everdark / sagemaker-playground

Experiments with AWS SageMaker Python SDK.

SageMaker Playground

The repository contains codes playing around with AWS SageMaker.

Local Mode

Deploy a pre-trained scikit-learn model (NOT trained by SageMaker SDK) to an local endpoint. This is especially useful when testing the user module for model inference task. The endpoint will be run by a docker container listening locally for HTTP request for model invocation.

# train a model
./train-withou-sagemaker.py

# package the resulting model
make model-package

# deploy the model
./deploy

To test model inference with application/json type:

curl --location '127.0.0.1:8080/invocations' \
--data '{
    "sepal_length": 1,
    "sepal_width": 1,
    "petal_length": 1,
    "petal_width": 1
}'

Or text/csv type:

curl --location '127.0.0.1:8080/invocations' \
--header 'Content-Type: text/csv' \
--data 'sepal_length,sepal_width,petal_length,petal_width
1,1,1,1'

Or application/jsonlines:

curl --location '127.0.0.1:8080/invocations' \
--header 'Content-Type: application/jsonlines' \
--data '{"sepal_length": 1, "sepal_width": 1, "petal_length": 1, "petal_width": 1}
{"sepal_length": 1, "sepal_width": 1, "petal_length": 1, "petal_width": 1}'