should mlr enable predict.type ="scores"

[MinhAnhL]

In Anomaly detection most algorithm return scores instead of probabilities or binary response variables. E.g. the autoencoder from h2o package (see issue #1820). We can turn the scores into probabilities with Platt's method.

Bernd and I already discuss a little bit about these two questions:

1. Do we need to keep the scores in case users prefer scores instead of probabilities?
2. If yes, any ideas how to do it?

[berndbischl]

this is a larger change to the core package and i would rather not do this now

[MinhAnhL]

For the Area under the Mass Volume Curve (AMV), we would need the scores. Need to find out if probability scores with Platt scaling would still work for AMV. https://drive.google.com/file/d/0B8Dg3PBX90KNUTg5NGNOVnFPX0hDNmJsSTcybzZMSHNPYkd3/view

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.