Describe the enhancement requested
Converting an ndarray to the table structure often can be achieved using np.tile() and np.repeat() on the ndarray labels.
If the data type in numpy is not zero copy to pyarrow, casting the expanded columns to pyarrow is expensive. It'd be more efficient to cast first, then use pa.array().repeat() and pa.array().tile().
Practically the low level building blocks of pa.Table.from_xarray(xr.Dataset).
Component(s)
Python, C++