numpy elementwise comparison with overloaded operator

Question

I have two numpy arrays containing objects with an overloaded comparison operator that returns another object, instead of True or False. How can I create an array containing the results of the individual comparisons. I would like result to be an array of objects like in the following

lhs = ... # np.array of objects with __le__ overloaded rhs = ... # another np.array result = np.array([l <= r for l, r in izip(lhs, rhs)])

but lhs <= rhs gives me an array of bools. Is there a way to get to the result to be the array of the results of the __le__ method calls without writing a python loop?

The documentation for np.less_equal (and the other comparison functions) says it returns "the truth value" of the comparison, so it may not be possible to do this without manually iterating over the array. — BrenBarn
– BrenBarn, Commented Jan 10, 2015 at 21:17

ryanpattison · Accepted Answer · 2015-01-11 01:06:30Z

Numpy's Github page on ndarray states that the comparison operators are equivalent to the ufunc forms in Numpy. Therefore lhs <= rhs is equivalent to np.less_equal(lhs, rhs)

From the output of np.info(np.less_equal)

 Returns ------- out : bool or ndarray of bool Array of bools, or a single bool if `x1` and `x2` are scalars.

To get around this you can use:

import operator result = np.vectorize(operator.le)(lhs, rhs)

The np.vectorize will allow you to use Numpy's broadcasting in the comparisons too. It will use your objects comparison and will return a numpy array of the results of those comparisons in the same form as your list comprehension.

The __le__ function that we're not allowed to see is operating in a plain old Python list comprehension not a numpy context. Therefore, np.less_equal is not called.
@msw lhs <= rhs is what the OP wanted to use, which calls np.less_than, I'm stating why it returns bool instead of the results of the comparisons as he expected.

Collectives™ on Stack Overflow

numpy elementwise comparison with overloaded operator

1 Answer 1

2 Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Related