need help with POC

[Piotr]

Hey Vizzu team!

We would like to add ipyvizzu story plot to MLJAR AutoML reports.

MLJAR AutoML trains several ML models. For each model we have the training information, it can be single value or list of values. Example log from training:

Iteration,TrainingLoss,ValidationLoss
0,0.6224042672864691,0.63477659479385
1,0.5852765052796838,0.6083788001751151
2,0.5536278422899589,0.5859192889100233
3,0.5253922667599724,0.5673112168974941
4,0.500639561006052,0.5514805711171018

And we have table with final loss value for each model. Right now we create a static report in Markdown and HTML. Report is documentation from AutoML training. It would be fantastic to make dynamic story from AutoML training.

Here is an example report:

Typically there are from 10 to 50 models, sometimes up to 100 ...

Is it possible to animate training logs (line charts) and show them in scatter plot of all models performance? Something like:

• training model 1, animate training logs,
• add final loss to leaderboard scatter plot,
• training model 2, animate training logs,
• add final loss to leaderboard scatter plot,
• training mode 3, animate training logs
• and so on ...

What do you think @petervidos, @aplonska?

[Peter Vidos]

Hey @pplonski,
I'm sorry for not getting back to you sooner.
I'm sure this could work. I would create a Vizzu-story (JS version of ipyvizzu-story -https://github.com/vizzuhq/vizzu-story-js), so the user can switch between the views using the presentation controls.
Some sort of racing chart could be built for each training log, then the final loss of that model could appear on the leaderboard scatter plot. Whether or not these are two animated stories or just one, depends on the data structure - and also the size of the tables.
Let's set up a short call to discuss the details if this sounds like something you'd be looking for.