Jeff Pasternack

WRT handlers/listeners, these are very much supported. To use these: (1) Subclass the `NeuralNetwork` class and override the `getPreparer(...)` method to return a new `Preparer` subclass you'll also define. (2)...

Do you have an example for the use of a per-label weights-array in DL4J? Most models in Dagli support per-example weights, but not neural networks, precisely because DL4J did not...

Thanks--so per-label weightings are certainly possible then. Unfortunately, this wouldn't be quite so trivial to implement in Dagli because we'd need to communicate the mapping of labels to indices to...

Incidentally, another (inconvenient) workaround would be to rephrase your multilabel problem by using a different binary NNClassification corresponding to each label, which you could then weight as desired. It'd be...

Thanks for reporting this. The `weights(...)` method itself does what's intended--it returns a vector of length `_inputWidth` (also the output width). Unfortunately, on investigation I discovered that ND4J considers a...

(You can prevent this corner case by adding two arbitrary "dummy examples", one with all "true" labels and one with all "false", but that strikes me as even _more_ hacky).

Thanks for the link regarding the fix in DL4J; you *might* be able to thus sidestep this by compiling their latest code and importing that into your project to replace...

This workaround is from quite a while ago, but I assume we discovered this issue ourselves. Reporting this bug to the DL4J team would have been the nice thing to...

Apologies for the late response, and thank you for raising this issue. I think the underlying problem here is that, at high thresholds, you're in the very unusual position where...

Hi--unfortunately, I do not speak Chinese. If possible, could you please restate your issue in English?