issue_comments

Thanks @dcherian ! This is a nice feature

@fmaussion any further concerns here? This looks good to me.

Done and rebased on master.

Thanks! In view of @shoyer ' comment we could revisit the API for datasets at a later stage.

Updated and rebased. Now these are the only options.

Yes, this is not valid syntax on Python 2.

On Wed, Feb 21, 2018 at 9:53 AM Deepak Cherian notifications@github.com wrote:

@dcherian commented on this pull request.

In xarray/plot/plot.py https://github.com/pydata/xarray/pull/1926#discussion_r169723308:

 add_legend = kwargs.pop('add_legend', True)

 ax = get_axis(figsize, size, aspect, ax)

• error_msg = 'must be either None or one of %r' % list(darray.dims) +
• if x not in [None, *darray.dims]:

I don't know what's wrong here. It runs locally; the tests pass and flake8 goes through plot.py without errors. Is it a python2 vs python3 thing?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/pydata/xarray/pull/1926#discussion_r169723308, or mute the thread https://github.com/notifications/unsubscribe-auth/ABKS1pXVXGKhgmkR8yP3epLPvqZCXNipks5tXFgpgaJpZM4SLdeS .

either by (i) specifying the name of the variable to the axis you want to plot it onto, or (ii) by specifying the name of the coordinate you want to use for the axis. Since (ii) is the default and only option for 1D and 2D plots until now, I just wondered if (i) is very necessary.

It does feel a little strange that if you had a DataArray with non-dimension coordinates you could make a plot without including any of the DataArray values

Good points. I'll change it so that only one of x and y can be specified.

My main thought is that this API would feel much more natural on a Dataset object, alongside a .plot.scatter() method.

It's very useful in oceanography (and meteorology) where we want to make plots of physical quantities like temperature against depth; it is much more natural to see depth on the y-axis than the x-axis. This is the original usecase @rabernat mention in #575.

The Dataset API is a more involved undertaking that I'm not confident of executing at this point.

My main thought is that this API would feel much more natural on a Dataset object, alongside a .plot.scatter() method.

That said, I suppose this could still be useful and I don't think it's harmful to expand the API here. It does feel a little strange that if you had a DataArray with non-dimension coordinates you could make a plot without including any of the DataArray values, e.g., xr.DataArray(..., dims=['x'], coords={'x': ..., 'y': ('x', ...)}, name='f').plot.line(x='x', y='y').

Do you mean da.plot(y='time', x=None, hue='lat')

Yes, I meant that there should be one and only way to specify which axis should do what: my understanding of your current implementation is that there are two ways to reach the same goal: either by (i) specifying the name of the variable to the axis you want to plot it onto, or (ii) by specifying the name of the coordinate you want to use for the axis. Since (ii) is the default and only option for 1D and 2D plots until now, I just wondered if (i) is very necessary. (or maybe I missed something)

@fmaussion when I added the hue argument, @shoyer suggested that we have a "fully explicit way to make these plots". In this case this would mean da.plot(x='temp', y='time', hue='lat') which seems a lot more readable than da.plot(y='coordinate', hue='lat')

It would also mean that x,y are different from hue (for hue we do specify dimension name).

@shoyer Thoughts?

EDIT:

@fmaussion Do you mean da.plot(y='time', x=None, hue='lat')?