arange(31))  # adding coordinates

coords = {
   "time": np.arange("2025-07-01", "2025-08-01", dtype="datetime64[D]),
   "column": np.arange(31),
}  # using a dictionary to define the coordinates

coords = [
   ("time", np.arange("2025-07-28T12:43:00", "2025-07-28T12:58:00", dtype="datetime64[s])),
   ("column", np.arange(1000)),
   ("row", np.arange(256)),
]  # using a list of tuples to define the coordinates

Time coordinates

Using the CF convensions described in section 4.4, we can define a simple time-coordinate variable as:

 double time(time) ;
   time:axis = "T" ;
   time:standard_name = "time" ;
   time:units = "days since 2024-01-01" ; // defaults to UTC

or

 double time(time) ;
   time:long_time = "sample time" ;
   time:axis = "T" ;
   time:standard_name = "time" ;
   time:units = "seconds since 2024-01-01 00:00:00" ; // defaults to UTC
   time:calendar = "proleptic_gregorian" ; // or utc, tai, etc.
   time:valid_min = 0. ;
   time:valid_max = 172800. ;

You can store any time-coordinate with all its attributes into an pyxarr Coords. However, often it handy to have the data read into a numpy.datetime64, this is an option which is offered in the pyxarr tools: dest_from_h5 and dest_to_h5. Both tools will only perform the conversion when the precision (level of detail or granularity present) is specified. This can be: days, hours, minutes, seconds, milli-seconds, micro-seconds, or nano-seconds.

TBW

Authors and acknowledgment

The code is developed by R.M. van Hees (SRON)

License

• Copyright: R.M. van Hees (SRON) (https://www.sron.nl)
• License: Apache-2.0