In this scenario, you will deploy Seldon Core to a Kubernetes environment.
You will then use Seldon Core to deploy and test a machine learning classifier- pre-trained on the Scikit-Learn Iris dataset.
An understanding of Kubernetes terminology and concepts will be helpful!
What is Seldon Core? 🤖
Seldon Core is an open source platform to deploy your machine learning models at scale on Kubernetes.
Seldon Core makes it simple to convert machine learning models into production-grade microservices following three steps:
- Containerise: Model binaries from popular frameworks (Scikit-Learn, Tensorflow, XGBoost) are readily containerised thanks to pre-built model servers. Custom models are supported using language wrappers (Python, Java) allowing you to take any model in these languages and containerise them.
- Deploy: Seldon Core extends Kubernetes by adding the custom
SeldonDeploymentresource. Seldon deployments support a range of complex inference patterns, such as canary rollouts and multi-armed bandits. Seldon deployments are built out of a number of core components e.g. Transformers, Predictors, Explainers, Routers, etc.
- Monitor: Logging can easily be configured to support tracing of network traffic to deployments, monitoring of request/response payloads, visualisation of real time model health. Alerting for outlier detection and concept drift can be setup.
✍️ Tutorial authored by Tom Farrand. Source here.
Deploying Seldon Core
Installing Helm ⛴
It's time to get your hands dirty! 🧑🔧
Prior to installing Seldon you will install Helm. Helm is a Kubernetes package manager which makes it dead easy to install other Kubernetes software with- including Seldon. 📦
Helm is straightforward to install. You will first grab the install script:
curl -fsSL -o get_helm.sh https://raw.githubusercontent.com/helm/helm/master/scripts/get-helm-3
Changing the permissions, to allow the script to be executed:
chmod 700 get_helm.sh
Running the installer:
If you are successful, you will see the following message:
helm installed into /usr/local/bin/helm