AWS Lambda with TensorFlow

This is a demo that integrates Tensorflow with AWS Lambda.

Requirements

• An AWS account

Get started

Once inside this example's directory, install Serephem.

$ npm i

Deploy

The Serephem plugin will take care of retrieving all the TensorFlow dependencies and injecting them into the Serverless Lambdas zip artifact prior to deploying to AWS.

Make sure you configure the serverless.yml file to your necessities before deploying (i.e. region, stage, profile...)

$ sls deploy -v

Remember to run sls remove when finished with this demo.

Lambdas

This project features 2 Lambdas:

• A TensorFlow addition of two numbers
• A NumPy that prints properties of a matrix

Tensorflow add

File

tensorflow_test.py

Description

It uses the TensorFlow library to add two integer numbers.

Input

Two integer numbers a and b. The Lambda test event JSON looks as follows:

{
  "a": 30,
  "b": 90
}

Result

The Lambda will add those numbers using the tf.add function. The result looks as follows:

{
  "message": "TensorFlow \"add\" function test: 30 + 90",
  "result": "120"
}

NumPy matrix properties

The TensorFlow package includes NumPy, since it is a dependency.

File

numpy_test.py

Description

Creates a 2x3 matrix of integers and returns some of its NumPy properties.

Input

None.

Result

{
  "message": "NumPy test",
  "event": {},
  "attrs": {
    "ndim": 2,
    "itemsize": 8,
    "dtype.name": "int64",
    "size": 6
  }
}

Local testing of Serephem plugin

In case you want to use this example to locally test the Serephem plugin, you can do the following:

1. Open a new terminal window and go to the Serephem root folder

2. Run npm link

3. Return to the terminal window for the tensorflow-lambda example

4. Run npm link serverless-ephemeral

5. Run sls package -v to see your changes