如何在 jupyter notebook 5 中逐行分析 python 3.5 代码

Question

How to find out execution time taken by each line of python code.

line_profiler works with ipython but doesnt work with jupyter notebook. I tried adding @profile to my function, it gives error saying name 'profile' is not defined. There is one way to do it by time.time() , but i was wondering if there is any inbuilt profiling function which can profile each line of my function and show me the execution time.

def prof_function():
    x=10*20
    y=10+x
    return (y)

Answer 1

You can use line_profiler in jupyter notebook.

1. Install it: pip install line_profiler
2. Within your jupyter notebook, call: %load_ext line_profiler
3. Define your function prof_function as in your example.
4. Finally, profile as follows: %lprun -f prof_function prof_function()

Which will provide the output:

Timer unit: 1e-06 s

Total time: 3e-06 s
File: <ipython-input-22-41854af628da>
Function: prof_function at line 1

Line #      Hits         Time  Per Hit   % Time  Line Contents
==============================================================
     1                                           def prof_function():
     2         1          1.0      1.0     33.3      x=10*20
     3         1          1.0      1.0     33.3      y=10+x
     4         1          1.0      1.0     33.3      return (y)

Answer 2

To get the execution time for each line and get a nice colour coded heatmap, I use this nice ipython magic.... https://github.com/csurfer/pyheatmagic

Installation:

pip install py-heat-magic

To profile every single line in notebook:

• duplicate your notebook.
• Merge all cells (highlight all and shift-m)
• Create a new cell at the top
• enter

%load_ext heat

At the top of your 2nd cell enter this at the 1st line:

%%heat  

You may have issues if you have more than 2000 lines of code.

Answer 3

Just a summary of @SA's answer

!pip install line_profiler
%load_ext line_profiler

def func():
    print('hi')

%lprun -f func func()

Answer 4

Install line profiler

conda install line_profiler

More info on http://mortada.net/easily-profile-python-code-in-jupyter.html