0.0.26

Full Changelog: 0.0.23...0.0.26

0.0.25

Full Changelog: 0.0.23...0.0.25

0.0.24

Full Changelog: 0.0.23...0.0.24

0.0.23

Full Changelog: 0.0.22...0.0.23

0.0.22

more magic

0.0.21

only increment step when training

0.0.20

can update less frequently

0.0.19

handle losses passed in as namedtuple. also validate the loss names b…

…eing passed in

0.0.18

accept tuple of losses too

0.0.17

loss weights should be settable for frozen case