Number of latent dimensions cannot be changed


I try to train a model with more than 2 latent dimensions (in v4.3.1). So, I increase the value to let’s say ‘3’, but as soon as I queue the job it jumps back to ‘2’ and the job will only train 2 latent dimensions. Is this a bug or or do I need to change another parameter at the same time?

Before I start (this example is going to crash since it sees a mismatch between the 3 components from 3DVA and the latent dimensions it tries to train):

After crash:

from the ‘Inputs and parameters’ tab of the crashed job

In this particular example I tried using 3DVA components (for better illustration), but it happens in the same way if doing it the default way.


@Moritz Please can you email us the job report for this job.

Hi, sorry, it took me a while to get back to this. In the meantime we upgraded to v4.4.0, and the error/behavior is gone. Specifying to train 3 latent dimensions now works as expected!