Thanks fora making available the code attached too your publication.
You highlight the fact that your use the userid (in addition to the binary vector of items viewed). Indeed, in contrary to usual AutoEncoder your input has one node more than the output.
Could you detail how you provide this userid :
- do you enter simply the userid (one integer I guess) (which is very different from the other node which are either 0 or 1 - does it not make problem ?)
- do you transform this id (as it seems to be the case, because you use an intializer) : in what way exactly do you transform this id ?
I am coding a very simple Autoencoder by using Keras (and dropout for noising), but, not yet an additionnal information like userid. I'll intend to do that and look for a convenient way.
Best regards.
Thanks fora making available the code attached too your publication.
You highlight the fact that your use the userid (in addition to the binary vector of items viewed). Indeed, in contrary to usual AutoEncoder your input has one node more than the output.
Could you detail how you provide this userid :
I am coding a very simple Autoencoder by using Keras (and dropout for noising), but, not yet an additionnal information like userid. I'll intend to do that and look for a convenient way.
Best regards.