Add nicer exception message for mistake in coords argument

[canyon289]

Fix for #213

If you guys like how this looks I'll change all the other plots to this pattern. Did a couple as an example

# Sample code
import arviz as az
data = az.load_arviz_data('non_centered_eight')
az.traceplot(data, var_names=('theta_t'), coords={'theta_t_dim_0': [0, 1, 100]})
az.traceplot(data, var_names=('theta_t'), coords={'theta_t_dim_0_t': [0, 1]})

[ColCarroll]

This looks good to me.

As an aside, there's lots of reorganizing to do - most functions in utils, plot_utils and xarray_utils should probably be in a more general spot (data_utils?), but that shouldn't happen here.

[ColCarroll]

can you verify that the stack trace is still informative? I think the pattern is

except ValueError as err:
    raise ValueError("Verify coords keys") from err

I might also print out more information - like "make sure {list(coords.keys()} match with the data data_vars ({list(data.data_vars)})"

[canyon289]

If you do from err the text from the original exception is lost. I was keeping it in because it was informative as to which coordinates were missing.

But I agree with your second statement which means the above point will be irrelevant! I'll make the changes and post another example.

[ColCarroll]

...are these right? A KeyError gets thrown for bad .values and a ValueError for bad keys?

[canyon289]

Maybe not. I'll verify. Thanks for double checking

[canyon289]

So yes,
Having a wrong key in coords dictionary raises a ValueError in xarray, and having a value in the coords dictionary raises a KeyError.

I wrote tests but just double checked by hand as well

Ill make the exceptions more clear

[canyon289]

There's a gist below of both xarray defined ValueError and KeyError, and also two Options at the bottom to consider for arviz

Let me know which one you guys like, (or any other suggestions to try) and I'll implement throughout the library.

https://gist.github.com/canyon289/cbd3744a36d8198c6d4ac796eae60e4b

[ahartikainen]

Could we have suggestion where to go to see correct coord/dim format?

For new users these things can be complicated.

Our docs + xarray docs?

[canyon289]

I agree it is complicated! There's two separate issues, one is the format, but the other is just trying to select something that doesn't exist.

Maybe we should have three exceptions?

1. Format is invalid
2. Format is valid but coord name doesn't exist
3. Format is valid but dim doesn't exist

[ahartikainen]

That does sound reasonable.

I think if we add any links, they should be on their own lines (or something users can click):

See.
https ://linktoxarrayprops.xyz/...

[canyon289]

Turns out there's four paths to three types of exceptions all shown in the gist.

I took the strategy of letting xarray raise the exception before handling it for two reasons

1. Xarray's exceptions have useful information as to what might be wrong so its worth capturing
2. I don't want arviz to raise exceptions in cases when xarray would not. If arviz handles the exception first we can't be certain of this.

https://gist.github.com/canyon289/51281affde8e8a8891d503e45dea0920

In regards to @ahartikainen's comment, I do agree we need an "xarray 101" type documentation. For this PR though would you guys be ok with merging this (once I update all plots and it looks ok) and I'll follow up with the documentation, and link in exception, in another PR?

[ColCarroll]

Looks good to me - is this all the places dataset.sel(**coords) happens? Even if it isn't, feel free to merge and we can followup with the rest.

[ahartikainen]

LGTM

[canyon289]

Updated to all the places with **sel. I'll merge if tests pasts. Thank you