Adding preprocessing to all models

[wielandbrendel]

No description provided.

[coveralls]

Coverage remained the same at 100.0% when pulling aef287a on wielandbrendel:preprocessing into 2c4d327 on bethgelab:master.

[jonasrauber]

Great contribution, @wielandbrendel !

Could you update the Keras ResNet example in the README to use the new preprocessing?
Then it should be ready to merge and I'll publish it as version 0.9.0.

_loss_fn was only added to implement the tests, right? It should be clear that _loss_fn should only used for these internal tests and should not be relied on by users because that interface is likely to change when we generalize the predictions and gradient calculations in future versions.

[coveralls]

Coverage remained the same at 100.0% when pulling 40dd264 on wielandbrendel:preprocessing into 2c4d327 on bethgelab:master.

[wielandbrendel]

@jonasrauber Good point, I forgot about the readme. Should be fixed now. Regarding _loss_fn: this will be refactored soon anyway when we generalise the handling of the loss. Lets not bother until then.

[jonasrauber]

Regarding _loss_fn: this will be refactored soon anyway when we generalise the handling of the loss. Lets not bother until then.
@wielandbrendel That's what I meant ;-)