Why is np.dot much slower the first time it is called in a python session?





.everyoneloves__top-leaderboard:empty,.everyoneloves__mid-leaderboard:empty,.everyoneloves__bot-mid-leaderboard:empty{ height:90px;width:728px;box-sizing:border-box;
}







2















I am trying to do series of big np.dot(a,x) operations, and the first one seems to take far longer than subsequent calls. In my problem, a is tall [n x 2] and x is [2 x 1]. My big matrix a is constant, it is just x that is changing. Here is a MWE:



import numpy as np



@profile

def do_work(a,x):

    tmp = np.dot(a,x)

    return tmp



@profile

def do_work_iter(a,x):

    tmp = np.dot(a,x)

    return tmp



if __name__=="__main__":



    n = 50000

    a = np.random.randn(n,2)

    x = np.random.randn(2,1)



    #

    tmp = do_work(a,x)



    #

    niter = 100

    for i in range(niter):

        x = np.random.randn(2,1)

        tmp = do_work_iter(a,x)


Using line_profiler, I get .155 s/call for the first call to np.dot and .00013 s/call for the subsequent ones. Is there some setup/error checking that numpy is doing the first time here? Is there a way I can bypass any of it? Or is there some kind of searching for blas functions that is taking all of the time?



I also ran profile and it gives the following:



ncalls  tottime  percall  cumtime  percall filename:lineno(function)

     1    0.000    0.000    1.514    1.514 {built-in method builtins.exec}

     1    0.000    0.000    1.514    1.514 <string>:1(<module>)

     1    0.000    0.000    1.514    1.514 speed_small.py:15(run)

   101    1.503    0.015    1.503    0.015 {built-in method numpy.core.multiarray.dot}

     1    0.000    0.000    1.491    1.491 speed_small.py:5(do_work)

   100    0.000    0.000    0.012    0.000 speed_small.py:10(do_work_iter)


so numpy.core.multiarray.dot is taking all of the time, it doesn't give much insight about anything further down the stack.



I am on Python 3.6 from Anaconda and have mkl installed (Windows 7).






python performance numpy







share|improve this question























  • I use line_profiler but my guess would be that you're incurring the import overhead the first time. After that, The module is cached

    – roganjosh
    Nov 21 '18 at 20:58













  • @roganjosh Isn't the module included before the script comes to if __name__ == '__main__': part ?

    – eozd
    Nov 21 '18 at 20:59













  • It's an external module and nothing in your code suggests you're using it (EDIT: that's a lie, you're using @profile decorator so I'll rethink) I don't know exactly what timings get bundled into the main library timings.

    – roganjosh
    Nov 21 '18 at 21:02








  • 1





    Loading of BLAS?

    – Ante
    Nov 23 '18 at 12:59


















2















I am trying to do series of big np.dot(a,x) operations, and the first one seems to take far longer than subsequent calls. In my problem, a is tall [n x 2] and x is [2 x 1]. My big matrix a is constant, it is just x that is changing. Here is a MWE:



import numpy as np



@profile

def do_work(a,x):

    tmp = np.dot(a,x)

    return tmp



@profile

def do_work_iter(a,x):

    tmp = np.dot(a,x)

    return tmp



if __name__=="__main__":



    n = 50000

    a = np.random.randn(n,2)

    x = np.random.randn(2,1)



    #

    tmp = do_work(a,x)



    #

    niter = 100

    for i in range(niter):

        x = np.random.randn(2,1)

        tmp = do_work_iter(a,x)


Using line_profiler, I get .155 s/call for the first call to np.dot and .00013 s/call for the subsequent ones. Is there some setup/error checking that numpy is doing the first time here? Is there a way I can bypass any of it? Or is there some kind of searching for blas functions that is taking all of the time?



I also ran profile and it gives the following:



ncalls  tottime  percall  cumtime  percall filename:lineno(function)

     1    0.000    0.000    1.514    1.514 {built-in method builtins.exec}

     1    0.000    0.000    1.514    1.514 <string>:1(<module>)

     1    0.000    0.000    1.514    1.514 speed_small.py:15(run)

   101    1.503    0.015    1.503    0.015 {built-in method numpy.core.multiarray.dot}

     1    0.000    0.000    1.491    1.491 speed_small.py:5(do_work)

   100    0.000    0.000    0.012    0.000 speed_small.py:10(do_work_iter)


so numpy.core.multiarray.dot is taking all of the time, it doesn't give much insight about anything further down the stack.



I am on Python 3.6 from Anaconda and have mkl installed (Windows 7).






python performance numpy







share|improve this question























  • I use line_profiler but my guess would be that you're incurring the import overhead the first time. After that, The module is cached

    – roganjosh
    Nov 21 '18 at 20:58













  • @roganjosh Isn't the module included before the script comes to if __name__ == '__main__': part ?

    – eozd
    Nov 21 '18 at 20:59













  • It's an external module and nothing in your code suggests you're using it (EDIT: that's a lie, you're using @profile decorator so I'll rethink) I don't know exactly what timings get bundled into the main library timings.

    – roganjosh
    Nov 21 '18 at 21:02








  • 1





    Loading of BLAS?

    – Ante
    Nov 23 '18 at 12:59














2












2








2


2






I am trying to do series of big np.dot(a,x) operations, and the first one seems to take far longer than subsequent calls. In my problem, a is tall [n x 2] and x is [2 x 1]. My big matrix a is constant, it is just x that is changing. Here is a MWE:



import numpy as np



@profile

def do_work(a,x):

    tmp = np.dot(a,x)

    return tmp



@profile

def do_work_iter(a,x):

    tmp = np.dot(a,x)

    return tmp



if __name__=="__main__":



    n = 50000

    a = np.random.randn(n,2)

    x = np.random.randn(2,1)



    #

    tmp = do_work(a,x)



    #

    niter = 100

    for i in range(niter):

        x = np.random.randn(2,1)

        tmp = do_work_iter(a,x)


Using line_profiler, I get .155 s/call for the first call to np.dot and .00013 s/call for the subsequent ones. Is there some setup/error checking that numpy is doing the first time here? Is there a way I can bypass any of it? Or is there some kind of searching for blas functions that is taking all of the time?



I also ran profile and it gives the following:



ncalls  tottime  percall  cumtime  percall filename:lineno(function)

     1    0.000    0.000    1.514    1.514 {built-in method builtins.exec}

     1    0.000    0.000    1.514    1.514 <string>:1(<module>)

     1    0.000    0.000    1.514    1.514 speed_small.py:15(run)

   101    1.503    0.015    1.503    0.015 {built-in method numpy.core.multiarray.dot}

     1    0.000    0.000    1.491    1.491 speed_small.py:5(do_work)

   100    0.000    0.000    0.012    0.000 speed_small.py:10(do_work_iter)


so numpy.core.multiarray.dot is taking all of the time, it doesn't give much insight about anything further down the stack.



I am on Python 3.6 from Anaconda and have mkl installed (Windows 7).






python performance numpy







share|improve this question














I am trying to do series of big np.dot(a,x) operations, and the first one seems to take far longer than subsequent calls. In my problem, a is tall [n x 2] and x is [2 x 1]. My big matrix a is constant, it is just x that is changing. Here is a MWE:



import numpy as np



@profile

def do_work(a,x):

    tmp = np.dot(a,x)

    return tmp



@profile

def do_work_iter(a,x):

    tmp = np.dot(a,x)

    return tmp



if __name__=="__main__":



    n = 50000

    a = np.random.randn(n,2)

    x = np.random.randn(2,1)



    #

    tmp = do_work(a,x)



    #

    niter = 100

    for i in range(niter):

        x = np.random.randn(2,1)

        tmp = do_work_iter(a,x)


Using line_profiler, I get .155 s/call for the first call to np.dot and .00013 s/call for the subsequent ones. Is there some setup/error checking that numpy is doing the first time here? Is there a way I can bypass any of it? Or is there some kind of searching for blas functions that is taking all of the time?



I also ran profile and it gives the following:



ncalls  tottime  percall  cumtime  percall filename:lineno(function)

     1    0.000    0.000    1.514    1.514 {built-in method builtins.exec}

     1    0.000    0.000    1.514    1.514 <string>:1(<module>)

     1    0.000    0.000    1.514    1.514 speed_small.py:15(run)

   101    1.503    0.015    1.503    0.015 {built-in method numpy.core.multiarray.dot}

     1    0.000    0.000    1.491    1.491 speed_small.py:5(do_work)

   100    0.000    0.000    0.012    0.000 speed_small.py:10(do_work_iter)


so numpy.core.multiarray.dot is taking all of the time, it doesn't give much insight about anything further down the stack.



I am on Python 3.6 from Anaconda and have mkl installed (Windows 7).






python performance numpy




python performance numpy






share|improve this question













share|improve this question











share|improve this question




share|improve this question










asked Nov 21 '18 at 20:55









Isaac AsherIsaac Asher

384




384













  • I use line_profiler but my guess would be that you're incurring the import overhead the first time. After that, The module is cached

    – roganjosh
    Nov 21 '18 at 20:58













  • @roganjosh Isn't the module included before the script comes to if __name__ == '__main__': part ?

    – eozd
    Nov 21 '18 at 20:59













  • It's an external module and nothing in your code suggests you're using it (EDIT: that's a lie, you're using @profile decorator so I'll rethink) I don't know exactly what timings get bundled into the main library timings.

    – roganjosh
    Nov 21 '18 at 21:02








  • 1





    Loading of BLAS?

    – Ante
    Nov 23 '18 at 12:59



















  • I use line_profiler but my guess would be that you're incurring the import overhead the first time. After that, The module is cached

    – roganjosh
    Nov 21 '18 at 20:58













  • @roganjosh Isn't the module included before the script comes to if __name__ == '__main__': part ?

    – eozd
    Nov 21 '18 at 20:59













  • It's an external module and nothing in your code suggests you're using it (EDIT: that's a lie, you're using @profile decorator so I'll rethink) I don't know exactly what timings get bundled into the main library timings.

    – roganjosh
    Nov 21 '18 at 21:02








  • 1





    Loading of BLAS?

    – Ante
    Nov 23 '18 at 12:59

















I use line_profiler but my guess would be that you're incurring the import overhead the first time. After that, The module is cached

– roganjosh
Nov 21 '18 at 20:58







I use line_profiler but my guess would be that you're incurring the import overhead the first time. After that, The module is cached

– roganjosh
Nov 21 '18 at 20:58















@roganjosh Isn't the module included before the script comes to if __name__ == '__main__': part ?

– eozd
Nov 21 '18 at 20:59







@roganjosh Isn't the module included before the script comes to if __name__ == '__main__': part ?

– eozd
Nov 21 '18 at 20:59















It's an external module and nothing in your code suggests you're using it (EDIT: that's a lie, you're using @profile decorator so I'll rethink) I don't know exactly what timings get bundled into the main library timings.

– roganjosh
Nov 21 '18 at 21:02







It's an external module and nothing in your code suggests you're using it (EDIT: that's a lie, you're using @profile decorator so I'll rethink) I don't know exactly what timings get bundled into the main library timings.

– roganjosh
Nov 21 '18 at 21:02






1




1





Loading of BLAS?

– Ante
Nov 23 '18 at 12:59





Loading of BLAS?

– Ante
Nov 23 '18 at 12:59












0






active

oldest

votes












Your Answer






StackExchange.ifUsing("editor", function () {
StackExchange.using("externalEditor", function () {
StackExchange.using("snippets", function () {
StackExchange.snippets.init();
});
});
}, "code-snippets");

StackExchange.ready(function() {
var channelOptions = {
tags: "".split(" "),
id: "1"
};
initTagRenderer("".split(" "), "".split(" "), channelOptions);

StackExchange.using("externalEditor", function() {
// Have to fire editor after snippets, if snippets enabled
if (StackExchange.settings.snippets.snippetsEnabled) {
StackExchange.using("snippets", function() {
createEditor();
});
}
else {
createEditor();
}
});

function createEditor() {
StackExchange.prepareEditor({
heartbeatType: 'answer',
autoActivateHeartbeat: false,
convertImagesToLinks: true,
noModals: true,
showLowRepImageUploadWarning: true,
reputationToPostImages: 10,
bindNavPrevention: true,
postfix: "",
imageUploader: {
brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
allowUrls: true
},
onDemand: true,
discardSelector: ".discard-answer"
,immediatelyShowMarkdownHelp:true
});


}
});














draft saved

draft discarded


















StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstackoverflow.com%2fquestions%2f53420348%2fwhy-is-np-dot-much-slower-the-first-time-it-is-called-in-a-python-session%23new-answer', 'question_page');
}
);

Post as a guest















Required, but never shown

























0






active

oldest

votes








0






active

oldest

votes









active

oldest

votes






active

oldest

votes
















draft saved

draft discarded




















































Thanks for contributing an answer to Stack Overflow!


  • Please be sure to answer the question. Provide details and share your research!

But avoid …



  • Asking for help, clarification, or responding to other answers.

  • Making statements based on opinion; back them up with references or personal experience.


To learn more, see our tips on writing great answers.




draft saved


draft discarded














StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstackoverflow.com%2fquestions%2f53420348%2fwhy-is-np-dot-much-slower-the-first-time-it-is-called-in-a-python-session%23new-answer', 'question_page');
}
);

Post as a guest















Required, but never shown





















































Required, but never shown














Required, but never shown












Required, but never shown







Required, but never shown

































Required, but never shown














Required, but never shown












Required, but never shown







Required, but never shown







-

Popular posts from this blog

Guess what letter conforming each word

Image
6 $begingroup$ This is an entry in the Fortnightly Topic Challenge #41: Short and Sweet. 1 letter only could be mean anything . With an additional letter in front , could be used as medicine component and something you definitely learn or at least heard once. Add another letter in front , you would possibly hear that many times . Add an additional letter in front , you would found an independent country . With an additional letter, in the end , you would found something related to apple . Add another letter in the end , you would found something bad . Hint: Each letter is unique . word knowledge letters chemistry share | improve this question ...
Read more

Port of Spain

Image
For ports in the country of Spain, see List of ports in Spain. City in City of Port of Spain, Trinidad and Tobago Port of Spain City City of Port of Spain Top to bottom (left to right): Skyline of Port of Spain from Gulf of Paria; National Academy for the Performing Arts, Independence Square, Eric Williams Plaza; MovieTowne, Port of Port of Spain Motto(s):  "We Gather Strength As We Go Along" Port of Spain Coordinates: 10°40′N 61°31′W  /  10.667°N 61.517°W  / 10.667; -61.517 Coordinates: 10°40′N 61°31′W  /  10.667°N 61.517°W  / 10.667; -61.517 Country   Trinidad and Tobago Jurisdiction City of Port of Spain Settled 1560 Incorporated (town) 1853 Incorporated (city) 1914 Government  • Mayor Joel Martinez  • Governing body City Corporation Area [1]  • City 12 km 2 (5 sq mi) Elevation [2] 66 m (217 ft) Population (2011) [1]  • City 37,074  • Rank 3rd  • Density 3,090/km 2 (8,000/sq mi)...
Read more

Run scheduled task as local user group (not BUILTIN)

Image
0 I am trying to run a scheduled task with User Group (not a single user). This is working fine when I in the field "when running a task, use the following user account:" insert BUILTIN User Group (for example: Administrators or Users). But when I try it with the created Local User Group (example: TestGroup) it is not working. The Users in the TestGroup are the same as in the Administrators group. What could be the problem? Are there some restrictions to the Created User Group compared to BUILTINUsers? scheduled-tasks taskscheduler windowstaskschedule share | improve this question asked Nov 15 '18 at 15:36 Far...
Read more