Visually comparable regressions - Githubissues

emorice / gemz

Gene expression model zoo

0 stars 0 forks source link

Visually comparable regressions #20

Open emorice opened 1 year ago

emorice commented 1 year ago

We have a basic regression setup, but when things change all we get is a thousand-line diff between yaml dumps.

For effective use, we need one of the following.

Directly regress files that can be visualized. This would mean for instance an html dump of the plots. The main obstacle is that plotly currently adds random ids in the file by default (blocking), and that the diffs may not be readable (minor).
Equivalent visualization of the thing. For instance a helper that could re-load the yaml serialization and generate the html from it. Then the points above would not matter much.

emorice commented 1 year ago

Update: we have a basic regression dashboard that does the second point above. Not closing because it's still not really usable when a change occurs as we can only see the reference file at the moment.