issue_comments

Yes, @shoyer is right : Looking at the numpy code, this array.squeeze(axis) -> array.squeeze() is happening. 1256 try: 1257 # First try to use the new axis= parameter 1258 return squeeze(axis=axis) 1259 except TypeError: 1260 # For backwards compatibility 1261 return squeeze()

I added the axis parameter in xarray and this fixed the problem.

``` PYTHONPATH=$HOME/sandbox/dev/xarray python -c 'import numpy as np ; import xarray as xr ; test = np.random.random((1,10,1)) ; print(np.squeeze(test, axis=0).shape); print(np.squeeze(xr.DataArray(test), axis=0).shape, np.version, xr.version) ' (10, 1) (10, 1) 1.13.1 0.10.0+dev64.gd554ece

```

I was under the impression that xarray.DataArrays behave the same as numpy.ndarrays when applying numpy functions, so that I didn't have to change any of my code.

We do attempt to make this work when feasible, but unfortunately it isn't always the case. In this case I do think this is doable, so I'm marking this issue as "Contributions welcome."

Yes, I did realize there are workarounds and DataArray.squeeze() is the preferred one.

In my mind this is a kind of incompatibility problem and it would be nice to fix this. I had existing numpy code, and later decided to add dimension and coordinate labels using xarray. I was under the impression that xarray.DataArrays behave the same as numpy.ndarrays when applying numpy functions, so that I didn't have to change any of my code.

At least for np.squeeze this is not the case. Perhaps there are other functions for which a similar issue arises. If you don't think this kind of compatibility is desired then it would be good to address this in the documentation.

This is (somewhat) expected. Basically, you should use the DataArray.squeeze() method instead.

The long story is that some but not all NumPy functions define themselves by (attempting) to call methods, e.g., np.squeeze(array) -> array.squeeze().

squeeze is a special and somewhat surprising case. Apparently np.squeeze(array, axis) -> array.squeeze(axis) -> array.squeeze() if the axis argument isn't handled. But we didn't bother to define an axis argument for DataArray.squeeze(), so this results in an error.

I suppose we could fix this by defining an axis argument, like we do for some methods like .mean(). But the short answer is that using xarray methods should be preferred.