[SciPy-User] How to numpy.vectorize functions with keyword arguments?

[nicky van foreest]

Hi,

Some time ago I built the code below, but now I take a look at it
again it seems to be the reverse of what you want. I first give the
arguments, and then vectorize, whereas you want a to give arguments to
an already vectorized function. Sorry for the confusion. Nevertheless,
here is what I built:

  delta = 0.01
  grid = arange(np.finfo(float).eps,15,delta)


    def G(self, i, k):
        def GG(i, k, x):
            if x < 10.:
                return 0.
            else:
                return 1.
        cdf = np.vectorize(functools.partial(GG, i, k))
        return cdf(self.grid)

I needed the cumulative distribution function on a grid, for specific
values of i and k. The code above shows a trivial example (in which I
actually don't need the i and k.)

hope this helps somewhat.

Nicky


On 22 June 2011 10:41, Christoph Deil <Deil.Christoph at googlemail.com> wrote:
>
> On Jun 21, 2011, at 3:28 PM, nicky van foreest wrote:
>
> Question 2: numpy.vectorized functions don't like being called with keyword
>
> arguments, the first line in __main__ raises a TypeError.
>
> Why does this happen? What is the standard method to make vectorized
>
> functions callable with keyword arguments?
>
> You might try partial functions, see functools.partial in the functools
> module.
>
> Nicky
>
> I tried using functools.partial but couldn't make it work.
> Can you give an example of how to use it to make a vectorized function
> accept keyword arguments?
> E.g. the following lines still give the same TypeError as before:
> a = np.array([10, 100, 1000])
> cube_find_x = ft.partial(find_x, scale='cube')
> print cube_find_x(a, cost)
> Also this is not really what I want, which is to make find_x accept keyword
> arguments without writing the wrapper myself,
> maybe there is a way to write a decorator handle_kwargs so that I can simply
> write something like the following code?
> @handle_kwargs
> @np.vectorize
> def find_x(...)
>     ....
>
>
>
> I found that writing a wrapper (wrapped_find_x) works, but I'd rather not
>
> litter my code with many such wrapper functions.
>
> In the example below it would be ok just using positional arguments, but I
>
> have many functions, each with ~10 keyword arguments.
>
> Christoph
>
>
> import numpy as np
>
> @np.vectorize
>
> def cost(x, scale='square'):
>
>     """Some complicated function that is supplied by the user"""
>
>     if scale == 'square':
>
>         return x ** 2
>
>     elif scale == 'cube':
>
>         return x ** 3
>
>     else:
>
>         return 0
>
> @np.vectorize
>
> def find_x(a, f, scale='square', maxiter=100):
>
>     """Uses an iterative algorithm to determine a result"""
>
>     x = 1
>
>     # just to avoid possibly infinite loop, maxiter should never be reached
>
>     for _ in range(maxiter):
>
>         if f(x, scale) > a:
>
>             break
>
>         x *= 2
>
>     return x
>
> def wrapped_find_x(a, f, scale='square', maxiter=100):
>
>     return find_x(a, f, scale, maxiter)
>
> if __name__ == '__main__':
>
>     print find_x(np.array([10, 100, 1000]), cost, scale='cube') # TypeError
>
>     print wrapped_find_x(np.array([10, 100, 1000]), cost, scale='cube') # OK
>
> _______________________________________________
>
> SciPy-User mailing list
>
> SciPy-User at scipy.org
>
> http://mail.scipy.org/mailman/listinfo/scipy-user
>
>
> _______________________________________________
> SciPy-User mailing list
> SciPy-User at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>
>
> _______________________________________________
> SciPy-User mailing list
> SciPy-User at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>
>