I wonder why they decided to get rid of MaxoutDense. Is there something better or is it so trivial to implement they decided to drop it?
It's trivial with 'max' merge param: https://github.com/fchollet/keras/pull/3128
Introducing Keras 2
11–20 of 70 posts
Re: Introducing Keras 2
#12Earlier quoted context omitted.
To put this quote in context: this isn't specifically about PyTorch. Every couple of months since mid-2015, a new deep learning framework gets released. In the following week, someone inevitably asks "will X get added as a Keras backend?". Supporting several backends is a strong positive. But chasing every new framework as a backend is a quick way to kill Keras, via bloat, support issues and general technical debt. W…
Yay, making deep learning uncool again! ;) ( http://www.fast.ai/about/ ) But seriously - does it even make use to have a "define by run" dynamic framework as a backend? It seems to me that keras is particularly suited to wrapping frameworks that define and run a computation graph.
With the functional API of Keras, it would definitely make sense. In fact I do think that imperative model definition would be great to have at some point in the future. We'll see :)
Re: Introducing Keras 2
#13Highly recommend this course - http://course.fast.ai/ that utilizes Keras as the main programming tool
I'm definitely going to give this a shot, thanks for the link. Approaching ML at a higher level is exactly what I need to develop a better interest in it. I realize that underpinnings are important, but waiting 30 minutes for mnist on to process on my localhost is just unbearably boring.
If you try the course, be sure to make use of the forums for it too: http://forums.fast.ai . As you'll see, they're extremely active and helpful for all deep learning students (and all practitioners in general).
Disclaimer: I teach the course. Although it is free and ad-free... :)
Re: Introducing Keras 2
#14Earlier quoted context omitted.
Yay, making deep learning uncool again! ;) ( http://www.fast.ai/about/ ) But seriously - does it even make use to have a "define by run" dynamic framework as a backend? It seems to me that keras is particularly suited to wrapping frameworks that define and run a computation graph.
With the functional API of Keras, it would definitely make sense. In fact I do think that imperative model definition would be great to have at some point in the future. We'll see :)
I'm intrigued!... The kernel calling overhead and lack of any GPU while/scan/map/etc for Pytorch seems like a limitation, but I guess on 2nd thoughts you can still do all the keras fit/predict stuff and auto-connecting up the layers.
Re: Introducing Keras 2
#15The mathematician in me has kept me from jumping into deep learning before I understand the mathematical and statistical underpinnings of the algorithms involved. Looking forward to reading through the latest book out by mit press and giving things a whirl with Keras which I've heard so much about.
what's the name of the book?
Re: Introducing Keras 2
#16The mathematician in me has kept me from jumping into deep learning before I understand the mathematical and statistical underpinnings of the algorithms involved. Looking forward to reading through the latest book out by mit press and giving things a whirl with Keras which I've heard so much about.
The only "math" in deep learning is given by reverse mode AD (or if you're into fancy stuff, "efficient computing of pullbacks"). The rest of it plain old hacking, and empirical tricks with the occasional variational doodads.
Re: Introducing Keras 2
#17Awesome. Yet "codebases written in Keras 2 next month should still run many years from now" given that deep learning is no new, how can they that confident that this API will remain relevant years down the line?
What does one thing have to do with the other? Regardless of how "relevant" Keras 2 stays years from now, code written with it now should still be capable of running then, that's the thing they claim.
Re: Introducing Keras 2
#18The mathematician in me has kept me from jumping into deep learning before I understand the mathematical and statistical underpinnings of the algorithms involved. Looking forward to reading through the latest book out by mit press and giving things a whirl with Keras which I've heard so much about.
what's the name of the book?
Re: Introducing Keras 2
#19The mathematician in me has kept me from jumping into deep learning before I understand the mathematical and statistical underpinnings of the algorithms involved. Looking forward to reading through the latest book out by mit press and giving things a whirl with Keras which I've heard so much about.
The only "math" in deep learning is given by reverse mode AD (or if you're into fancy stuff, "efficient computing of pullbacks"). The rest of it plain old hacking, and empirical tricks with the occasional variational doodads.
You clearly haven't read the book.
Re: Introducing Keras 2
#20I love Keras but I think this update broke more things than you realized. For example it's no longer possible to get the validation set score (val_acc) during training which renders early stopping impossible. This was a documented feature on your FAQ.
Is the old documentation still available? I'd like to wait before I upgrade.
Edit:typo