Add example of apply_ufunc

[jemmajeffree]

I've written up an example of how to calculate the trend and significance at each gridpoint using xr.apply_ufunc to pass things through to scipy.stats.linregress, so it's vectorised and much faster than using a for loop

Is this something that's of use to the COSIMA community?

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[ashjbarnes]

This looks great! I've added a comment to suggest a bit more motivation at the top of the notebook :)

[anton-seaice]

Thankyou for doing this Jemma!

I think this could go in Tutorials rather than Examples ?

(I will also point out xarray has a polyfit which does the same as the notebook :) )

If we move to the intake catalog, then the chunking warnings will go away :)

And it would be good to fix up the figure to be tripole aware ... although I guess its not actually essential.

[navidcy]

@jemmajeffree thanks!

I pushed few edits... space after commas etc

the PR looks good to me! I run the notebook and it's all good!

I just don't understand what np.arange(sst.time.shape[0])/12 does...
Seems something hardcoded to me... something related with months (or the conspicuous coincidental 12)... but I have no idea.

If we can reduce the hardcoding it'd be great. Otherwise a comment next to it explaining what's happening will be useful for somebody reading this example down the road. But I am happy to merge this even as is rather than keep ruminating over details!

[navidcy]

Also @ashjbarnes suggestion is nice. If you can incorporate this before merging it'd be nice!

[navidcy]

@jemmajeffree I think I'll include a sentence along what @ashjbarnes suggested and merge. Unless you plan to edit/work on this?

[navidcy]

(We can always open a new PR in the future and revisit-enhance the recipe!)

[jemmajeffree]

I've updated the example here /g/data/x77/jj8842/cosima-recipes/Examples/Apply_scipy_function_to_every_gridpoint.ipynb, but I can't work out how to push it back to the PR and merge with existing changes here. I can try follow instructions, or I'm very happy to have someone else work out how to merge them again.

I've:
- added some text as suggested by @ashjbarnes
- tried to clarify what is meant by the np.arange(sst.time.shape[0])/12), which is specific to the linear regression, so I guess hardcoded to this example but I can't see how you'd generalise when you don't know what other functions might be used with this code and what variables they would take

Also noting the comments by @anton-seaice, I know you could do this specific task faster with xr.polyfit, but some things you can't and linregress is an approachable example

Feel free to rewrite any of my explanations, or add anything you think they're lacking

[navidcy]

We can zoom and I show you?

[navidcy]

oh I see... probably you can't push because I pushed few edits and you need to merge them! damn... I made it hard....

[navidcy]

I'll sort it out!

[navidcy]

new example -> https://cosima-recipes.readthedocs.io/en/latest/Recipes/Apply_function_to_every_gridpoint.html

thanks @jemmajeffree!

[jemmajeffree]

Thanks @navidcy for sorting that out!