machine learning inference with AWS Lambda and Layers

bash.sh

pipenv lock -r > requirements.txt

zip -r MyMLfunction_layer1.zip python

aws s3 cp MyMLfunction_layer1.zip s3://mybucket/layers/

# layer 1
aws lambda publish-layer-version --layer-name MyMLfunction_layer1 --content S3Bucket=mybucket,S3Key=layers/MyMLfunction_layer1.zip --compatible-runtimes python3.6

# layer 2
zip MyMLfunction_layer2_model20190104v1.zip *.joblib
aws lambda publish-layer-version --layer-name MyMLfunction_layer2 --zip-file fileb://MyMLfunction_layer2_model20190104v1.zip --compatible-runtimes python3.6

# lambda
zip MyModel20190104v1_lambda_function.zip lambda_function.py
aws lambda create-function --function-name MyModelInference --runtime python3.6 --handler lambda_function.lambda_handler --role arn:aws:iam::xxxxxxxxxxx:role/MyModelInference_role --zip-file fileb://MyModel20190104v1_lambda_function.zip

# Link the layers
aws lambda update-function-configuration --function-name MyModelInference --layers arn:aws:lambda:eu-west-1:xxxxxxxxxxx:layer:MyMLfunction_layer1:1 arn:aws:lambda:eu-west-1:xxxxxxxxxxx:layer:MyMLfunction_layer2:1

export_model.py

from sklearn.externals import joblib 
# Save to file in the current working directory 
joblib_file = “MyModel20190104v1.joblib” 
joblib.dump(model, joblib_file) 
# Load from file
joblib_model = joblib.load(joblib_file)

get-layer-package.sh

#!/bin/bash

export PKG_DIR="python"

rm -rf ${PKG_DIR} && mkdir -p ${PKG_DIR}

docker run --rm -v $(pwd):/foo -w /foo lambci/lambda:build-python3.6 \
    pip install -r requirements.txt --no-deps -t ${PKG_DIR}

lambda_function.py

import base64
import json
from sklearn.externals import joblib
PREP1_FILE_NAME = '/opt/PreProc1_MyModel.joblib'
PREP2_FILE_NAME = '/opt/PreProc2_MyModel.joblib'
MODEL_FILE_NAME = '/opt/MyModel.joblib'
def predict(data):
    # Load the model pre-processors
    pre1 = joblib.load(PREP1_FILE_NAME)
    pre2 = joblib.load(PREP2_FILE_NAME)
    clf = joblib.load(MODEL_FILE_NAME)
    print("Loaded Model and pre-processors")
    # perform the prediction
    X1 = pre1.transform(data)
    X2 = pre2.transform(X1)
    output = dict(zip(data, clf.predict(X2)[:]))
    return output
def lambda_handler(event, context):
    predictList = []
    for record in event['Records']:
       # decode the base64 Kinesis data
       decoded_record_data = base64.b64decode(record['kinesis']['data'])
       # load the dynamo DB record
       deserialized_data = json.loads(decoded_record_data)
       predictList.append(json.loads(deserialized_data['textblob']))
       result = predict(predictList)
    return result

Why you put the folder in /opt/ ?