Use of 5 seats in Company License packs, role of multiple annotators

Hi, while installation and running Prodigy, I faced some doubts regarding the following,

  • How to use 5 seats that are provided with the license since we are using it for multiple users by running it on multiple ports. What is the use of 5 seats that are provided?

  • As we have a large number of entities can we add labels through UI or any other method instead of using command prompt ?

  • If multiple annotators are working on the same large file, will their work complement each other ? .. i.e. if some annotator is done tagging 500 lines. will that file portion be skipped for other annotators or will they still have to go through tagging the same 500 lines again, which I think will be redundant for another annotator and collaboration.

The seats are developer seats, so every developer who has access to the Prodigy Python library, CLI, back-end, database etc. will require a seat. If you have 5 seats, this means that 5 developers can work with Prodigy at the same time. Annotators who only have access to the web app do not require a seat.

Prodigy expects you to provide your label scheme upfront when you set up the annotation task – modifying them at runtime is very counterproductive, because you can easily end up with inconsistent data.

But instead of providing a comma-separated list of labels on the command line, you can also set the --label argument to a text file with one label per line.

Yes, you can configure whether everyone should see and annotate the same examples or whether everyone should see different examples using the feed_overlap setting. You can read more about this here: https://prodi.gy/docs/api-web-app#multi-user-sessions If you re-start the server, annotations that are already present in the current dataset will also be skipped.

1 Like

Thanks. I will go through the feed_overlap setting.

Our use-case as a team includes using pre-annotations from our base models, rectify annotations and improve further on the model. Does the above line also imply that the feed_overlap setting will skip the pre-annotated lines when we import our datasets?