[FIXED] Find the minimum value for each unique key without using a for loop

Issue

I have a numpy array with keys (e.g. [1, 2, 2, 3, 3, 2]) and an array with values (e.g. [0.2, 0.6, 0.8, 0.4, 0.9, 0.3]). I want to find the minimum value associated with each unique key without using a for loop. In this example, the answer is {1: 0.2, 2: 0.3, 3: 0.4}. I asked ChatGPT and New Bing but they keep giving me the wrong answer. So, is it really possible to do this without a for loop?

Edit 1: What I'm trying to achieve is the fastest speed. Also, in my case, most keys are unique. I considered using np.unique to acquire every key and then compute the min value for every key, but clearly it requires a for loop and a quadratic time. I also considered sorting the arrays by keys and apply np.min on the values of each key, but I doubt its efficiency when most keys are unique. Additionally, according to the comments, pandas.DataFrame has a groupby method which might be helpful, but I'm not sure if it's the fastest (perhaps I'm going to try on my own).

Edit 2: I don't necessarily need a dict as the output; it can be an array of unique keys and an array of min values, and the order of keys doesn't matter.

Solution

The naive, python solution is something like:

result = {}

for key, value in zip(keys, values):
    current = result.get(key)
    if current is not None:
        result[key] = min(current, value)
    else:
        result[key] = value

It should be relatively fast.

If you really need to squeeze performance out of this, you should use numba

import numba

@numba.jit(nopython=True)
def group_min(keys, values):
    result = {}

    for key, value in zip(keys, values):
        current = result.get(key)
        if current is not None:
            result[key] = min(current, value)
        else:
            result[key] = value

    return result

Make sure to read over the numba docs to understand how to squeeze as much performance as you can out of it.

Answered By - juanpa.arrivillaga