search

wolfe-pack / wolfe

Wolfe Language and Engine
https://wolfe-pack.github.io/wolfe
Apache License 2.0
135 stars 17 forks source link

Matrix Factorization Model Wrapper #140

Closed riedelcastro closed 9 years ago

riedelcastro commented 9 years ago

In the spirit of the linear classifier and linear chain interfaces, which hide some of the Wolfe term and domain internals, it would be good to have a similar thing for matrix factorization.

rockt commented 9 years ago

Do you think our Tensor data structure and it's sampling mechanisms should go in there as well?

riedelcastro commented 9 years ago

I am wondering if it could be a simpler, possibly immutable representation of the tensor, to keep the complexity of this low.

rockt commented 9 years ago

Done with an initial version of it... can be made prettier though. Next, use this for a spec that fails once MF doesn't work anymore.

riedelcastro commented 9 years ago

Cool! Looks nice, we can iterate on this. One step could be to get rid of the var thetaStar, and use the style of the linear classifier where theta star is mixed in, either learned, or loaded. The other step is replacing the Predictor logic with the fun(...) method as used in the other examples.

On Sat, Mar 28, 2015 at 6:04 PM, Tim Rocktäschel notifications@github.com wrote:

Done with an initial version of it... can be made prettier though. Next, use this for a spec that fails once MF doesn't work anymore.

— Reply to this email directly or view it on GitHub https://github.com/wolfe-pack/wolfe/issues/140#issuecomment-87274707.

rockt commented 9 years ago

Like this?