Jörg Bornschein

This way the features are still called "features" and we can use Silhouettes as a drop-in replacement for other datasets. "features" seems to be pretty standard. It's indeed a bit...

I agree -- having a log (and plot) server that just receives updates and decides what and how to plot afterwards seems better. In that case, maybe blocks should not...

Hi, thanks for the contribution! I think it would be better to give this example a separate number instead of cramping it together with the other scatter-gather example. I guess...

That sounds like a good idea. In the same vein: I started a "DetailedStepMonitor" that could be part of a CompositeStep and would create max/median/mean variables for each parameter receiving...

Correction: ... that would create max/median/mean variables for each parameterupdate per _StepRule_. I had unstable learning dynamics and it was very useful to see the magnitude of individual parameter updates...

That would be an option. I see two aspects that might make it desirable to explicitly activate this detailed monitoring: 1) these are a lot of variables: per step-rule, per...

@rizar Yes, you are right, having the AUXILIARY role does not imply they will be monitored on the test/verifikation dataset. I just wanted to point out that maybe we should...

Ouch, nice catch! I think you are right. Instead of using T.tril(W , k=-1) to extract the lower triagular matrix, it should have been T.triu(W, k=1) to extract the upper...