Python Multiprocessing Tutorial: Run Code in Parallel Using the Multiprocessing Module
ฝัง
- เผยแพร่เมื่อ 21 ก.ย. 2024
- In this video, we will be learning how to use multiprocessing in Python.
This video is sponsored by Brilliant. Go to brilliant.org/cms to sign up for free. Be one of the first 200 people to sign up with this link and get 20% off your premium subscription.
In this Python Programming video, we will be learning how to run code in parallel using the multiprocessing module. We will also look at how to process multiple high-resolution images at the same time using a ProcessPoolExecutor from the concurrent.futures module. Let's get started...
The code from this video can be found at:
bit.ly/multipro...
Threading Tutorial - • Python Threading Tutor...
List Comprehensions Tutorial - • Python Tutorial: Compr...
Exception Handling Tutorial - • Python Tutorial: Using...
Pillow (PIL) Tutorial - • Python Tutorial: Image...
✅ Support My Channel Through Patreon:
/ coreyms
✅ Become a Channel Member:
/ @coreyms
✅ One-Time Contribution Through PayPal:
goo.gl/649HFY
✅ Cryptocurrency Donations:
Bitcoin Wallet - 3MPH8oY2EAgbLVy7RBMinwcBntggi7qeG3
Ethereum Wallet - 0x151649418616068fB46C3598083817101d3bCD33
Litecoin Wallet - MPvEBY5fxGkmPQgocfJbxP6EmTo5UUXMot
✅ Corey's Public Amazon Wishlist
a.co/inIyro1
✅ Equipment I Use and Books I Recommend:
www.amazon.com...
▶️ You Can Find Me On:
My Website - coreyms.com/
My Second Channel - / coreymschafer
Facebook - / coreymschafer
Twitter - / coreymschafer
Instagram - / coreymschafer
#Python
Hey everybody! I hope you find this video helpful. I'm finally moved into my new home, so I should be back on a schedule of regular releases very soon. Hope everyone has a great weekend!
Please make a video on asyncio too..
@@deadfool6105 yes yes
Congrats on the move
Corey, can you make jupyter notebooks on these tutorials also. By the way congratulations!
Bless you man!
Why do people give thumbs down to someone who teaches for free? And a small detail ... the man is a grandmaster! Thank you sir for doing this. Greetings from the Netherlands
There are jealous because they're programming in Cobol.
I think it's the Australians. You see, they have everything upside down there, when they thumbs up, it's thumbs down for us. Corey is the best.
I think it's some kind of obsessive compulsive disorder.
~0.78% of people who voted, thumbed down down.
.. because its not working for them.. so they wasted their time getting 'taught' something that didn't work. .hence thumbs down.
There are courses that take thousands of dollars for teaching what you're teaching. you're teaching very well and you give it free.
Respect.
Great video!
If you are running this in Windows and a later version of Python (I'm running 3.7), you need to run all multiprocessing under main(), then run if __name__ == '__main__': main()
Make sure the do_something() function is declared outside main().
current
Thank you! Just the fix I was looking for. This is apparently true of macs as well
You fixed my Programm🤝
Only the code that spawns the subprocess needs to be guarded.
thank you so mutch
31:12 Mr. Schafer finishes with the theory and goes into the real-world example with processing images
35:37 implementing multi-processing to single thread/process operation
40:04 easy switch from Processes to Threads
Love Your videos Mr. Schafer. Especially Your communication skills. No BS, just clear communication. I also love that You don't edit out mistakes, and summarize after several steps. Thank You for putting the effort and sharing.
I just started an internship in Germany and you have no idea how helpful these classes series are to young people like me that want to succeed in the professional world of today :)
when i watch your vidieos i belive that if you can't explain something easy to understand, absolutely nobody on this world can do that. thanks you alot Corey, you make my life easier.
Thnx sir I got placed in a MNC all bcz of you learned python a lot and still learning a lot whenever I have any doubts stack overflow and you r my mentor thnx again preparing for higher studies now..
i pursuing a college degree and I'm learning python in different resources and i learned the basics , will you please recommend me a good resource for mastering python language and to go to next level??
It's one of those tutorials that has you reaching for the like button several times during the video, and once more after it's ended.
Well, then let's hope this urge posesses you an uneven number of times or it would be unfair to the author.
love the honesty, i feel like i've understood the concepts even more.
I am not in a position right now to donate through patreon. All I can do is watch your videos without skipping ads and give it a like.
That helps a ton too! Thanks!
This video is gold, every second of it. You are very talented at teaching complex concepts, you can't even begin to fathom how much I appreciate your work. Thanks!
when getting to learn the idea of data parallel and distributed data parallel, they stuck me a lot, but after watching your 2 videos, they help me a lot. Thanks sir, you're a life saver !
Of all the resources out there that I have looked to get a start and grasp on this subject (some cost money), this video is on point!
Corey's channel has done more for me in 2 weeks than any other resource I have tried over the last 10 years.
I struggle a bit to run this trivial pgm on my venv python 3.11.1. Ultimately, this one has run properly ... Thanks a lot anyway for this tutorial, which I definitely like a lot ! Hope this helps:
import time
from multiprocessing import Process
start = time.perf_counter()
def do_something(seconds=1):
print(f'Sleeping {seconds} second(s)...')
time.sleep(seconds)
return f'Done Sleeping...{seconds}'
if __name__ == '__main__':
p1 = Process(target=do_something, args=[2])
p2 = Process(target=do_something, args=[3])
p1.start()
print(f"Process 1 pid {p1.pid}")
p2.start()
print(f"Process 2 pid {p2.pid}")
p1.join()
p2.join()
finish = time.perf_counter()
print(f'Finished in {round(finish-start, 2)} second(s)')
the longest deja vu in my life so far
same, lmao
Same LOL
I swear
I needed to check github repo first to confirm that I was not watching the same tutorial yesterday ;))
Hahaha! Same
Still. After 3 years - it's still helpful. Even counting that I need to test this out - nothing that much has changed in Python for MP. Thanks for such an excellent explanation!
Another wonderful video. Thanks to your videos I've gained an understanding of how to build and deploy a Django web application, using matplotlib for data visualisations, python multi-threading, python-multiprocessing, exception handling, python unit test cases and more. Your content is consistently excellent because of how practical and well thought out your videos are. Extremely useful! Please keep making these videos! Thanks a lot! You're helping a lot of people with this high quality content!
Even these years, this video is really informative about that topic
We need at least a dozen of Corey Schafer in our programming learning life. I wish everyone would follow your method. You are simply the best! God bless.
oh man, you don't even know how long I was looking for a video on multiprocessing.
This has been incredibly helpful for me and my team. I'm a network engineer working on multiprocessing SSH connections and this made everything make sense. Definitely subscribed after seeing this.
I was pretty scared of touching this topic, but you made it so easily to understand!
In newer versions of Python instead of using a list in args you should use tuple. So instead of writing this args=[1.5] write this args=(1.5,). It's okay to use the list version, but you will get a warning. Good tutorial btw.
Not anymore ...
self = reduction.pickle.load(from_parent)
AttributeError: Can't get attribute ...
Nice explanation about paralel processing! Thank you!
Nice explanation about paralel processing! Thank you!
Nice explanation about paralel processing! Thank you!
Nice explanation about paralel processing! Thank you!
Nice explanation about paralel processing! Thank you!
Thanks Corey! Next please do async!!
Yes, this is concept I haven't used but would really like to learn about.
Yes please, async and when to use async/asyncio/threading/multiprocessing.
Yes would strongly appreciate this
Yes, async will a perfect next. Thanks Corey for the handwork.
Yes please and then about locks,race conditions and queues as well. Thanks a lot.Loved your explanation.
By substituting a recursive Fibonacci call as the process you can really get a handle on how this helps.
```
def recursive_fib(n):
if n == 0 or n == 1:
return n
return recursive_fib(n-1) + recursive_fib(n-2)
```
pass ``args=[55]`` and watch your CPU utilization hit 100%. Very satisfying (even if it is useless).
@Corey Schafer,
Thank you very much for taking the time to make this video. I have written my first-ever parallelized code and the speedup is invaluable.
For years I was apprehensive about learning parallelization methods because they seemed very complicated. You, my friend, have opened the flood gates. This work puts me one step closer to completing my Master's.
Absolutely amazing video. Well if anybody is getting any error for the same code as in the video then use the main function in your code. I had encountered it (don't know why)...
Just came here to say thanks for this. I just completed my first ever paid work programming, made a simple image slicing and joining script. Client mentioned he wanted it to run though 20000 images in under an hour after I built a demo. I rebuilt it using concurrent.futures and your video saved me hours of work! Plus it runs super quick. Thanks so much!
Curious how much something like this pays? I wrote a similar but much more complex script for myself recently
Corey is great. When you help people, and I mean really HELP people, he is just a saint. I am on Python 3.9.12 and my code does not execute properly, even though I am copying verbatim and I a stuck at the 10:20 mark. After I put in the p1/p2.start() statements, the script runs without printing the "sleep" statements. The final print statement does execute.
But after I put in the p1/p2.join() statements, I get a runtime error:
An attempt has been made to start a new process before the
current process has finished its bootstrapping phase.
This probably means that you are not using fork to start your
child processes and you have forgotten to use the proper idiom
in the main module:
if __name__ == '__main__':
freeze_support()
...
same here
another one with same error
I have been looking for videos like this for like over 1 year, and today I randomly saw your video. Very helpful, easy to understand, thanks a lot Corey!
Hi Corey, great video covered a lot, one thing I was hoping to see here which I did not was using Pool from Multiprocessing to pass multiple parameters to a class or function. I ended up finding it after some digging using starmap with zip instead of map. To pass constants with this method import itertools and use the repeat inside the zip. Hope this helps anyone looking for this.
From my experience, the only issue with the image example is that there are limits to the number of read/write ops an SSD can do, so if you assign too many processes you quickly run into hung threads. A pure CPU example with no read or write ops could arguably demonstrate the point better, but the video was fantastic and I am probably just splitting hairs...
Nice work with the real world example not only does it help the learning process after the foundation has been laid. It also encourages thought on personal projects that will help you apply this new-found knowledge.
I had to make an index by checking 150,000 numbers over 110,000 entries:
Before Multiprocessing: 3558 seconds
After Multiprocessing: 115 seconds
Many, many thanks!
Brah i was stuck at work tryna do this. Searched your name and braaaah! Your videos are so helpful.
same
I watch both of threading and multiprocessing video together without skipping a moment. I enjoy your video more than movie.
thx so much, already implemented that into my webscraping scripts and the speed gain is tremendous. you’re explaining this so well an clear. thx a lot.
Hear trying to achieve the same
By 22:48 this video had saved a project I've been working on for close to three years when the volume of data started taking to long to process and the memory load cause it to swap badly.
Thank you. :)
That's the module I used to parallelize and optimize Mosaik smart grid co-simulations in my bachelor's thesis, which I finished this week! :D
That last piece showing how to switch from processes to threads for testing was super cool. Thanks for this great video.
When faced with a python challenge, I ask myself "What would Corey Schafer do?" I then, somewhat ironically, ask myself "Why am I wasting my time asking myself rhetorical questions, when I could be watching one of his outstanding videos."
This is A GREAT VIDEO!!!Thanks for explaining each step of the code
Thanks for the smooth and didactic intro to multiprocessing, I've done this a year ago during a fast and furious degree and needed a refresher. I have a case in which this will come in handy although it's hundreds of lines of code, but I bet I can wrap them all into a function to be the executor's target.
A function with hundreds of lines might not be such a good idea lol. Try learning Classes and Methods and I’m sure you can incorporate multiprocessing in a much more encapsulated way
@@drygordspellweaver8761 right you are, god knows what insanity I was up to 2 years ago.
@@bigbangdata haha must have been a blast
I learned Flask and SQLAlchemy from this guy. I know this video will be good.
This is the second video I've watched from your channel and was blown away at the quality of the content and the clarity of the concepts explained. Once I get into a data analyst role, I will support you on Patreon. Until then, I will watch every other video! 😁Thanks a million!
Thanks! That helps more than you know. I really appreciate it!
Wow! The use of map() is brilliant way to automatically batch the jobs for each process. Thanks for showing this!!
Just a reminder, on Windows you need to add "if __name__ == '__main__'"
Thanks! I usually run my videos through my Windows laptop before posting just to make sure it's the same for both operating systems, but I didn't do that this time. I appreciate the comment!
@@coreyms Somehow, me too facing errors without using main function. Even after using the if __name__ == '__main__', i got my output in different way .
done sleeping 5 second(s).....
done sleeping 4 second(s).....
done sleeping 3 second(s).....
done sleeping 2 second(s).....
done sleeping 1 second(s).....
sleeping 3 second(s)....
sleeping 2 second(s)....
sleeping 1 second(s)....
sleeping 4 second(s)....
sleeping 5 second(s)....
Finished in 5.5 second(s)
Just curious (python beginner), where would that go in the script?
@@whoa_that_ruled
import multiprocessing
import time
start = time.perf_counter()
def do_something():
print('Sleeping for 1 second')
time.sleep(1)
print('Done sleeping...')
if ___name___ == '__main__':
p1 = multiprocessing.Process(target=do_something)
p2 = multiprocessing.Process(target=do_something)
p1.start()
p2.start()
p1.join()
p2.join()
finish = time.perf_counter()
print(f'Finished in {round(finish-start,4)} seconds')
Add a shebang directive and you wouldn’t need one.
Waoovv. It's a great video. Even the person who know nothing about coding will understand what you're talking about :) Thanks for description...
Thank you so much for this quality of videos on Python!... Cant tell you how much i`ve learned with them. Keep it coming! :D
PS. Any GUI tutorials coming?
PS2. Greetings from Chile!
Thank you, Corey, for such a wonderful explanation. The best part of this video is you have used the real-time problem, which helps to understand the concept in more depth.
Have never understood multithreading and multiprocessing this clearly ever before in my life. Would greatly appreciate if you can post a video on locks. Deadlock and livelock and how to resolve them. Thanks for the service you are doing to so many people.
The reason it finished the 1s future in the Multiprocessing part after the 2 and (just after) the 3s future is that the PoolExecutor limited the threads to 4 (you said yourself that it chooses the number of concurrent threads based i.e. your hardware). And your list of times to wait is reversed in duration. So it only started the future for the 1s duration AFTER the first future finished, the one with a duration of 2s. So the 1s future finished after 2+1= 3s plus overhead, so slightly after the 3s future.
Edit: Aaand you realized that just shortly after. Was wondering why noone else wrote this before.
No one mentioned it because it’s literally spelled out in the video 🤦♂️
@@drygordspellweaver8761 I did wait some time to write that comment.
You meanwhile were too lazy to read my whole comment.
@@9SMTM6 Literally everyone who watched the video understood what was going on. Only you saw fit to go out of your way to write a comment to "explain" what was literally explained in the video like 5 seconds after it happened. You are insufferably arrogant and dense.
Beautiful sir, I really love the manual way for when you are going to work with different functionalities, like in a producer consumer architecture. But use pools when you are doing a reusable module.. Thanks for this tutorial! My respects! 😎
Hi Sir,
As you mentioned, I am interested to learn advanced threading concepts like race condition, mocks, etc. Please make videos on that too. I will be waiting... Between really nice video. Thanks for the effort.
Thank you Corey, it is brilliant to start from the basics. Really helpful approach and advice !
Dude you are amazing! Excellent excellent stuff! Truly great educator and we own you a lot.
I have a question. What if my function takes more than one argument. Will the ProcessPoolExecutor instance (executor) be able to map the function to however number of arguments for every process?
It’s the map method that would handle that. The function then would naturally need to accept multiple arguments
Almost the identical video to the multithreading video. I learned enough to know that IO bound is right choice for multithreading.
You could have put the image processing and the reading and writing from disk in separate functions you could have used threads for the IO and procesess for the processing
I am thankful to you for this video.
I have come to know that yes conceptually threads and multiprocessing are different but it is exactly same in terms of implementation. Only methods and internal concept is different.
Excelent tutorial!, just for completeness, i'd love to see some tutorial on python coroutines, thanks a lot.
What is that
Me too
You are the best teacher I have ever had. I wish I could give this video a million likes.
Any computer vision tutorial coming ? Would be awesome !
15:06 my favorite part of the video
I had to add
if __name__ == '__main__':
for it to work in Python 3.7.6
I did not get the print output, but the images were processed
The same for me!
def main():
if __name__ == "__main__":
main()
and it worked, anyone can help to explain why?
thanks man , you are a savior , after watching your videos on multiprocessing and threading i could finally make a project i wanted to make for a long time - python which counts the number of times i press each key and store them in an excel file for each day
At 40:40 you're omitting one important thing: GaussianBlur() is an extension using a library written in C. So it can processing the image in a separately managed thread in C and relinquish execution to the python interpreter to process python bytecode on other python threads.
So it's essentially a bad example, because at the low-level it behaves like multiprocessing in the sense that it can run things outside the python interpreter (multiprocessing does that in a separate process, PIL does that in a private thread, in C).
So your assessment that the workload is I/O bound is incorrect. It is mostly CPU bound, but the processing you chose released the GIL, that's an unfortunate choice.
Sir, this video is a fucking treasure. It Saved me a bunch of time reading shitty tutorials on google.
31 people didn't find a piece of code they could copy and paste for their work.
#indians
sorry
your sopposed to follow along
NΛVΛZ don’t be jealous
57 now
Wow!! I kind of understood multi processing, but this polished up everything I know into a very sparkly vision.... Thank you!!!
What's the difference between time.perf_counter() and time.time()? Why should I use one over the other?
The resolution of perf_counter() is higher than that of time().
But he is eventually rounding off😅
thx Corey for this great video! I'm a biomedicine master student from Sweden. This tutorial enabled me to use multiprocessing in a program I write for dna analysis during my master thesis!:)
Bro, can you tell me why this error pops up?? I'm trying to print the return value of the function.
concurrent.futures.process.BrokenProcessPool: A process in the process pool was terminated abruptly while the future was running or pending.
Very good video, though yes, this is really frustrating me too!
@@concert_music You have to encase your code in that
> if ___name___ == "__main__":
thing. This is done to avoid unwanted side effects, and multiprocessing and concurrent.futures libraries both force programmers (at least on Windows; Corey is on Mac) to do that. Hope that helps! :D
Took me a while to find the issue: stackoverflow.com/questions/61860800/running-a-processpoolexecutor-in-ipython
Basically I added those lines on top when I'm using MacOS on jupyter-lab:
import multiprocessing as mp
mp.set_start_method('fork')
The best instructor in the world... you explain tough things very easily... better than all others(including paid)
Instructor 100%👌
Content 100%👌
Quality 100%👌
Explanation 100%👌
Thankyou so much ❤ keep inspiring us...
Thanks!
That's so humble you replied sir... sir can you please upload numpy and pandas tutorial
@@shivstn98 I hope you already found that from his playlist, there are 11 videos for Pandas.
for i inrange(1,1000000):
print ("Thank you")
print("Thank you
"*1000000)
# :)
@@jimmy2643 wow. I didn't know this was possible... thank you!
I am about to get into using Pandas to gather data from multiple spreadsheets and output cleaned up versions of those sheets as well as generate some simple reporting from the data, and I am so glad that I came across your multiprocessing video! While my use case would not take more than maybe a minute or two to run, I'm big on efficiency and if it can be run in under 15 seconds instead that is a win for me.
It's great video. However, I receive the following error when run your code to print results in my windows. But the code works when I run it in aws environment.
BrokenProcessPool Traceback (most recent call last)
in
3 results = executor.map(do_something, secs)
4
----> 5 for result in results:
6 print(result)
7
~\AppData\Local\Continuum\anaconda3\lib\concurrent\futures\process.py in _chain_from_iterable_of_lists(iterable)
474 careful not to keep references to yielded objects.
475 """
--> 476 for element in iterable:
477 element.reverse()
478 while element:
~\AppData\Local\Continuum\anaconda3\lib\concurrent\futures\_base.py in result_iterator()
584 # Careful not to keep a reference to the popped future
585 if timeout is None:
--> 586 yield fs.pop().result()
587 else:
588 yield fs.pop().result(end_time - time.monotonic())
~\AppData\Local\Continuum\anaconda3\lib\concurrent\futures\_base.py in result(self, timeout)
430 raise CancelledError()
431 elif self._state == FINISHED:
--> 432 return self.__get_result()
433 else:
434 raise TimeoutError()
~\AppData\Local\Continuum\anaconda3\lib\concurrent\futures\_base.py in __get_result(self)
382 def __get_result(self):
383 if self._exception:
--> 384 raise self._exception
385 else:
386 return self._result
BrokenProcessPool: A process in the process pool was terminated abruptly while the future was running or pending.
@Hongbing , even I am getting the same error in windows. Did you manage to get around it - if so then how ?
@@surajitchakraborty1903 I'm getting similar errors in pycharm - not exactly the same, but similar (also on windows)
Very clear and concise explanation - well done sir! Other kudos running on other processor that hasn't finished yet... 🤣
Corey is a divine teacher...
thanks a lot Mr Schafer!! your videos helps me a lot, plus the tips u give to keep the code clean are making my projects look awesome
This has helped me take a task that usually takes 3 hours, down to 30 minutes. Thank you! Best thing I ever did was watch these
I do have to thank you @Corey Schafer your work in these videos. I learn a lot from them and in the end I use them on my work to automate some tasks. Some love from Portugal
Seriously learning more from your vids on Python than the Python 'class' I took one time. Clear and helpful. Everytime I watch I learn something new.
Great Video!!! As always, the best, most logical explanation on TH-cam.
I came to this video with a working knowledge of python threads, processes and asynchronous coroutines.
I must say, this video was very well-executed. It helped me arrange my own thoughts on the matter, and I think anyone new
to the concept of multiprocessing would find Corey's explanation easy to understand.
I also commend his insistence on showing a practical example, as it helps a lot of people bring new concepts into perspective.
Congratulations on a job well done!
Yeah sure you are a thread expert that’s why you watched a tutorial video.
People and their cringey little egos... lmao 🤦♂️
Corey, you're the best. Thanks again from brazil.
This video came on time. I work with AI with python and, at university, I'm seeing operating system. Now I can join OS and python for do faster algorithms.
Thanks Corey. I really like your videos!
incredible video! you've covered more information than all articles all together that i've read so far - thank you for that!
I joined the C# start to finish course last year and am so glad that you did this video. Hope you are keeping well. Thanks for this well explained tutorial
You are a real real giga chad for providind this for free !
Yep, this video was 100% helpful. Even the first 10 minutes was helpful, to start with Multiprocessing.
Do you have video or lessons for TCP IP connections or\and Bluetooth connections?
I teach children's in robotic-Club, we make games on python and robots on Arduino and Lego Mindstorm. So main idea connect to the robot by bluetooth on server PC, and using TCP send information to all client PC's and draw GUI using Real Data.
Many years ago I made programm in that way, but not in Python.
Your video was like quick start - simple, clear and easy to make, and easy to understand
One of the best video of Multiprocessing I have ever seen. Thanks for all your effort and teaching us.
This video was a great introduction for a Python newbie like myself, with previous experience using a Future API in R.
You're a grandmaster. I really appreciate you put all stuff in ONE video. So easy to navigate. Thank you!
Outstanding video! Never learned so much about multiprocessing in python - in just 45 Minutes. Thank you, Corey!
This is the best explanation of that topic I have watched so far!
Many thanks, Corey!
Awesome video! I have gone through 20min and will go through rest again later... people should learn from
This how to create a teaching video with high amount of content per minute and not waste time of unnecessary sentences.
I got enough in 20 minutes to create my stock scraping process in my program.
my new favorite python teacher! TY
He asked us to run a code in which there are 12 proceses, you divide them by 3 every 4 run a task
4=>put
4=>corect
4=>get
We have a queue of 20
What is required is that the processors responsible for put put random numbers in queue
The process spacing of the correct response decreases the value by 2 from the random number, and the process spacing that results in the get read the values in the queue
I told you, you have to put in order to get the value of receiving a branch.
And between the process of putting, correcting and getting, it must be done in 1.5 minutes