/TfLambdaDemo

TensorFlow AWS Lambda Demo

Primary LanguagePythonMIT LicenseMIT

TfLambdaDemo

TensorFlow AWS Lambda Demo

This project is inspired by mikepm35/TfLambdaDemo, and its original article is posted here.

How to Setup

  1. Install serverless npm package via

    npm install -g serverless

  2. Setup a AWS serverless environment via

    serverless create --template aws-python.

  3. In order to run TensorFlow, we need to assign a specific version (3.6.5) of Python in its virutal environment by typing

    virtualenv --python=python3.6.5 tflambdademo && source tflambdademo/bin/activate

  4. For this demo, we are going to pick TensorFlow version 1.13.1. We also can chooser other versions of TensorFlow. However, AWS Lambda has a limitation of Unzipped size must be smaller than 262144000 bytes.

    For my understanding. v.1.13.1 is a smaller size version. Newer versions of TensorFlow framework are usually larger ones, that might cause us can't upload files to the server. Choosing a version of TensorFlow by inserting the following command,

    pip install tensorflow==1.13.1 && pip freeze > requirements.txt.

  5. Add nodeploy Python packages into the serverless.yml, also add exclude for the packages which you don't want to be uploaded to the server.

    package:
  exclude:
    - node_modules/**
    - tflambdademo/**

    After that, your unzip file size will be dropped dramatically, and you should be able to upload this zip file to AWS and without the unzip file size is too large to upload problem.

  6. Deploy your code via a serverless command. (you must lauch your Docker app in advance.)

    serverless deploy -v

How to run

  1. After deploying to AWS Lambda, you will get your server id as below,

    POST - https://<id>.execute-api.us-west-2.amazonaws.com/dev/upload
POST - https://<id>.execute-api.us-west-2.amazonaws.com/dev/train
POST - https://<id>.execute-api.us-west-2.amazonaws.com/dev/infer

  2. Use the server id to communicate with your AWS Gateway APIs. First of all, we can start to upload our training data to AWS.

    curl -X POST https://<id>.execute-api.us-east-1.amazonaws.com/dev/upload

    And, it will response a folder prefix defined by the epoch. {"epoch": "1556995xxx"}

  3. Ask AWS Lambda to traing our data from the folder.

    curl -X POST https://<id>.execute-api.us-east-1.amazonaws.com/dev/train -d '{"epoch": "1556995xxx"}'

  4. In the final step, we will ask AWS Lambda to execute an infer based on our training result.

    curl -X POST https://<id>.execute-api.us-east-1.amazonaws.com/dev/infer -d '{"epoch": "1556995xxx", "input": {"age": ["34"], "workclass": ["Private"], "fnlwgt": ["357145"], "education": ["Bachelors"], "education_num": ["13"], "marital_status": ["Married-civ-spouse"], "occupation": ["Prof-specialty"], "relationship": ["Wife"], "race": ["White"], "gender": ["Female"], "capital_gain": ["0"], "capital_loss": ["0"], "hours_per_week": ["50"], "native_country": ["United-States"], "income_bracket": [">50K"]}}'

    The result will look like this

    [{"logits": [1.088104009628296], "logistic": [0.7480245232582092], "probabilities": [0.25197547674179077, 0.7480245232582092], "class_ids": [1], "classes": ["1"]}]