Aditya Raman

Nov 30, 2020

8 min read

Seasons of Serverless, Lovely Ladoos

Seasons of Serverless

Challenge 2: Lovely Ladoos

It’s Diwali season in India! A very popular delicacy that Indians eat during Diwali are ladoos, balls of flour dipped in sugar syrup. Can you create a machine learning model to detect a proper ladoo? Learn more about this challenge.

Teaser

Solution

  1. The Solution creates an Image Classifier using Custom Vision API
  2. The solution also creates a Serverless Endpoint which is used to
  • List all the images stored on the Azure Blob Storage
  • Upload a new image on the Azure Blob Storage
  • Predict the already uploaded image using Custom Vision API

Serverless Compute

  • Function as a Service: Modular way of writing business logics as a different application(s) which is/are triggered with the event(s). eg., Azure Functions
  • Stateless Compute Containers​: Service which does not require to read/store their previous states after each start.
  • Event-triggered and Ephemeral​: Triggered only by events, and they exist until the time it is triggered. If not triggered for a period, the service gets down allowing other services to use their resources.
  • Dynamically Allocated​: Services gets down when not in use so their resources might be used by other services, that is why they have to be redeployed(behind the scene) thus dynamically allocated.
  • Pay as you go​: Pay for the services that have actually been used.

Advantages of using Serverless

  • Lower Operational Cost​
  • Easily Scalable​
  • Billing is based upon Usage​
  • Easy Deployment​
  • Low Cost​

Disadvantages

  • Complex Architecture​
  • Execution Time​
  • Execution Frequency​

Azure CLI

Azure CLI Commands

az login
az account list-locations \
    --query "[].{Region:name}" \
    --out table
# To see all the available subscriptions
az account list # To set the subscription
az account set --subscription <Subscription-ID or Subscription-Name>

Create Azure Resource Group

az group create --name <Your-Resource-Group-Name> --location eastus

Blob Storage

az storage account create \
    --name <Blob Storage Name> \
    --resource-group <Resource Group Name> \
    --location eastus \
    --sku Standard_ZRS \
    --encryption-services blob

Create Container

az ad signed-in-user show --query objectId -o tsv | az role assignment create \
    --role "Storage Blob Data Contributor" \
    --assignee @- \
    --scope "/subscriptions/<Subscription>/resourceGroups/<Resource Group Name>/providers/Microsoft.Storage/storageAccounts/<Blob Storage Name>"az storage container create \
    --account-name <Storage Account> \
    --name <Container Name> \
    --auth-mode login

Azure Functions

Creating Functions Project

# This will generate a folder containing two files host.json and local.settings.json
func init Seasons-of-Serverless-Solution-Lovely-Ladoos --worker-runtime nodecd Seasons-of-Serverless-Solution-Lovely-Ladoos

Creating Functions Template

# Here a pre-built template for the Http Trigger will be created
# with two files, function.json and index.js
# We are going to write our logics in the index.js# To create blob endpoint
func new --template "Http Trigger" --name blobs# To create prediction endpoint
func new --template "Http Trigger" --name predict  # To create blob endpoint

Initialize your nodeJs project with npm

npm init

List all the blobs in a Blob Storage Container

const { BlobServiceClient, StorageSharedKeyCredential, newPipeline } = require('@azure/storage-blob');const sharedKeyCredential = new StorageSharedKeyCredential(
  process.env.AZURE_STORAGE_ACCOUNT_NAME,
  process.env.AZURE_STORAGE_ACCOUNT_ACCESS_KEY
);const pipeline = newPipeline(sharedKeyCredential);
const containerName = process.env.CONTAINER_NAME;const blobServiceClient = new BlobServiceClient(`https://${process.env.AZURE_STORAGE_ACCOUNT_NAME}.blob.core.windows.net`, pipeline );const containerClient = blobServiceClient.getContainerClient(containerName);const listBlobsResponse = await containerClient.listBlobFlatSegment();

Uploading the image to the Blob Storage

const { BlobServiceClient } = require('@azure/storage-blob');
const streamifier = require("streamifier");const containerName = process.env.CONTAINER_NAME;const blobServiceClient = BlobServiceClient.fromConnectionString(process.env.AzureWebJobsStorage);const containerClient = blobServiceClient.getContainerClient(containerName);;const blockBlobClient = containerClient.getBlockBlobClient(fileName);const result = await blockBlobClient.uploadStream(streamifier.createReadStream(Buffer.from(fileData)), fileData.length);

Predict the image with image URL

// Import Dependencies
const msRest = require("@azure/ms-rest-js");
const { PredictionAPIClient } = require("@azure/cognitiveservices-customvision-prediction");// Get the env variables
const projectId = process.env.PROJECT_ID
const publishedName = process.env.PUBLISHED_NAME
const predictionKey = process.env.PREDICTION_KEY
const endPoint = process.env.PREDICTION_ENDPOINT
const predictionResourceId = process.env.PREDICTION_RESOURCE_IDconst predictor_credentials = new msRest.ApiKeyCredentials({ inHeader: { "Prediction-key": predictionKey } });const predictor = new PredictionAPIClient(predictor_credentials, endPoint);const results = await predictor.classifyImageUrl(projectId, publishedName, { url: imageUrl });

Custom Vision API

  • Log in the Azure Portal, search for Cognitive Services → Custom Vision or search for Custom Vision, and click Create.
Create Custom Vision
  • Select the subscription, resource group you created above.
    Select the pricing tier best suited for your purpose, well here I would go with Free F0 Pricing Tier.
  • It takes a few minutes to deploy
  • Navigate to ‘Go to prediction resource’ ie., customvision.ai portal
  • Create a New Project
Name: <Any Name>
Description: <Any Description>
Resources: <The Resource just created>
Project Types: Classification
Classification Types: Multiclass (Single tag per image)
Domains: Food
  • On the Custom Vision Portal, click on ‘Add Images’, and upload the images of “ladoos” and tag them as “ladoo”. At the end Upload all the files.

For your simplicity, I have attached the dataset: ramanaditya/Seasons-of-Serverless-Solution-Lovely-Ladoos/tree/main/datasets

  • Similarly, upload the images of doughnut and sesame buns.
  • Once all the images are uploaded click on Train, you can select any training options, here I have selected “Quick Training”.
  • Once the training is completed you will find the beautiful statistics
  • Click on “Publish” to publish your prediction model to be used in the Azure Functions.
  • While publishing save the “Published Name” to your .env file
  • Navigate to settings and store Project ID, Prediction Endpoint, Prediction Key and Prediction Resource ID in the .env file.

Run locally

func start

Check the working of API Endpoints

GET: /api/blobs

list

POST: /api/blobs

upload

GET: /api/predict?imageurl=<imageUrl>

prediction

Chefs for the Challenge

  1. Jasmine Greenaway, Cloud Advocate (Microsoft)
  2. Rama Soumya Naraparaju, Microsoft Student Ambassadors
  3. Aditya Raman (me), Microsoft Student Ambassadors

About the Author

Aditya Raman
Seasonsofserverless
Serverless
Microsoft
Azure
Nodejs

--

--

More from Aditya Raman

Back End Developer | Software Engineer | DevOps Engineer | Data Science | Microsoft Learn Student Ambassador | Mentor MLH | Full Stack Developer

AboutHelpTermsPrivacy

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Aditya Raman

Aditya Raman

20 Followers

Back End Developer | Software Engineer | DevOps Engineer | Data Science | Microsoft Learn Student Ambassador | Mentor MLH | Full Stack Developer

Help

Status

Writers

Blog

Careers

Privacy

Terms

About

Text to speech