Here, we will walk you through a tutorial to use DVC Studio for collaboration on your ML projects. You will need access to a GitHub, GitLab or Bitbucket account which has access to the Git repositories you want to connect. DVC Studio creates views from repositories when you connect to them.
The following video provides you a quick overview of DVC Studio.
DVC Studio identifies datasets, metrics and hyperparameters in your Git repositories. These values can either be in DVC repositories or you can add custom files with the required values. Read more about the different ways in which you can prepare your Git repository for use with DVC Studio here.
By identifying the datasets, metrics and hyperparameters in your Git repositories, DVC Studio creates a view, which is an interactive, tabular representation of all your ML experiments. In this view, you will not only see your complete experiment history, you can also generate plots, compare experiments, and run new experiments.
In your browser, open https://studio.iterative.ai. Sign in with your Github, GitLab, or Bitbucket account.
When you first login, an example view is already created for you to explore, and you can add more views.
When you first login, you will find that there already exists a
connecting to an example DVC project. Use this view to explore the features that
DVC Studio has to offer.
DVC Studio automatically identifies datasets, metrics and hyperparameters in
your ML experiments. Each view on the dashboard displays the metrics. In the
figure above, you can see that
roc_auc metrics are displayed.
You can dive deep into all the experiments committed to the repo. For this, open
the view by clicking the view name (in this case,
A table will be generated as shown below. This includes metrics, hyperparameters and information about the datasets. All these values are flattened and neatly presented for you to evaluate and compare the experiments.
This tabular display has the following components:
Various buttons for performing actions:
You can connect to additional repositories and add more views as needed. You'll find out how to do this in the next section.