ML Microservice in Torch

Simple Waffle web server implementing the following REST API.

Introduction

The Torch scientific library has some exceptional ML libraries. To facilitate using them I have written a simple REST API that should work for most ML models, as well as a Docker container around the whole thing.

Usage

Docker

Build: docker build -t lua-ml-microservice . Run: docker run -it lua-ml-microservice

Host

Run: th app.lua

Configure

Edit app.lua to add your AWS credentials and handlers for your ML model.

local bucket = s3:connect{
  awsId="<Your AWS ID here>",
  awsKey="<Your AWS Secret Key here>",
  bucket="<target bucket name here>",
}

And to handle each file passed in the POST request edit ml.lua:

function M.handleTrainFile(file)
  -- TODO: Placeholder for real ML processing per row
  -- Dear ML Engineer who's gonna read this,
  -- Feel free to train your model inside this function.
  -- Each file will be received, printing them will show
  -- their format.
  print(file);
end

REST API

POST /train

Body:

{
  files: [ <S3 file path>, ...]
}

Example:

{
  files: [ 32AD86AB68CB4368830506D352772A12/bluetooth/490569736771195.csv ]
}

Error: 500

POST /input

Body:

{
  files: [ <S3 file path>, ... ]
}

Error: 500

POST /output/

Body:

{
  files: [ <s3 file path>, ... ]
}

Response:

{
  output: { output vector }
}

Error: 500