Question

I would like to use numpy.frompyfunc to generate an unbound ufunc from an unbound member method. My concrete but failing attempts look like

class Foo(object):
    def f(x,y,z):
        return x + y + z

Foo.g = numpy.frompyfunc(Foo.f,3,1)
i = Foo()
i.g(5,6,7)

where the last line fails with
“TypeError: unbound method f() must be called with Foo instance as first argument (got int instance instead)”. That error makes sense to me as Foo.f is unbound.

Despite it being deprecated, I thought I’d give new.instancemethod a shot:

import new
Foo.h = new.instancemethod(numpy.frompyfunc(Foo.f,3,1),None,Foo)
j = Foo()
j.h(5,6,7)

where the last line now fails with “TypeError: return arrays must be of ArrayType” which I don’t understand.

My goal is to monkey patch Foo with ufunc-ready versions of its members. They must be member methods as they depend on a Foo instance’s state (though that dependency is not shown here).

Any suggestions?

Answer 1

import numpy as np

class Foo(object):
    def f(self,x,y,z):
        return x + y + z

Add a g method to Foo:

def g(self,x,y,z):
    return np.frompyfunc(self.f,3,1)(x,y,z)
Foo.g = g

Calling g from an instance of Foo:

i = Foo()
print(i.g(np.array([5,6]),np.array([6,7]),np.array([7,8])))
# [18 21]