Reclaim memory usage in Jupyter

edit ✏️

Today I ran into an issue where we had a one-off script that just needed to work, but it was just chewing threw memory like nothing.

It started with a colleague asking me How do I clear the memory in a Jupyter notebook, these are the steps we took to debug the issue and free up some memory in their notebook.

How do I clear the memory in a Jupyter notebook?

Pre check the status of memory.

There are a number of ways that you can check the amount of memory on your system. The easiest is not necessarily my first go to is free... literally free .

check for free space

$ free -h
             total       used       free     shared    buffers     cached
Mem:           15G        15G       150M         0B        59M       8.7G

Generally my first go to is a bit more graphical, and not available on a stock stystem, but far more useful.... htop . htop is a terminal process explorer that shows cpu usage, mem usage, and running processes.


sudo apt-get install htop # install it from your package repo

htop in use

First step throw more swap at it

Often before going through the process of getting a larger instance underneath the notebook you can hobble home with a bit more swap file. It may not be pretty or fast, but gets the job done in a pinch.

Check for free disk

$ du

Filesystem      Size  Used Avail Use% Mounted on
/dev/asdasd        200G   50G  150G   25% /

Make sure you check your free disk space first, filling both memory and disk can be bad news

make a swap file and activate it

mkdir ~/swaps
sudo fallocate -l 50G $SWAPFILE
sudo chmod 600 $SWAPFILE
sudo mkswap $SWAPFILE
sudo swapon $SWAPFILE

You can see the results with either swapon or free.

sudo swapon --show
free -h

How to Add Swap Space on Ubuntu 20.04

linuxize how to add swap space on ubuntu 20.04

More details on creating swapfiles checkout linuxize . It is my favorite linux tutorial site!

Refactor - functions

keep big datasets inside functions returning aggregations

Sometimes there is a clear quick and simple way to just let the python garbage collector. Often we pull in large datasets to create features then aggregate them down into smaller datasets that can be then joined into other datasets. This pattern of pulling in big_data , processing then aggregating can be a simple one.

let the garbage collector take care of big data

def process():
   big_data = get_big_data()
   smaller_data = <some aggregation>
   return smaller_data
data = process()

If your notebook is following this type of pattern a simple del won't work because ipython adds extra references to your big_data that you didnt add. These are things that enable features like _ , __ , ___ , umong others.


check out more on reset from the ipython docs

The last resort I would lean on here is an ipython specific feature %reset and %reset_selective . These will flush out all user define variables or selecive ones based on a regex respectively.

Following two example are directly from the ipython docs


In [6]: a = 1

In [7]: a
Out[7]: 1

In [8]: 'a' in get_ipython().user_ns
Out[8]: True

In [9]: %reset -f

In [1]: 'a' in get_ipython().user_ns
Out[1]: False

In [2]: %reset -f in
Flushing input history

In [3]: %reset -f dhist in
Flushing directory history
Flushing input history


In [2]: a=1; b=2; c=3; b1m=4; b2m=5; b3m=6; b4m=7; b2s=8

In [3]: who_ls
Out[3]: ['a', 'b', 'b1m', 'b2m', 'b2s', 'b3m', 'b4m', 'c']

In [4]: %reset_selective -f b[2-3]m

In [5]: who_ls
Out[5]: ['a', 'b', 'b1m', 'b2s', 'b4m', 'c']

In [6]: %reset_selective -f d

In [7]: who_ls
Out[7]: ['a', 'b', 'b1m', 'b2s', 'b4m', 'c']

In [8]: %reset_selective -f c

In [9]: who_ls
Out[9]: ['a', 'b', 'b1m', 'b2s', 'b4m']

In [10]: %reset_selective -f b

In [11]: who_ls
Out[11]: ['a']

Develop faster utilizing autoreload in ipython

The above tips will help you reclaim used memory in ipython, but the following tip is one that single handedly is the reason I use Ipython for faster development over anything else.

Autoreload in Ipython

autoreload-ipython one of my biggest productivity boosts.