[FIXED] Fastest way to generate shuffled range array form zero to x

Issue

I'm trying to find the fastest way to generate a shuffled array of indices. A shuffled range from 0 to a number, so to say.

Currently I've compared these two methods:

setup = "import numpy as np; idxlen = 1000; rng = np.random.default_rng(); matrix = -np.ones([100,idxlen],dtype=int)"
test1 = "rand_idxs = np.arange(idxlen); 
         rng.shuffle(rand_idxs); 
         matrix[0][:] = rand_idxs"
test2 = "matrix[0][:] = rng.choice(range(idxlen),idxlen,replace=False)"
print(timeit.timeit(test1, setup, number = 10000))
print(timeit.timeit(test2, setup, number = 10000))

Returning the following timings:

0.1846354000736028
0.950088300043717

I assume that under the hood, the rng.shuffle() method randomizes the indices of the array past to it, and rearranges it according to those indices. Now, since I'm generating an array containing every integer from 0 to a certain number, this practically corresponds to generating randomized indices. So I can't shake the feeling that there is a faster way to do what I'm trying to achieve. Is there?

Solution

If you remove range from the test two then you will see changes and find test2 is faster then first

import timeit
setup = "import numpy as np; idxlen = 1000; rng = np.random.default_rng(); matrix = -np.ones([100,idxlen],dtype=int)"
test1 = "rand_idxs = np.arange(idxlen); rng.shuffle(rand_idxs); matrix[0][:] = rand_idxs"
test2 = "matrix[0][:] = rng.choice(idxlen,idxlen,replace=False)"
print(timeit.timeit(test1, setup, number = 10000))
print(timeit.timeit(test2, setup, number = 10000))

Now for testing

for idxlen=1000

0.15045420000024023
0.24143790000016452

for idxlen=10000

1.4334205999984988
1.3395810999973037

for idxlen=100000

14.489788099999714
8.599601199999597

Also there is one more method that is np.random.Generator.permutation

And the testing for this method are these

test3 = "matrix[0][:] = rng.permutation(idxlen)"

for idxlen=1000

0.1567758999990474

for idxlen=10000

1.4174363000020094

for idxlen=100000

13.964398899999651

---

In the end the comparison of timing between all three

Graph for number=10000

Code for graph

Answered By - codester_09