Unsupervised learning using keras

[vinayakumarr]

Has any body tried to do unsupervised learning using keras. Could you please help me..i am new to keras

[marcj]

Has any body tried to do unsupervised learning using keras.

Yes. Is this your only question?

[vinayakumarr]

Yes....unsupervised learning in deep learning using keras. I know auto encoder is unsupervised...but is there any deep learning algorithm for unsupervised learning

[farizrahman4u]

You will have to be a lot more specific.

[vinayakumarr]

I have data which is of 9 class label and 1800 features. This is basically a classification. I want to apply unsupervised learning. Could you please suggest us regarding this and also is there ant basic code to do this.

[farizrahman4u]

If you have labels then why unsupervised?

[vinayakumarr]

I would like to try unsupervised classification. To try out new things

[swayson]

Never tried it myself, but perhaps this paper could be of use:

NEURAL NETWORK-BASED CLUSTERING USING PAIRWISE CONSTRAINTS

http://arxiv.org/pdf/1511.06321v5.pdf

[abishekk92]

Given the context, I am not sure what exactly you're looking to achieve, that being said, you can take look at how to implement an autoencoder using Keras.

[gaw89]

@swayson I have been trying to implement the method in that paper the past couple of weeks, but for the life of my I can't make it work. Do you happen to know if anyone has tried it/had any success with it yet?

[rjpg]

I saw the link : https://blog.keras.io/building-autoencoders-in-keras.html

Is well done BUT is missing something critical :

How to apply autoencoders in classification using keras ?

Like , extend the examples to : given an image of a digit it tells you what is the digit it self (using autoencoders ?)

For this we need to know how to (after autoencoder unsupervised pre training):

1-to slice the NN in half on the last encoder (before the decode starts) 2- freeze the the wights of the encoders 3- add or concat a new NN in front of the last encoding layer 4- (post) train this concat NN with a softmax output layer to classify the digit ...

Example with all this would be super

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed after 30 days if no further activity occurs, but feel free to re-open a closed issue if needed.

[jolespin]

Here I give an example of using an autoencoder on the iris dataset. You can use that embedding w/ a clustering algorithm in scikit-learn. I have been experiencing an issue where one of the dimensions gets zeroed out :/

https://stackoverflow.com/questions/47297526/how-to-prevent-simple-keras-autoencoder-from-over-compressing-data