Model Development

How do you train models in Iguazio?

Simply put, we are orchestrating Python code. You can bring your own training code or utilize one of our pre-built training functions.

I need to build Machine Learning pipelines. What tools do you provide?

Kubeflow is integrated into our environment. It is enhanced to support our security and leverage access to data inside the cluster. In addition, we use the open-source framework MLRun for serverless function orchestration, experiment tracking, feature store, and more.

If I have a Dev, QA, and Prod environment, and I trained my model on Dev, how can I deploy my model to Prod?

It is possible to connect to Iguazio clusters externally via environment variables. Using this technique, it is simple to deploy a model from a cluster or central CI/CD environment.

Do you support ML and Deep Learning?

Yes we do. We support ML and Deep Learning frameworks such as Tensorflow, XGBoost and others.

Does Iguazio offer experiment tracking?

Yes. Using our open-source project MLRun, we enable tracking of Artifacts, inputs, outputs, code, parameters, models, metrics, hyper-parameters, data sources, plots/charts, and more. This allows users to track the lineage of a model from start to finish.

Do you support distributed training?

Yes, Iguazio runs on top of Kubernetes cluster and has built-in frameworks such as Dask and Horovod to run distributed training. MLRun makes it easy to transform your code to run on a distributed cluster along with tracking and monitoring the training jobs.

Do you support conda environment and option to do pip install?

Yes, users can create various conda environment and install any library using pip.

Can you integrate with Git?

Yes, Git can be used as the repository for saving the data scientist's code.