Mixing of dask.array and numpy.ndarray in Numba @guvectorize #6855
Description
It appears that dask.array and numpy.ndarray cannot be mixed when using Numba's @guvectorize decorator - see the minimal example below. It would be a nice feature if dask could resolve this (or at least raise a more helpful exception!). Regardless, this is a truly helpful package!
Minimal Example
import dask.array as da
import numpy as np
import numba as nb
# large array
x = np.random.random((300000, 100, 10))
y = da.from_array(x) # for the sake of the example
# small vector
v = np.ones(10)
w = da.from_array(v)
@nb.guvectorize([nb.void(nb.f8[:, :], nb.f8[:], nb.f8[:])],
'(m, n), (n) -> (m)', target='parallel')
def test(arr, vec, out):
# really just a test function
out[:] = (arr + vec).sum(axis=1)
out1 = test(x, v) # works
out2 = test(y, w).compute() # works
out3 = test(y, v) # fails with ValueErrorThe following ValueError is raised ... (on dask version 2.30.0)
[...]
ValueError: `dtype` inference failed in `apply_gufunc`.
Please specify the dtype explicitly using the `output_dtypes` kwarg.
Original error is below:
------------------------
ValueError('test: Input operand 1 has a mismatch in its core dimension 0, with gufunc signature (m, n), (n) -> (m) (size 10 is different from 1)',)
Traceback:
---------
File "<...>/miniconda3-latest/envs/<...>/lib/python3.6/site-packages/dask/array/core.py", line 377, in apply_infer_dtype
o = func(*args, **kwargs)