Geoff Ruddock

Jupyter notebooks

Jupyter is an open-source tool for executing Python code in an interactive notebook environment.



This is the boilerplate code I use to initialize every notebook.

You can add boilerplate imports to ~/.ipython/profile_default/startup/ to be executed every time the kernel is initialized.

import os, sys
import datetime as dt
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

%matplotlib inline
%reload_ext autoreload
%autoreload 2

from IPython.core.interactiveshell import InteractiveShell
InteractiveShell.ast_node_interactivity = 'all'  # display all output cells

from IPython.core.display import display, HTML
display(HTML("<style>.container { width:100% !important; }</style>"))  # make full width

pd.set_option('float_format', lambda x: '%.3f' % x)
pd.set_option('display.max_rows', 200)
np.set_printoptions(suppress=True, linewidth=180)

SQL syntax highlighting

Add the following lines to ~/.jupyter/custom/custom.js:'kernel_ready.Kernel',
               .highlight_modes['magic_text/x-mysql'] = {'reg':[/^%%sql/]} ;
                if (cell.cell_type == 'code'){
        }) ;
    }) ;


Suppress output

Add ; to the end of the line, useful when you want to prevent text output when plotting.

data = np.random.exponential(size=1000)
sns.histplot(data, kde=False)


data = np.random.exponential(size=1000)
sns.histplot(data, kde=False);


Check python version

from platform import python_version


Tips & tricks

Idempotent pip installs

If your notebook has dependecies, you can make it “one-click runnable” using !pip install -Uqq module.

This will silently install or upgrade a pip package, showing no output unless an error occurs.

Source: StackOverflow > pip install options unclear

Progress bars w/ tqdm

Source: how to make a nested tqdm bars on jupyter notebook

from time import sleep
from tqdm.notebook import tqdm
from IPython.display import clear_output

iters_outer = 3
iters_inner = 5
for i in tqdm(range(iters_outer), desc='Outer'):
    for j in tqdm(range(iters_inner), desc='Inner', leave=(i==iters_outer-1)):


The tqdm progress bars do not render properly when this notebook is converted to markdown, but below is a screenshot of what it looks like in-notebook.

TQDM output


Our test function

from time import sleep
from random import random

def my_func():
    sleep(random())  # Random number in range [0, 1]
    return True

Timing execution

timeit magic


%timeit my_func()
565 ms ± 69.1 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
%timeit -r 100 -n 1 my_func()
The slowest run took 829.94 times longer than the fastest. This could mean that an intermediate result is being cached.
552 ms ± 302 ms per loop (mean ± std. dev. of 100 runs, 1 loop each)
%timeit -r 1 -n 100 my_func()
507 ms ± 0 ns per loop (mean ± std. dev. of 1 run, 100 loops each)
%timeit -r 10 -n 10 my_func()
576 ms ± 79.9 ms per loop (mean ± std. dev. of 10 runs, 10 loops each)
%timeit -r 50 -n 2 my_func()
The slowest run took 7.53 times longer than the fastest. This could mean that an intermediate result is being cached.
477 ms ± 168 ms per loop (mean ± std. dev. of 50 runs, 2 loops each)

So the best approach is to call timeit with arguments -r <n> -n 1, since otherwise it will underestimate the variability in run times.

Line profiling

!pip install -Uqq line_profiler

%reload_ext line_profiler

%lprun -f my_func my_func()

Memory profiling

Can be used for:

!pip install -Uqq memory_profiler
%reload_ext memory_profiler

%memit my_func()
peak memory: 51.43 MiB, increment: 0.00 MiB


%debug magic (docs)

comments powered by Disqus