About updates in adadelta

[MilesQLi]

I think in the original paper of adadelta, there is no step_rate.

[f0k]

That's right. Feel free to add a note to the documentation. In Lasagne, we have (a bit sloppy): "In the paper, no learning rate is considered (so learning_rate=1.0). Probably best to keep it at this value."