recommendation

🎥🎞🍿 A movie recommendation example using matcha

In this example, we'll show you how to use matcha to setup a default cloud environment on Azure and hook up a movie recommendation pipeline to run on that environment.

If you're wondering what on earth matcha is (besides the drink) then check out our main repository here and our documentation - don't forget to come back to try out this example!

🚦 Getting Started

There's a bit of a setup required before unleashing matcha, the steps below will guide you through this.

✅ Pre-requisites

Before you start, this example workflow requires the Azure CLI to be installed. See here for how to do that.

You will also need to ensure you have installed Docker and that the Docker daemon is running on your machine.

Finally, you also need to ensure that you have Terraform installed on your machine.

🧬 Cloning

Before you start, this example workflow requires the Azure CLI to be installed. See here for how to do that.

You will also need to ensure you have installed Docker and that the Docker daemon is running on your machine.

Clone this repo:

git clone git@github.com:fuzzylabs/matcha-examples.git

Go to the recommendation example directory:

cd matcha-examples/recommendation

Log into Azure via your terminal:

az login

Create a virtual environment:

python3 -m venv venv 
source venv/bin/activate

There is a requirement for the Python version being used to be 3.8+. We recommend making use of pyenv to manage your versions.

Install matcha:

pip install matcha-ml

Provisioning

You need to be in the recommendations directory before running this!

matcha provision

Once that's finished, crack on!

Set up the environment:

This will install the requirements for the example (see requirements.txt) and setup ZenML:

./setup.sh

You may need to give the setup.sh file the correct permissions to run, if so then do the following: chmod +x setup.sh.

▶️ Running the example

Once setup.sh has completed, do the following to run the training pipeline:

python run.py --train

Once training has finished, we can deploy our trained model by doing the following:

python run.py --deploy

We can also run both training and deployment with one command:

python run.py --train --deploy

[Optional] Run the tests:

python -m pytest tests

🙋‍♀ Query the deployed model

✅ You've trained a model

✅ You've deployed it

❓ And now you want to get predictions.

We've created a handy inference script which you can use to send a user_id and a movie_id to the deployed model get a predicted rating:

python inference.py --user 100 --movie 100

And the output should be something similar to:

User 100 is predicted to give the movie (100) a rating of: 4.2 out of 5

Alternatively, you can curl the endpoint with the following:

curl -XPOST -H 'Content-Type: application/json' -d '{"data": {"ndarray": [{"iid": "302", "uid": "196"}]}}' <endpoint_url>

The output will be the raw predictions sent back by the model!

Destroy resources

Even though we've chosen a sensible default configuration for you, leaving the resources you've provisioned in this example running on Azure is going to run up a bill.

To deprovision the resources, run the following command:

matcha destroy