Is anyone familiar with neural networks around? I am new to neural nets, as well as to using them from Mathematica, and I am regularly running into trouble with specifying the inputs and outputs correctly.

@kirma Thank you very much!

@Jacccy Great if it helped. :)

@CarlLange When I use a ValidationSet in NetTrain, it will automatically return the net with the lowest validation loss. Is this right? Can I also get the net after the final training round? Also, can I control how exactly a final net is selected (based on loss, error rate, etc.)? This must have to do with TrainingProgressMeasurements and TrainingStoppingCriterion, but it's unclear how they affect the selection of the final result.

@Szabolcs I actually don't know - I haven't done a lot with the NN framework in 12, and a lot of changes went into how the training works.

In version 11.3, I was pretty sure it would return the net after the final training round, rather than net with lowest loss

Anybody know if there's a secret way to get notebooks to store large images, plots etc out-of-core somehow? Large (multi-MB) plots kill my notebook

Given a Dataset named ds, I can plot foo against bar^2 as ListPlot[ ds[All, {#foo, #bar^2}&] ]

Is there an easy way to map a 3rd column "abc" to the colour of points?

I can use Style, but it feels clumsy ...

Use span for each letter making the output html super long. And this did not happen in live rendering.

@b3m2a1 It seems that using the style and position to keep html and nb rendering into the same result. Wel, How to get the position and size from the nb in mathematica?

@GalAster what do you mean? I guess you're right I could try to statically render using NodeJS and ExternalEvaluate or something and then dump that into my Markdown