cumsum providing correct behavior for non-coordinate DataArrays?

[pwolfram]

In the case of a DataArray without coordinates, should cumsum work without specifying an axis, e.g., da.cumsum() is valid? This is not currently the obtained behavior.

In [1]: import xarray as xr
imp
In [2]: import numpy as np

In [3]: da = xr.DataArray(np.arange(10))

In [4]: da
Out[4]: 
<xarray.DataArray (dim_0: 10)>
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
Dimensions without coordinates: dim_0

In [5]: da.cumsum(axis=0)
Out[5]: 
<xarray.DataArray (dim_0: 10)>
array([ 0,  1,  3,  6, 10, 15, 21, 28, 36, 45])
Dimensions without coordinates: dim_0

In [6]: da.cumsum()
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-5-4ff0efc782ee> in <module>()
----> 1 da.cumsum()

/Users/pwolfram/src/xarray/xarray/core/common.pyc in wrapped_func(self, dim, axis, skipna, keep_attrs, **kwargs)
     17                              keep_attrs=False, **kwargs):
     18                 return self.reduce(func, dim, axis, keep_attrs=keep_attrs,
---> 19                                    skipna=skipna, allow_lazy=True, **kwargs)
     20         else:
     21             def wrapped_func(self, dim=None, axis=None, keep_attrs=False,

/Users/pwolfram/src/xarray/xarray/core/dataarray.pyc in reduce(self, func, dim, axis, keep_attrs, **kwargs)
   1146             summarized data and the indicated dimension(s) removed.
   1147         """
-> 1148         var = self.variable.reduce(func, dim, axis, keep_attrs, **kwargs)
   1149         return self._replace_maybe_drop_dims(var)
   1150 

/Users/pwolfram/src/xarray/xarray/core/variable.pyc in reduce(self, func, dim, axis, keep_attrs, allow_lazy, **kwargs)
    898             if dim is None and axis is None:
    899                 raise ValueError("must supply either single 'dim' or 'axis' "
--> 900                                  "argument to %s" % (func.__name__))
    901 
    902         if dim is not None:

ValueError: must supply either single 'dim' or 'axis' argument to cumsum

[fmaussion]

I'm not sure that whether coordinates should play a role here, but rather the dimensionality of the data (i.e. maybe xarray could infer that the data has one dimension and work on that axis?).

I personally find the numpy behavior with ndarrays (flattening the array before applying cumsum) quite dangerous, I would rather like it to raise an error...

a = np.arange(12).reshape((4, 3))
np.cumsum(a)
array([ 0,  1,  3,  6, 10, 15, 21, 28, 36, 45, 55, 66])

[pwolfram]

Thanks @fmaussion, it just seems strange that if the data is one-dimensional we return an error. I would agree that we probably want an error for dimensionality larger than one. I think the thing to change here is to make da.cumsum() work for one-dimensional da. But, it may not be worth the effort to fix this issue relative to other issues.

[shoyer]

Yes, it would be reasonable to not require specifying the dimension for one-dimensional arguments.

I recall this coming up before for other methods, but can't remember the particulars now.

[shoyer]

Someone also wanted a default dimension for diff: #1131

[stale]

In order to maintain a list of currently relevant issues, we mark issues as stale after a period of inactivity

If this issue remains relevant, please comment here or remove the stale label; otherwise it will be marked as closed automatically