A bash script for creating a layer on AWS Lambda (Python 3.7) that includes NLTK and the Punkt sentence tokenizer

nltk_punkt_lambda_layer.sh

LAYER_NAME=NltkPunkt
REGION=us-east-1

mkdir -p build/nltk_punkt/python

docker run -v $(pwd):/out lambci/lambda:build-python3.7 /bin/bash -c \
  "pip install nltk -t /out/build/nltk_punkt/python;
  cd /out/build/nltk_punkt/python;
  python -c \"import nltk; nltk.download('punkt', '/out/build/nltk_punkt/python')\";
  rm tokenizers/punkt.zip;
  rm tokenizers/punkt/*.pickle
  "

# add nltk data to path
INIT_FILE=build/nltk_punkt/python/nltk/__init__.py
LINE=$(awk '/from __future__/{ print NR}' $INIT_FILE | tail -n 1)  # last line importing __future__
N_LINES=$(wc -l $INIT_FILE | awk '{ print $1 }')   # number of lines in file
S1=$(head -n $n $INIT_FILE)
S2=$'\nimport os; os.environ["NLTK_DATA"]="/opt/python"'
S3=$(tail -n $((N_LINES-LINE)) $INIT_FILE)
echo "$S1 $S2 $S3" > $INIT_FILE

cd build/nltk_punkt
zip -r ../../nltk_punkt.zip python/
cd ../..
                                 
echo deploying $LAYER_NAME to $REGION
aws lambda publish-layer-version --layer-name $LAYER_NAME \
                                 --region $REGION \
                                 --zip-file fileb://nltk_punkt.zip \
                                 --compatible-runtimes python3.7

rm -rf build
rm *.zip