Thank you for the video! I love quick little algorithm comparison showcases. A quick note: on my machine, my own version of the trial division algorithm (up to 1 million) took 12 seconds to run while printing the numbers and just over one second without printing the numbers. This is because system calls (which are required to print to the terminal) are around a thousand times (or more!) slower to run than typical math operations, depending on the language and environment. This also explains why the sieve and trial division algorithms took about the same time for the first million primes, they both spent most of the time printing instead of actually doing math. It's worth keeping this in mind for future comparisons. Again, thank you for the video; I would love to see more naive vs. advanced algorithm implementation comparisons!
Hi! Really glad to hear that you enjoyed the video😊 I’ve also tried the algorithm without printing and realised they were quicker, however I felt like it would be a little more visually interesting if the prime numbers were printed out in the video 😅
Hello, mathematician here. Your method is awesome, but you only need to check for divisors up to the floor of sqrt(p), what i mean is, if you wanna check if one thousand is prime, you only need to check if 1000 is divisible by the numbers 2 to 31 (square root of 1000). Hope this helps, this makes it a lot faster.
The algorithm explanation is genuinely super easy. Great job as always :D Also a quick question; if we use a faster compiling language than python will the time change be noticeable or same?
Thanks!🙏🏻 Yes the time difference if we use a faster language (e.g C/C++) should be noticeable! I just used python because it’s the language I am most familiar with 😁
I think, the main problem with faster algorithms here is results output. Most of the time this code would spend printing the results. If it is true, the way of working with i/o is more important (in terms of performance) than programming language choice
Yes, I actually did this a while back with 3 languages and this sieve. Nodejs, C++, and Python Also, a quick note, these tests were automated back to back, mixing the order. along with that I did not have each number printed into the console because printing, especially on Windows, is slow. On going to 1m, and averaged over 5 times, the times were: Python: 1797.4 ms Node: 149.8 ms (12x faster) C++: 46.7 ms (38.4x faster) then for 10m, again averaged over 5 times: Python: 23752.1 ms Node: 1214.1 ms (19.6x faster) C++: 699.8 ms (33.9x faster) Also, here are the peak ram usages going from the 1m run, then 10m run, were Python: 86.7 mb, 772.9 mb Node: 36.6 mb, 119.7 mb C++: 4.9 mb, 15.6 mb C++ is so much better here btw because I used bit arrays (std::vector), basically each number bool takes up a single bit in the array. (this is how the sieve works btw)
Question: Does the respective script exclude numbers ending in 0, 2, 4, 5, 6, 8? Because these are certainly not prime numbers. Where can I get the scripts?
Good catch, this was actually an oversight when I made the video. The trial division algorithm actually had the square root as the limit, I just forgot to explain it during the video. Hope that answers your question!