suboptimal model saved

[dstoekl]

When I train a model kraken first seems to run a trial validation run on a subset then a complete validation run. If I later interrupt the training process, it does not save the best model trained on the whole dataset but the small validation run if its accuracy was highest:

[dstoekl]

command used: ketos -v train -o modelname -i basemodel.mlmodel -d cuda:0 -B 30 -w 0 -r 0.0005 --lag 20 --augment --workers 1 -t train.arxlist -e test.arxlist -f binary --resize add

[mittagessen]

Fixed in main.