All recommendations disappear during training

[reckart]

All recommendations disappear during training

[reckart]

The changes in 8d3ecdd fix the problem only for the case where there is only a single recommender. If there are multiple recommenders, then it can still happen that one is finished but others are still training when a rendering happens... and it looks like that makes them disappear again.

Probably either need to change the way that predictions are stored or force all recommenders for an entire project to complete before switching predictions. Probably better the former than the latter.

[reckart]

Correction: we presently already run all recommenders for an entire project in the PredictionTask. However, instead of setting the incoming predictions at after the loop which goes over all the layers/recommender, we currently update the incoming predictions inside the loop. That is wrong.

Refactoring the code such that recommenders could run in parallel would be more effort at this time compared to what seems to be an easy fix to the current problem by just moving the setting of the incoming predictions after the end of the loop.