Review experiments

Dotscience enables users to collaborate and offer feedback directly on individual runs.

Let’s take for example, two ML engineers working on a problem. The first engineer has the following run:

The accuracy looks quite low, so they ask one of their colleagues for help. To allow other users to see your work you can add them as collaborators by heading to settings.

And scrolling down until you reach “collaborators”

You can type in the username from the search username box - this allows you to search for users in case you don’t know their exact username.

When you add collaborators they will show in this list. From the other engineer’s view point, they now see this project in their “shared with me” section of projects:

And when they come to view the project, they can see the project as a readonly user - they can browse around the project and see statistics, but won’t be able to run the project or change any settings.

Now, our second ML engineer can see the run, and can get to the same page we saw earlier:

Here they might suggest how to improve this run by leaving a comment:

Seeing Colleagues progress

Another way to view other team members’ progress is to use forks. The best way to do this is to set up the project from the first team mate’s perspective:

And then add your colleagues as collaborators:

If there’s any files that all three users would like to have initially, it’s usually a good idea to have them in this project from the start so that the other employees can use the exact same files from the beginning. I’m going to do this by opening Jupyter and using the file manager, but there are other options suggested in the Runs view.

Once added using the file browser, they should appear here:

Then let’s head back and stop Jupyter from the settings tab on the project (Click Runs -> Settings)

You should now see a run in the runs panel which confirms those files got uploaded:

Now Charlotte and Ben can fork the project by logging in, selecting the project from “shared with me”…

…and clicking “fork”…

Which takes them to a copy of the project which they can edit and run:

Let’s say Charlotte runs the project and updates the metadata using JupyterLab, resulting in her fork having some extra runs:

Checking “show all visible forks” allows us to see Charlotte’s work:

If Ben then forks the project and does some work, we can now compare their work to see how they’re doing as a team: