page_type	languages	products	description	urlFragment
sample	python html	azure-functions	Predict ImageNet Classes with PyTorch and Azure Functions	functions-python-pytorch-tutorial

Make machine learning predictions with PyTorch and Azure Functions

Run locally

Note, the instructions below assume you are using a Linux environment.

Activate virtualenv

1. mkdir start
2. cd start
3. python -m venv .venv
4. source .venv/bin/activate

Initialize function app

1. func init --worker-runtime python
2. func new --name classify --template "HTTP trigger"

Copy resources into the classify folder, assuming you run these commands from start

1. cp ../resources/predict.py classify
2. cp ../resources/labels.txt classify
3. Add the following dependencies to start/requirements.txt, installing some numerical libraries and PyTorch itself:

azure-functions
requests
numpy==1.15.4
https://download.pytorch.org/whl/cpu/torch-1.4.0%2Bcpu-cp36-cp36m-win_amd64.whl; sys_platform == 'win32' and python_version == '3.6'
https://download.pytorch.org/whl/cpu/torch-1.4.0%2Bcpu-cp36-cp36m-linux_x86_64.whl; sys_platform == 'linux' and python_version == '3.6'
https://download.pytorch.org/whl/cpu/torch-1.4.0%2Bcpu-cp37-cp37m-win_amd64.whl; sys_platform == 'win32' and python_version == '3.7'
https://download.pytorch.org/whl/cpu/torch-1.4.0%2Bcpu-cp37-cp37m-linux_x86_64.whl; sys_platform == 'linux' and python_version == '3.7'
https://download.pytorch.org/whl/cpu/torch-1.4.0%2Bcpu-cp38-cp38-win_amd64.whl; sys_platform == 'win32' and python_version == '3.8'
https://download.pytorch.org/whl/cpu/torch-1.4.0%2Bcpu-cp38-cp38-linux_x86_64.whl; sys_platform == 'linux' and python_version == '3.8'
torchvision==0.5.0

1. Install dependencies with pip install --no-cache-dir -r requirements.txt

Update the function to run predictions

1. Add an import statement to classify/__init__.py

import logging
import json
import azure.functions as func

from .predict import predict_image_from_url

1. Replace the entire contents of the main function with the following code:

def main(req: func.HttpRequest) -> func.HttpResponse:
    image_url = req.params.get('img')
    logging.info('Image URL received: ' + image_url)

    results = predict_image_from_url(image_url)

    headers = {
        "Content-type": "application/json",
        "Access-Control-Allow-Origin": "*"
    }

    return func.HttpResponse(json.dumps(results), headers = headers)

Run the local function

1. Run func start from within the start folder with the virtual environment activated.
2. Run http://localhost:7071/api/classify?img=https://raw.githubusercontent.com/gvashishtha/functions-pytorch/master/resources/assets/Bernese-Mountain-Dog-Temperament-long.jpg

Create an Azure function

Run the following in the Azure Cloud Shell to create a sample function app with a Python runtime:

#!/bin/bash

# Function app and storage account names must be unique.
storageName=mystorageaccount$RANDOM
functionAppName=myserverlessfunc$RANDOM
region=westeurope
pythonVersion=3.7

# Create a resource group.
az group create --name myResourceGroup --location $region

# Create an Azure storage account in the resource group.
az storage account create \
  --name $storageName \
  --location $region \
  --resource-group myResourceGroup \
  --sku Standard_LRS

# Create a serverless function app in the resource group.
az functionapp create \
  --name $functionAppName \
  --storage-account $storageName \
  --consumption-plan-location $region \
  --resource-group myResourceGroup \
  --os-type Linux \
  --runtime python \
  --runtime-version $pythonVersion \
  --functions-version 2

Publish to Azure

1. func azure functionapp publish <appname> --build local
2. Test by using the suggested URL and appending &img=https://raw.githubusercontent.com/gvashishtha/functions-pytorch/master/resources/assets/Bernese-Mountain-Dog-Temperament-long.jpg at the end of the query string.

License

See LICENSE.

Contributing

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.