to_dataset lossy

[tdimiduk]

I am looking at saving a DataArray to netcdf. This appears to require turning into a dataset. However, the conversion to and from datasets appears to be lossy.

d = DataArray(...)
n = d.to_dataset().to_array()

looses attrs and the name of the DataArray gets moved to a "variable" coordinate.

I can manually add the attrs to the created dataset, but it seems like they should either be transfered by default, or have an option to preserve them. Is there a reason this is not done?

[fmaussion]

If you don't use to_array() but simply acces the variable everything is fine:

import xarray as xr
da = xr.DataArray([1, 2, 3], name='da', attrs={'myattr':3})
da.to_dataset().da
<xarray.DataArray 'da' (dim_0: 3)>
array([1, 2, 3])
Coordinates:
  * dim_0    (dim_0) int64 0 1 2
Attributes:
    myattr: 3

[tdimiduk]

Ah thanks. Would you be be interested in a pull request to make that more clear in the docs?

[shoyer]

Sure, improvements to the docs are always welcome!

Note that in the next release of xarray we will have a DataArray.to_netcdf method (and open_dataarray).

[tdimiduk]

Oh, that will be excellent. When is that coming out?

On Fri, Oct 14, 2016 at 4:12 PM, Stephan Hoyer notifications@github.com
wrote:

Sure, improvements to the docs are always welcome!

Note that in the next release of xarray we will have a DataArray.to_netcdf
method (and open_dataarray).

—
You are receiving this because you modified the open/close state.
Reply to this email directly, view it on GitHub
#1047 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AArz9M4l7AKamroG_E8swXNbZoxxxzJIks5qz-IigaJpZM4KXT2N
.

[shoyer]

I'm trying to get a couple of big PRs (#1017 and #1028) into the next major release. So probably 2-4 weeks.