Building autoencoders in Keras

[hooloovoo_zoo]

What a lovely tutorial. It should also be noted that autoencoders are useful for supervised learning as feature generator for more effective (problem specific) techniques than NNs like GBMs.

[nomailing]

Could you please elaborate on this. I would really like to know if autoencoders are still useful for classification if I have only labels for a small part of my training data. Is unsupervised pretraining still useful or was it completely replaced by other techniques as the article somehow seems to suggest?

[isseu]

I don't understand the image denoising. He put noise in the images, but never use them in the code.. It's an error from the author or I missed something?

edit: Author fixed it

[glial]

I hadn't heard of Keras, but it looks easy to use. Can anyone comment on how it compares to the other deep learning helper packages out there?

[ogrisel]

It's both simple to use and very easy to customize to build ad-hoc architectures with custom nodes. The development is very active and it's well documented.

It can also use either tensorflow or theano as runtime.

[callesgg]

if you just want it and dont want to build it: http://getnarrative.com/

[msandford]

This has nothing to do with autoencoders.

[callesgg]

Posted that on wrong link :/