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/
captainRubik has joined #mlpack
captainRubik has quit [Ping timeout: 260 seconds]
govg has quit [Ping timeout: 256 seconds]
< zoq>
rcurtin: It's too early to say that :)
govg has joined #mlpack
vpal has joined #mlpack
vivekp has quit [Ping timeout: 240 seconds]
< cult->
just make sure you will use fancy words like deep learning
mikeling has joined #mlpack
vpal has quit [Ping timeout: 255 seconds]
vivekp has joined #mlpack
root_____ has joined #mlpack
root_____ has left #mlpack []
layback has joined #mlpack
layback has quit [Quit: leaving]
vinayakvivek has joined #mlpack
govg has quit [Ping timeout: 240 seconds]
govg has joined #mlpack
n2lmp has joined #mlpack
< n2lmp>
i'm trying to verify the output of the lars function. Is there a way to get sklearn to output the same values? If I specify alpha for lasso_path, it just outputs one single set of coefs instead of the path that mlpack outputs
< n2lmp>
(if i get disconnected i will check the logs on the website so feel free to respond)
< rcurtin>
n2lmp: you can check the last vector in mlpack's alphaPath and compare it with what scikit gives
< rcurtin>
the path is just the set of alpha values that LARS takes at each iteration, if I remember right
< n2lmp>
but in the mlpack lars function we fix the lambda parameters right?
< n2lmp>
perhaps i'm not understanding what's going on
< n2lmp>
but i thought fixing alpha in lasso_path would be like selecting lambda in mlpack lars?
< n2lmp>
basically in mlpack I do: lars -i x.csv -r y.csv -l 0.4 -L 0
< n2lmp>
in sklearn I do: linear_model.lasso_path(x,y,alphas=[0.4,],eps=1e-16, verbose=True)
mikeling has quit [Quit: Connection closed for inactivity]
< rcurtin>
n2lmp: you're right, lambda is fixed in mlpack LARS
< rcurtin>
let me quickly see the scikit documentation
< rcurtin>
yes, so to me it looks like scikit's coef_path_ variable will be the same thing that mlpack's alphaPath is
< rcurtin>
the final coefficients of the regression in scikit will be coef_, I believe
< rcurtin>
ah hang on sorry I have misspoken from the beginning
< rcurtin>
when I said "mlpack's alphaPath" I meant "mlpack's betaPath"
< rcurtin>
so the last element of LARS::BetaPath() (which is a std::vector<arma::vec>) should be the regression coefficients
< rcurtin>
and that std::vector<arma::vec> BetaPath() itself should be the same as scikit's coef_path_
< rcurtin>
I'm not fully understanding how you set the L1 or L2 penalty in the regression in scikit's implementation
< rcurtin>
so I dunno if alphas=[0.4,] will be the right way to imitate mlpack's '-l 0.4'
pratik has joined #mlpack
pratik has quit [Client Quit]
n2lmp has quit [Ping timeout: 260 seconds]
layback has joined #mlpack
vinayakvivek has quit [Quit: Connection closed for inactivity]