How do I create an AWS Lambda layer for seaborn?

Question

I have been unsuccessful in creating a layer for seaborn to be used on AWS Lambda.
I have tried creating a python environment and installing seaborn using the following commands:
python -m venv python
cd python/Scripts
activate
pip3 install seaborn

The package installs, and then I organize the folder using this heirarchy:
aws_lambda_python37_layer_seabornpythonlibpython3.7site-packagesseaborn
(site-packages contains a number of other folders as well, which are dependencies of seaborn I assume)
I copied the heirarchy from a layer someone else created for matplotlib, which works correctly as a layer on AWS Lambda.
When I try to import seaborn however, I receive this error:
"errorMessage": "Unable to import module 'lambda_function': No module named 'seaborn'"
I am using python 3.7

Marcin · Accepted Answer

I just created the custom layer with seaborn and can confirm that it works. The layer explicitly installs some dependencies, but not uses numpy and scipy. These two packages must be provided by AWS managed layer AWSLambda-Python37-SciPy1x.
Thus you need to use two layers in your function. One custom created below, and the second one is AWSLambda-Python37-SciPy1x.
The technique used includes docker tool described in the recent AWS blog:

How do I create a Lambda layer using a simulated Lambda environment with Docker?

Thus for this question, I verified it  as follows:

Create empty folder, e.g. mylayer.

mkdir mylayer && cd mylayer

Go to the folder and create requirements.txt:

echo seaborn > requirements.txt 
echo matplotlib >> requirements.txt
echo pyparsing >> requirements.txt
echo cycler >> requirements.txt
echo certifi >> requirements.txt
echo pillow >> requirements.txt
echo six >> requirements.txt
echo pandas >> requirements.txt
echo kiwisolver >> requirements.txt
echo pytz >> requirements.txt

Run the following docker command:

Note the use of --no-deps to skip any dependencies appart
from those specified above.
docker run -v "$PWD":/var/task "lambci/lambda:build-python3.7" /bin/sh -c "pip install --no-deps -r requirements.txt -t python/lib/python3.7/site-packages/; exit"

Create layer as zip:

zip -9 -r mylayer.zip python

Create lambda layer based on mylayer.zip in the AWS Console.
Don't forget to specify Compatible runtimes to python3.7.

Test the layer in lambda using the following lambda function:

import json

import seaborn

def lambda_handler(event, context):

print(dir(seaborn))

The function executes correctly:
['FacetGrid', 'JointGrid', 'PairGrid', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '__version__', '_orig_rc_params', 'algorithms', 'axes_style', 'axisgrid', 'barplot', 'blend_palette', 'boxenplot', 'boxplot', 'categorical', 'catplot', 'choose_colorbrewer_palette', 'choose_cubehelix_palette', 'choose_dark_palette', 'choose_diverging_palette', 'choose_light_palette', 'clustermap', 'cm', 'color_palette', 'colors', 'countplot', 'crayon_palette', 'crayons', 'cubehelix_palette', 'dark_palette', 'desaturate', 'despine', 'distplot', 'distributions', 'diverging_palette', 'dogplot', 'external', 'factorplot', 'get_data_home', 'get_dataset_names', 'heatmap', 'hls_palette', 'husl_palette', 'jointplot', 'kdeplot', 'light_palette', 'lineplot', 'lmplot', 'load_dataset', 'lvplot', 'matrix', 'miscplot', 'mpl', 'mpl_palette', 'pairplot', 'palettes', 'palplot', 'plotting_context', 'pointplot', 'rcmod', 'regplot', 'regression', 'relational', 'relplot', 'reset_defaults', 'reset_orig', 'residplot', 'rugplot', 'saturate', 'scatterplot', 'set', 'set_color_codes', 'set_context', 'set_hls_values', 'set_palette', 'set_style', 'stripplot', 'swarmplot', 'utils', 'violinplot', 'widgets', 'xkcd_palette', 'xkcd_rgb']

Joe Carter · Answer

The accepted answer to my post works correctly for Linux, and works for Windows as well, with one small tweak.
For Windows users that stumble across this post, here is what I had to do to create my own layer:
The reason Docker was not running properly on my computer was caused by the version of Windows 10 I had. I had Windows 10 Home, and I needed Professional or Enterprise to run Docker properly. After I upgraded to Windows Professional I was able to properly install Docker.
With Docker properly installed, I was then able to run @Marcin's command.
But since this is for Windows, I needed to change "$PWD" to "%cd%".
Like this:
docker run -v "%cd%":/var/task "lambci/lambda:build-python3.7" /bin/sh -c "pip install --no-deps -r requirements.txt -t python/lib/python3.7/site-packages/; exit"
After running this command, you should instantly be prompted by docker to allow file access to the specified directory. Allow it to let the command complete.
Once the command completes, be sure to zip just the folder titled python. Since this is for Windows, I don't believe @Marcin's zip command will work. But any zipping software is sufficient. I used 7-Zip just fine. Right-click on the python folder, select "add to python.zip" and voila! your zipped layer is ready to be added to AWS Lambda.

How do I create an AWS Lambda layer for seaborn?

2 Answers

Add your own answers!

Ask a Question