Tutorial

In part 1, we talked about effective management and versioning of large datasets and the creation of reproducible ML pipelines. Here we'll learn about experiment management: generation of many experiments by tweaking configurations and hyperparameters; comparison of experiments based on their performance metrics; and persistence of the most promising ones

In most cases, training a well-performing Computer Vision (CV) model is not the hardest part of building a Computer Vision-based system. The hardest parts are usually about incorporating this model into a maintainable application that runs in a production environment bringing value to the customers and our business.

In this guide we will show how you can use CML to automatically retrain a model and save its outputs to your Github repository using a provisioned AWS EC2 runner.

We're going to look at how you can prevent stale models from remaining in production when the data starts to differ from the training data.

Sharing experiments with teammates can help you build models more efficiently.

You can easily make changes to your dataset using DVC to handle data versioning. This will let you extend your models to handle more generic data.

You can work with pretrained models and fine-tune them with DVC experiments.

Using DVC, you'll be able to track the changes that give you an ideal model.

DVC 1.0 optimized data synchronization to and from remote storage. Here's how we did it.