ChanServ changed the topic of #mlpack to: "mlpack: a fast, flexible machine learning library :: We don't always respond instantly, but we will respond; please be patient :: Logs at http://www.mlpack.org/irc/
ib07 has quit [Remote host closed the connection]
ib07 has joined #mlpack
ib07 has quit [Remote host closed the connection]
ib07 has joined #mlpack
ib07 has quit [Max SendQ exceeded]
ib07 has joined #mlpack
ib07 has quit [Max SendQ exceeded]
ib07 has joined #mlpack
ImQ009 has joined #mlpack
ImQ009 has quit [Quit: Leaving]
< rcurtin>
shrit[m]: I guess for the RL code you mean?
< shrit[m]>
rcurtin: exactly
< shrit[m]>
I think we have only two out of 12 methods.
< rcurtin>
sounds good to me
< rcurtin>
I know very little about the RL code though, so probably @zoq or others are the right people to ask
< rcurtin>
I looked at the cereal build---I think it's basically there for AppVeyor, but I am guessing that MSBUILD isn't set right
< rcurtin>
I think maybe it should be C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\MSBuild\Current\Bin\MSBuild.exe
< zoq>
ersonally, I would improve the converter first, and see if we can load pre-trained models instead of training them from scratch.
< rcurtin>
just guessing by the comment below the `matrix` block :-D
< zoq>
*Personally
< zoq>
But if someone likes to implement more methods or likes to provide pre-trained models, happy to review.
< shrit[m]>
I agree, if someone is interested and have time. I do not have any time right now. I think it should be easy to implement all of them since we already have two.
< abernauer[m]>
Is it necessary to print the ratings for the movie recommendations example. For the user I did recommendations for they didn't actually rate any of the recommendations.
< zoq>
abernauer[m]: That's fine.
< abernauer[m]>
So just don't print anything vs the titles and NAs ?
< rcurtin>
so it's up to you how you want to put the tutorial together, but the more information you can give to the reader the better
< rcurtin>
personally, just hearing about it from the messages here (and so remember I don't have the full picture) I would think it might be useful to have the actual recommendations printed, so the reader can see that there were recommendations produced
< rcurtin>
but, I don't know the full story---so you can make a judgment call :)
< zoq[m]>
What is the reason to not print the recommendations?
< abernauer[m]>
Yeah I printed the recommendations in the notebook. Just was wondering if I should add another code cell with the ratings printed along the side.
< rcurtin>
ohh, you mean the predicted ratings themselves... I see... I'd say, up to you... I guess they could be useful to look at?
< rcurtin>
I thought you meant whether to print the recommendations themselves, sorry for the misunderstanding :)
< abernauer[m]>
Yeah my sentece was a bit ambiguous.
< abernauer[m]>
*sentence
< AakashkaushikGit>
Hey, when trying to replace the boost::visitors basically the apply_visitors with the actual Layer* object do i need to redefine the visitors classes or what do i exactly need to do?, i have been reading about the visitor patterns through boost and without boost too. But by what we are trying to do it feels like those functions need to be defined in the linear class itself to be called through a Layer*, I am not
< AakashkaushikGit>
really sure how to move forward from here and have been stuck there.
< AakashkaushikGit>
(edited) ... actual Layer* object do i need to redefine the visitors classes or what do i exactly need to do?, i have been reading about the visitor patterns through boost and without boost too. But by what we are trying to do it feels like those functions need to be defined in the linear class itself to be called through a Layer*, I ... => ... actual `Layer*` object do i need to redefine the visitors classes or what
< AakashkaushikGit>
do i exactly need to do? I have been reading about the visitor patterns through boost and without boost too. But by what we are trying to do it feels like those functions need to be defined in the linear class itself to be called through a `Layer*`, I ...
< AakashkaushikGit>
And so many files interconnecting do confuse me a bit.
< rcurtin>
probably each visitor's functionality needs to be replaced with a virtual method in BaseLayer, and then instead of using the apply_visitor() call, you can call the virtual method directly
< rcurtin>
so it's probably easiest to start with, e.g., ForwardVisitor
< rcurtin>
take a look at how that's used in ffn_impl.hpp; you can see it's a loop over all layers, and at each layer we call apply_visitor() with the output of the previous layer
< rcurtin>
instead of using apply_visitor(), the idea would be to call layers[i].Forward() directly (or something like that), and since those are virtual functions they will call the appropriate implementations
< rcurtin>
`layers` will also need to be a std::vector<BaseLayer*> (so I guess it should be layers[i]->Forward() not layers[i].Forward())
< rcurtin>
anyway, play with it and see what you think... the code can be a bit hard to understand, but probably is easier to understand if you start in ffn_impl.hpp, understand how the visitor is being used, then substitute a virtual method implementation in place of that visitor
< abernauer[m]>
ok I joined the two data sets on the column movieId and then just took the mean rating of each of the recommendations printed those to give more context on the quality of the movies.
< AakashkaushikGit>
Thank you @rcurtin , I will start with what you said in the morning itself, excited to see the results and i got to know a lot of the codebase just through this :D
< rcurtin>
actually, no need to reproduce---I can see that libmlpack.so isn't even included in the OS X wheel... so next I just need to figure out why in the world that is...
< zoq>
rcurtin: Just seen the message.
< rcurtin>
it seems like every release to PyPI for OS X has never contained libmlpack.dylib (I guess it is not libmlpack.so on OS X :))