issue_comments

See #1236 for a proposed fix. After merging it, I will release v0.9.1 and issue the delayed release announcement for xarray v0.9.

Nice diagram! Do you think it's worth to also add Variable and IndexVariable (along with inheritance)?

Yes, but probably only for a separate "internal/advanced API" diagram. I want this to focus on the user facing and public API.

Although this wouldn't fully solve the problem, maybe an html repr for the notebook would help here?

Indeed, I think this is quite possible -- we could squeeze a lot more information into an HTML repr. For example, with a little bit of JavaScript (or maybe CSS these days?) we could highlight all appearances of a dimension name when you hover over it. Maybe we can find someone with design talent/interest to work on this?

Perhaps more broadly documentation-wise, it might be good to add a terminology list. For example, that could clarify the difference and relation between dimensions, labels, indices, coordinates, etc.. There are dimensions without coordinates, dimensions that are labelled or unlabelled, there are coordinates that are indices, coordinates that are not indices. I'm still figuring out how all of those relate to each other and how I use them.

Agreed, this is definitely worth doing. I also made a diagram recently to summarize the xarray data model that I'd like to put in the docs:

(feedback on the diagram is very welcome if anyone has suggestions!)

In the SO post, the OP also wondered about the "empty" coordinates. Any plan to change this too? Maybe a - instead of empty?

I think I want to remove the appearance of *empty* for coordinates, and maybe for data variables, too. The intent was to remove confusion about missing fields in the repr, to avoid something like <xarray.Dataset> Dimensions: () for xr.Dataset().

It gives a subtle signal that the user is doing something wrong, but that isn't necessarily the case.

I'd add another suggestion to the list that @shoyer proposed, which is simply to do nothing with the unindexed dimensions

Yes, this is what I did originally, and I think it's a very elegant solution.

Unfortunately, for large Dataset objects, there were valid complaints it can be easy to miss dimensions without coordinates. There's some threshold, probably around five or so items, at which point it becomes hard to see at a glance what's missing from a list.

I don't see it working well to make repr change suddenly when a certain threshold number of items is reached, but maybe there's another way (a different method? an option?) we can use for exposing this information to the power-users who need to see it without making it confusing for newcomers.

Any use case for dimensions which don't have an index (i.e., an IndexVariable) but have coordinates? This would be confusing too:

I suppose we do something even more explicit, e.g., "Dimensions without a corresponding coordinate". That feels too long to me, though.

I don't like "Dimensions without index variable" because it emphasizes that they don't have an index rather than that they don't have a variable.

For now I think "Dimensions without coordinates" is my favorite.

OK, let's go back to the drawing board.

Let ds = xr.Dataset({'foo': (('x', 'y'), [[1, 2], [3, 4]])}, {'x': [1, 2]}).

Current repr (v0.9.0): <xarray.Dataset> Dimensions: (x: 2, y: 2) Coordinates: * x (x) int64 1 2 Unindexed dimensions: y Data variables: foo (x, y) int64 1 2 3 4

Some alternatives:

• Rename "Unindexed dimensions" to "Dimensions without coordinates". This is clearer but pretty verbose: ``` <xarray.Dataset> Dimensions: (x: 2, y: 2) Coordinates:

• x (x) int64 1 2 Dimensions without coordinates: y Data variables: foo (x, y) int64 1 2 3 4 ```

• Same as above, but with everything on one line (less obtrusive): ``` <xarray.Dataset> Dimensions: (x: 2, y: 2) Coordinates:

• x (x) int64 1 2 Data variables: foo (x, y) int64 1 2 3 4 Dimensions without coordinates: y ```

• Mark Dimensions with an index with * before there name. @crusaderky suggested this previously but I rejected it as ugly: ``` <xarray.Dataset> Dimensions: (*x: 2, y: 2) Coordinates:

• x (x) int64 1 2 Data variables: foo (x, y) int64 1 2 3 4 ```

• Use another marker character in Dimensions, maybe ' after dimension names for dimensions with indexes? ``` <xarray.Dataset> Dimensions: (x': 2, y: 2) Coordinates:

• x (x) int64 1 2 Data variables: foo (x, y) int64 1 2 3 4 ```

We may not have gotten this right yet. See StackOverflow: What are “unindexed dimensions” and why are coordinates empty?

See #1221.

OK, I'm going to put together a PR to change this behavior to add the Unindexed dimensions section instead of putting these dimensions under Coordinates.

This feels a little overly verbose (I still feel like there may be some sort of solution with symbols here), but it's explicit and fully self-explanatory, which is a big advantage of over what we currently have.

Sorry, I'm not sure to understand: do you mean you'd rather stay with the current behavior as you said in #1017 (comment) ?

I like the currently implemented behavior, which is the second bullet in https://github.com/pydata/xarray/pull/1017#issuecomment-249908900. We could potentially switch to the behavior of the first bullet (issue TypeError instead), but I think this would be less convenient.

I actually think that .sel() shoud not work on y. I understand that this would break a bunch of existing code, but this is a confusing (if not inconsistent) behavior.

See https://github.com/pydata/xarray/pull/1017#issuecomment-249908900 and discussion below for comments on this. This does make .sel a little more complex, but I think it's pretty reasonable given that the fall-back behavior is well defined and not value dependent like pandas's soon to be deprecated .ix.

I wonder if instead this should be taken as an indication that we need to make the repr more self explanatory, e.g., by writing something like *missing* instead of -. This would be consistent with how we use *empty* in the Dataset repr for empty coords or data_vars.

Current version: <xarray.Dataset> Dimensions: (dim_0: 3) Coordinates: o dim_0 (dim_0) - Data variables: var (dim_0) int64 1 2 3 Alternative A: write *missing* instead of -: <xarray.Dataset> Dimensions: (dim_0: 3) Coordinates: o dim_0 (dim_0) *missing* Data variables: var (dim_0) int64 1 2 3 Alternative B: write *missing* instead of -, use * for the marker instead of o: <xarray.Dataset> Dimensions: (dim_0: 3) Coordinates: * dim_0 (dim_0) *missing* Data variables: var (dim_0) int64 1 2 3

I think I like Alternative B best. @MaximilianR @crusaderky @benbovy any opinions?

Agreed, we should definitely document this and highlight the change in the release notes. I'm still not super happy with using o in the new repr, but it is the best we came up with so far.