Replacing selected elements in a list in Python

Question

I have a list: mylist = [0, 0, 0, 0, 0]

I only want to replace selected elements, say the first, second, and fourth by a common number, A = 100.

One way to do this:

mylist[:2] = [A]*2
mylist[3] = A
mylist
[100, 100, 0, 100, 0]

I am looking for a one-liner, or an easier method to do this. A more general and flexible answer is preferable.

Solution 1

Especially since you're replacing a sizable chunk of the list, I'd do this immutably:

mylist = [100 if i in (0, 1, 3) else e for i, e in enumerate(mylist)]

It's intentional in Python that making a new list is a one-liner, while mutating a list requires an explicit loop. Usually, if you don't know which one you want, you want the new list. (In some cases it's slower or more complicated, or you've got some other code that has a reference to the same list and needs to see it mutated, or whatever, which is why that's "usually" rather than "always".)

If you want to do this more than once, I'd wrap it up in a function, as Volatility suggests:

def elements_replaced(lst, new_element, indices):
    return [new_element if i in indices else e for i, e in enumerate(lst)]

I personally would probably make it a generator so it yields an iteration instead of returning a list, even if I'm never going to need that, just because I'm stupid that way. But if you actually do need it:

myiter = (100 if i in (0, 1, 3) else e for i, e in enumerate(mylist))

Or:

def elements_replaced(lst, new_element, indices):
    for i, e in enumerate(lst):
        if i in indices:
            yield new_element
        else:
            yield e

Solution 2

def replace_element(lst, new_element, indices):
    for i in indices:
        lst[i] = new_element
    return lst

It's definitely a more general solution, not a one-liner though. For example, in your case, you would call:

mylist = replace_element(mylist, 100, [0, 1, 3])

Solution 3

Numpy supports this if you're not opposed to using an np.ndarray:

>>> a = np.zeros(5)
>>> a[[0,1,3]] = 100
>>> a
array([ 100.,  100.,    0.,  100.,    0.])

Comments

[elwc]

Good explanation regarding the new list and mutation.

[abarnert]

+1. On top of being more general, I think this is the most readable way to do the mutable replacement, despite (or, rather, probably because of) not being a one-liner.

[Abhijit]

Sometimes I do miss the valarray::gslice

[abarnert]

@Abhijit: Well, nearly every case where valarray makes sense in C++, numpy makes sense in Python—and numpy's extended slicing is even simpler than gslice for almost all cases. For example, see mgilson's answer to this question. (Of course "nearly every" and "almost all" aren't quite the same as "every" and "all"… but it's rare that I find myself writing some numpy code and missing a C++ feature…)

[abarnert]

The main problem with this is that in Python 3, it doesn't work, because you just get back an iterator that you never iterate. The lesser problem is that in Python 2, it builds a list you have no need for. You can solve the first problem by explicitly doing list(map(…)). Or you can solve both by using six.map and passing the result to an iter_discard function, which you can implement as collections.deque(it, max_size=0). Which is all a whole lot more thought than you should ever put into a solution you weren't a huge fan of in the first place, of course…

[abarnert]

+1. Even if this doesn't directly answer the OP's question (and I'm not sure it doesn't), whenever you find yourself dealing with sequences of numbers and something seems harder than it should be, the first question you should ask is probably "am I opposed to using an np.ndarray here?"

[RocketDonkey]

@abarnert Belated +1 to your answer (informative as always), and great points regarding this implementation (never thought about passing the results of map to an iter_discard function). I'll add in an update for posterity's sake, and thanks as always for the helpful info :)

[abarnert]

I really wish iter_discard were part of the 3.x standard library. The argument against it was that this would want to encourage people using map for side effects instead of results (and, if you really want it, the deque trick has been added to the documentation for itertools somewhere). I can buy that… but for these kinds of questions, people always do that anyway (and then want to timeit the results with every version of Python they can find).

[RocketDonkey]

@abarnert Right you are, looks like the consume recipe. That is interesting - I have always felt like I was doing something not-quite-right when using map for the side effects you mention (the unnecessary list byproduct being the strongest indicator). That deque technique definitely alleviates that feeling, so thanks for the new trick to add to the rotation :)