Keras package for AWS Lambda

This is a reference application for Inference with Keras on AWS Lambda along with deployable package dependencies

This repo is an extension of ryfeus’s Keras Lambda pack. It seemed like the repo was created for doing training on Lambda, which isn't the ideal use case, hence porting it to do Inference.

Instructions

• I love doing everything in Jupyter these days, hence providing a notebook to code, deploy and test using Jupyter. Take a look at the keras-lambda-notebook.ipynb. If you'd still prefer the traditional way, look below

• Create a Lambda function from the CLI by running the following commands:

cd keras-lambda/src
zip -9r lambda_function.zip  * 
aws lambda update-function-code --function-name keras-lambda-app --zip-file fileb://lambda_function.zip

• Test the Lambda function:

aws lambda invoke --invocation-type RequestResponse --function-name keras-lambda-app --region us-west-2 --log-type Tail --payload '{"url": "https://images-na.ssl-images-amazon.com/images/G/01/img15/pet-products/small-tiles/23695_pets_vertical_store_dogs_small_tile_8._CB312176604_.jpg"}' output

• Update the Lambda function

aws lambda update-function-code --function-name keras-lambda-app --zip-file fileb://lambda_function.zip  --region us-west-2

Creating the package on your own

I didn’t find any inference specific ports of Keras Tensorflow on Lambda, hence decide to create this package

: You can reproduce this by doing the following canonical steps

wget https://github.com/ryfeus/lambda-packs/tree/master/Keras_tensorflow/Pack.zip

pip install h5py 
pip install pillow

# replace function param
sed -i -e 's/require_flatten/include_top/g' keras/applications/imagenet_utils.py

# Lets make those binaries lean
find ./ -name “*.so” | xargs strip

# Also remove those .pyc files
find . -name \*.pyc -delete

# Deploy on Lambda