At the bottom of the Data Collections home page, you can find a Popular Datasets section containing the popular datasets in your organization’s Dataiku instance.
Popular Datasets are datasets that are considered the most relevant for reuse or publication to data collections, workspaces, or feature stores.
If you have the relevant permissions, you can use a popular dataset in your own projects or publish it into a Workspace, a Data Collection, or the Feature Store.
A dataset is considered popular if it satisfies the following conditions:
It has a recent last build date.
It has been shared with multiple projects.
It is used in at least one downstream recipe in a project it is shared with, and that recipe has been run at least once.
Optionally DSS administrators can strengthen these conditions by requiring a dataset to be trending, or part of a least one Data Collection.
DSS administrators can enable or disable Popular Datasets and tune the settings used for the computation.
To configure Popular Datasets, go to Administration > Settings > Misc.
The following parameters can be configured to drive the conditions a popular dataset must fulfill:
Max # days since last rebuild
The maximum number of days since the last build of your dataset. This parameter cannot be set to 0.
Max # days since last used by a new recipe
The maximum number of days since the dataset has had a new downstream recipe created in a shared project. This parameter cannot be set to 0. This parameter is also used when checking whether a dataset is trending.
Min # shares
The minimum number of projects a dataset must be shared with to be considered popular (excluding the source project). This parameter cannot be set to 0.
Only from data collections
If true, only consider a dataset as popular if it is part of at least one Data Collection.
Only trending datasets
If true, only consider a dataset as popular if it is trending. Trending datasets refer to datasets that exhibit an increasing pattern of new recipe creation over specific temporal windows, determined by analyzing historical usage data.
Popular datasets are not detected across multiple DSS instances.