Add to_pandas method for Dataset

[gcaria]

• Closes Add Dataset.to_pandas() method #255
• Tests added
• Passes pre-commit run --all-files
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[max-sixty]

Seems very reasonable to me; others' thoughts?

[shoyer]

The model we have with DataArray.to_pandas() is that it converts into a corresponding pandas object without changing the number of dimensions. to_dataframe()/to_series() will flatten multiple dimensions into a MultiIndex, but that isn't the role of to_pandas().

Thus:

• 0D xarray.DataArray -> 0D NumPy scalar
• 1D xarray.DataArray -> 1D pandas.Series
• 2D xarray.DataArray -> 2D pandas.DataFrame
• 3D or higher xarray.DataArray -> error

If we treat a Dataset like a DataArray with one extra dimension (corresponding to variables), then it would make sense to have:

• 0D xarray.Dataset -> 1D pandas.Series
• 1D xarray.Dataset -> 2D pandas.DataFrame (same as to_dataframe())
• 2D or higher xarray.Dataset -> error

I guess this is basically what you have here, except for raising an error in the final case (the error message should mention to_dataframe().

[shoyer]

Looks great, thank you!

[mathause]

Thanks @gcaria!