Franz Király

Hm, I start thinking there is a design issue here. I don't think loss funtions (or measures) should do dispatch at all on distribution types or method specific properties. Why...

>The pdfSquared2norm computes the integral from -Inf to Inf and it only depends on the training data. I'm not sure if we can call directly from the prediction The prediction...

yes, indeed: - priors ought to be hyper-parameters, and we haven't agreed on a representation, especially in the context of the sets6 discussion - composition/reduction should be compatible with mlr3pipelines,...

Regarding Bayes, perhaps it's premature to look at this at all, without thinking carefully about a Bayesian mlr interface - since the issue with priors is potentially also of relevance...

and, obviously, any suggestions for the wishlist are welcome too

>Oh and can I suggest adding some baselines? e.g. Gaussian with mean = sample mean, variance = sample var? That's a special case of two methods already there: - the...

In line with "one feature, one issue" principle (which @RaphaelS1 mentioned in communication elsewhere) - should this be split in individual issues, and the list moved to wiki? Issues can...

ok, let me know when. Just trying to comply with local best practice conventions.

quick question about the recipe: do we always need to create a separate network? Or are there a few common ones? E.g., for regression/classification? Should the regressor/classifier not use the...

So, @RakshithGB, @mloning, I'm not sure whether this is merely a question of implementation, or also of extending the PMML and ONNX specifications. PMML's [time series model specification at v4.4](http://dmg.org/pmml/v4-4/TimeSeriesModel.html)...