Pull requests about nolearn HOT 5 CLOSED

Pull requests are welcome. If you have something you'd like to improve, having code ready that implements the change helps a lot, and it's more likely it'll be done sooner (once there's consensus that the general idea is good -- you might wanna ask if unsure). You'll probably want to add a test for a new feature.

from nolearn.

Daniel
I already have in my nolearn dev, and am using a function to load weights by layer type.
I wanted to add that if it is helpful to others.

from nolearn.

I've added a little comment to the tutorial where you've mentioned the same thing; I think the recent addition of a 'name' for Lasagne layers is good news for us, and it means that layer weights can be copied maybe more reliably than by type. Check my comments there. I think it'll be useful to see your code anyway, also to see how you've implemented the AE, something that others might want to try as well.

from nolearn.

Here it is https://gist.github.com/run2/e7162610e759a3453156. It is nothing fancy. I can improve it further if it is helpful to others

from nolearn.

Thanks for sharing this. I think it'll ultimately be more useful to match layers by name here. Lasagne just recently added names to layers. We should use these, and then copying weights becomes just a matter of matching names.

from nolearn.