Single most favorite v2 feature

For the past few days I’ve been migrating a project of mine from v1 to v2 - finally a good excuse to learn the new library in depth after watching the videos months ago :smile: )

Throughout this learning process I’ve got couple of wow moments that I found impressive, fun and easy to work with. For example, v2 is pretty similar to v1 in terms of workflow mechanics so I didn’t have much trouble during the migration and new data API is much more flexible.

Of course, this is just tip of the iceberg that I can surface with my own experience so far. Aligned with this, I thought it would be a good idea to open this thread to allow everyone to share their single most favorite feature in v2 that wasn’t straightforward or possible in v1:

Feature: 1-2 sentence
Reason: 1-2 sentence

It can potentially help others to find out about these features and how they are useful in different settings and use cases. One feature can have multiple different reasons depending on the way you use it :slight_smile:

I will go first;

Feature: splits with subset() method in DataSource.
Reason: It allows to have multiple test sets without creating additional dataloaders like it’s used to be done in v1. Gives flexibility to semantically define unlimited subsets from a single datasource; can also be useful for bootsrapping, cross-validation, active learning.