verne.freenode.net changed the topic of #mlpack to: http://www.mlpack.org/ -- We don't respond instantly... but we will respond. Give it a few minutes. Or hours. -- Channel logs: http://www.mlpack.org/irc/
nilay has joined #mlpack
< nilay>
zoq: delta of next layer could have any size.. so in order to ensure this works always, should we do a full convolution followed by a submat..?
nilay has quit [Ping timeout: 250 seconds]
mentekid has joined #mlpack
< zoq>
nilay: The size of the delta has to match with the layer, however, you used the correct delta dimension in the first place, but since we never used wPad > 0 before, the gradient function isn't correct ... we have to use the paddedInput if wPad > 0.
mentekid has quit [Ping timeout: 244 seconds]
mentekid has joined #mlpack
marcosirc has joined #mlpack
< rcurtin>
marcosirc: could you run a quick timing simulation on the new boost priority_queue implementation, just to make sure it is as fast as the STL implementation?
< rcurtin>
I would do it myself but I think it would be easier for you to do it, since you already have everything set up and you won't need to make the STL numbers again :)
< marcosirc>
rcurtin: Ok, sure! I will do that now! :)
< rcurtin>
great, thanks
< rcurtin>
if the numbers are fine I'll merge it
travis-ci has joined #mlpack
< travis-ci>
mlpack/mlpack#1295 (master - 256d77d : Ryan Curtin): The build was broken.
sumedhghaisas has quit [Ping timeout: 250 seconds]
JMarler has joined #mlpack
< JMarler>
\HELP
< JMarler>
Hi all, unsure if I am in the correct place.
< JMarler>
Currently using mlpack in a realtime audio plugin as part of a final research project for an Msc
< JMarler>
I'm classifying (or attempting to) small scale data, specifically single audio buffers/frames of 256-512 sample points. Each with 18 features. I'm wondering whether it would be of any use for me to submit a pull request which allowed certain methods like naive bayes to classify a single instance/column vector ?
< JMarler>
i.e. arma::uword NaiveBayes::classify(arma::Col<T> data)
< JMarler>
fairly new to ML so maybe a worthless request
< JMarler>
It would make for a nice addition in my code and keep things a lot cleaner. Unable to allocate a matrix on the audio callback thread due to realtime constraints so means keeping around an arma::mat variable as a class member to be used purely to hold a single Col (current instance) to be classified.
< rcurtin>
hi JMarler
< rcurtin>
that PR would be fine with me
< rcurtin>
in the existing code your only option would be to pass a matrix with one column
< JMarler>
Great, cheers PR - will tackle it over the next day or so and submit the request.
< rcurtin>
and then receive back a predictions vector of length one
< JMarler>
Yeah going down that approach at the moment but seems a bit of a smell. Might be a few places across mlpack where similar approaches could be useful. The project will test a variety of models/methods in realtime audio context so if I come across more I'll submit a request for each. Thanks a bunch!
< rcurtin>
absolutrly, the contributions would be appreciated :)
< JMarler>
Considering a small library to wrap around mlpack for use by the audio development community (particularly users of the JUCE framework). Probably pull the code out from this research project after and refactor it into library for audio classification so great to know the mlpack community is so responsive to ideas! Good stuff :)
< JMarler>
Hopefully make a few contributions along the way if I can.
< JMarler>
Cheers for now.
< rcurtin>
:)
JMarler has quit [Quit: Page closed]
Wizzle has joined #mlpack
< Wizzle>
Hey guys, I am new to the community. I just installed MLpack on my machine and I'm going through the tutorials. I learn best from studying examples of the code. I was wondering if anyone had an example of the Collaborative Filtering algorithm. I have found the example on the site but I am looking for one that has actual data and not just a template. Any help would be appreciated
< rcurtin>
Wizzle: taking a look at cf_test.cpp in src/mlpack/tests/ might be useful
< rcurtin>
that loads a small subset of the GroupLens dataset
< Wizzle>
rcurtin: thanks for the help and I will take a look. I don't mind sharing anything I create with the community
< rcurtin>
sure, contributions are always welcome :)
Wizzle has quit [Ping timeout: 250 seconds]
marcosirc has quit [Quit: WeeChat 1.4]
Wizzle has joined #mlpack
< Wizzle>
I apologize but one last question. Can MLpack process real time data? I saw in CF-test.cpp that a .cvs file was loaded. The data I am trying to input into the CF algo is stored in a subgraph
< Wizzle>
Sorry might be a stupid question but I am just trying to see if my solutions is feasible