How do I create a numpy array using a function?

Question

I am using np.fromfunction to create an array of a specific sized based on a function. It looks like this:

import numpy as np
test = [[1,0],[0,2]]
f = lambda i, j: sum(test[i])
matrix = np.fromfunction(f, (len(test), len(test)), dtype=int)

However, I receive the following error:

TypeError: only integer arrays with one element can be converted to an index

Just for reference: [Looking at numpy source](https://github.com/numpy/numpy/blob/v1.7.0/numpy/core/numeric.py#L1611) `fromfunction` just creates `indices` array and passes it to user function. So `i` and `j` are `array`s, not `int`s. — mg007, May 28 '13 at 20:15
You can see how it performs that task in this similar question: http://stackoverflow.com/questions/400739/what-does-mean-in-python — sdasdadas, May 28 '13 at 20:23

score 16 · Accepted Answer · answered May 28 '13 at 20:03

16

The function needs to handle numpy arrays. An easy way to get this working is:

import numpy as np
test = [[1,0],[0,2]]
f = lambda i, j: sum(test[i])
matrix = np.fromfunction(np.vectorize(f), (len(test), len(test)), dtype=int)

np.vectorize returns a vectorized version of f, which will handle the arrays correctly.

answered May 28 '13 at 20:03

Florian Rhiem

3

passing the function through `np.vectorize` is _the_ trick documentation fails to point out, thank you so much – blue May 11 '16 at 14:05
1

Excellent trick! It ought to be in the docs. I had similar problem with trivial scenario `f = lambda i, j: 3.14` Running `np.fromfunction(f, (2, 6))` would simply give me a scalar 3.14 ! By contrast, `np.fromfunction(np.vectorize(f), (2, 6))` returns the expected 2x6 matrix whose entries are all 3.14 – Julian - BrainAnnex.org Jan 04 '19 at 04:44

1 Answers1