[mlpack-git] [mlpack] improve speed of SparseAutoencoder and make it more flexible (#451)

[Marcus Edel]

> If you agree with it, I would make the changes.

I agree, using the network as a single parameter simplifies the parameters of the error function. Additionally, if someone needs more parameters in the future, the parameters can be extracted from the network.

> As you can see, to extract the features from the input, I do not need to forward the whole layers, I would need to access underlying network of the FFN, could I add a public api to access the network?

I don't see any convincing argument against a public function to access the network. So please go ahead.

> When implementing sparse autoencoder by FNN, I find out there are some copy could be eliminated, I think the first one should be deal with is the Forward function. [...] I would like to deal with this issue after this pull request done, any idea?

Yeah, let us discuss the issue in another thread. I guess we can change the current API, as you suggested and remove the current getter function.

---
Reply to this email directly or view it on GitHub:
https://github.com/mlpack/mlpack/pull/451#issuecomment-164465767
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.cc.gatech.edu/pipermail/mlpack-git/attachments/20151214/db723a32/attachment.html>