# Usage of xhistogram compared to np.digitize

**URL:** <https://discourse.pangeo.io/t/usage-of-xhistogram-compared-to-np-digitize/1381>\
**Category:** Science\
**Created:** [April 9, 2021, 8:41pm UTC](https://discourse.pangeo.io/t/usage-of-xhistogram-compared-to-np-digitize/1381 "2021-04-09T20:41:25Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![raybellwaves](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pangeo.io/raybellwaves/32/782_2.png) [@raybellwaves](https://discourse.pangeo.io/u/raybellwaves)\
**Post date:** [April 9, 2021, 8:41pm UTC](https://discourse.pangeo.io/t/usage-of-xhistogram-compared-to-np-digitize/1381/1 "2021-04-09T20:41:25Z")

</div>

I have recently been going though a course on geospatial python which uses xarray and rioxarray 😀  
[https://carpentries-incubator.github.io/geospatial-python/](https://carpentries-incubator.github.io/geospatial-python/)

There are a few times in the course the authors move to numpy functions. Obviously there is nothing wrong with that. I would just like to make awareness of xarray functionality or packages in the pangeo stack that may do something similar and stay within an xarray/dask framework for scalability. i.e. I found an example of where the authors could use quantile with an xarray object over using np.percentile on the values of the arrays ([use da.quantile in notebook 05 · Issue #43 · carpentries-incubator/geospatial-python · GitHub](https://github.com/carpentries-incubator/geospatial-python/issues/43))

Similarly, i’m curious if [xhistogram](https://github.com/xgcm/xhistogram) (which leverages functionally of xarray and dask) can be used over np.digitize.

np.digitize is used in this lesson [Raster Calculations in Python – Introduction to Geospatial Raster and Vector Data with Python](https://carpentries-incubator.github.io/geospatial-python/07-raster-calculations/index.html) down in the section on “Classifying Continuous Rasters in Python”.

Rather than using the data in the course here is an MCVE:

```
# numpy version
import numpy as np
arr = np.random.lognormal(mean=1, sigma=1.0, size=(1, 1367, 1697)) -1
bins = [-1, 2, 10, 20, np.inf]
expected = np.digitize(nparry, bins)

# xhistogram version
import xarray as xr
da = xr.DataArray(arr, name="foo")
...
actual = ...values

# Check both give same result
import numpy.testing as npt
npt.assert_almost_equal(expected, actual)
```

---

<div class="post-metadata">

**Author:** ![RichardScottOZ](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pangeo.io/richardscottoz/32/752_2.png) [@RichardScottOZ](https://discourse.pangeo.io/u/RichardScottOZ)\
**Post date:** [April 10, 2021, 12:06am UTC](https://discourse.pangeo.io/t/usage-of-xhistogram-compared-to-np-digitize/1381/2 "2021-04-10T00:06:42Z")

</div>

Quantile and percentile have dask limitations and multidimensional limitations, perhaps?