Trained model with L2 regularisation. Didn't really see much of a difference to L1. Checked everything again... turned out I had re-run the L1 setup instead of the L2. *facepalm*

When the variables and numbers start dancing....

... and suddenly everything looks similar.

... finally all kinds of errors and explosions.

Love these moments when brain cannot differentiate e.g. between hyphen and underscore, 1 or 2... Or 1 and l... Stuff like that.

And you're sitting in front of the error and you don't know what went wrong till you realize that brain played a few pranks on you. -.