Skip to main content

​
Implicity datasets

The simplest way to create a dataset is to collocate a group of tasks in a directory.
which can be run with:

​
Explicit datasets

You may want to define multiple datasets containing overlapping tasks. For example, you may have a directory of tasks defining an implicit dataset that you want to subset into categories. In this case, copying and pasting the tasks is inconvenient, inefficient, and adds maintenance overhead. Instead, you should create a dataset.toml manifest, which contains pointers to the task directories.

​
Create a dataset.toml manifest

To create a dataset.toml manifest, run
which outputs a dataset.toml manifest in the current directory.
Usually, <org> is the name of your company and <name> is the name of the dataset.

​
Add and remove tasks

To add tasks to the dataset, run
To remove tasks from the dataset, run
You can also add and remove all tasks from another dataset

​
Run the dataset.toml

To run the dataset.toml manifest, run

​
Publish a dataset

To share your dataset with team members or publicly using Harbor Hub, run
Think of Harbor Hub like PyPI or NPM rather than GitHub. Because tasks are software, development typically happens in a version-controlled repository and versions are published to Harbor Hub. Once a dataset is published, anyone with access can run it using the harbor run -d <org/name> command.