`frompyfunc` requires `dtype=object` when used with `accumulate`

[hpaulj]

np.frompyfunc(lambda x,y:x+y,2,1).accumulate(np.arange(10))

raises an error:

ValueError: could not find a matching type for <lambda> (vectorized).accumulate, requested type has type code 'l'

accumulate only works if the dtype is set explicitly to object (either in the input, the out array, or as a kwarg).

np.frompyfunc(lambda x,y:x+y,2,1).accumulate(np.arange(10,dtype=object))

This behavior may be consistent with the note in its doc that The returned ufunc always returns PyObject arrays.. Also when vectorize uses a ufunc generated by this function, it converts the args to object type, and then converts the output to a specified (or deduced) otype.

However, is this error message required? A straight call of the ufunc returns an object array without any warning or error message:

np.frompyfunc(lambda x:2*x,1,1)(np.arange(4))
# array([0, 2, 4, 6], dtype=object)

Or is this just documentation issue? Does its __doc__ need an added note?

[jakirkham]

Probably does some casting under the hood (speculating as I have not looked). Maybe that is an option for accumulate too?

[WarrenWeckesser]

I don't know when it was fixed, but in the current development branch, the example does not generate an exception:

In [1]: import numpy as np

In [2]: np.__version__
Out[2]: '1.22.0.dev0+1696.g5cc7ef066'

In [3]: np.frompyfunc(lambda x,y:x+y,2,1).accumulate(np.arange(10))
Out[3]: array([0, 1, 3, 6, 10, 15, 21, 28, 36, 45], dtype=object)