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/
mm93 has quit [Remote host closed the connection]
togo has quit [Quit: Leaving]
outmanipulateGit has joined #mlpack
< outmanipulateGit>
Hey everyone, I'm Mani Tyagi, a 3rd year student at IIT Delhi. I look forward to contribute to mlpack reinforcement learning being an AI research student. For now I'm having trouble installing dependencies. It showsUnable to locate package libensmallen-dev
< outmanipulateGit>
Unable to locate package libstb-dev
< rcurtin>
outmanipulateGit: are you sure the version of Ubuntu or Debian you are using is new enough? you may need to upgrade
< Param-29Gitter[m>
Hello, @zoq @rcurtin I have added a comment of all the things that you guys wanted to know about #2169. Please have a look once you are free. Speed Up percentage for 1900 samples is 48% and it increases with increase in number of samples.
Rishi69 has joined #mlpack
Rishi69 has quit [Remote host closed the connection]
< outmanipulateGit>
@matrixbot I'm using Ubuntu 18.04 I don't suppose that's an issue right?
birm[m]1 has joined #mlpack
< birm[m]1>
It may not be a bad idea for someone with appveyor permission to cancel the builds associated with all but my latest commits on the model repo. Otherwise it'll block for up to 12 hours.
< rcurtin>
outmanipulateGit: hmm, ensmallen is available on at earliest 19.10. but mlpack's CMake configuration can automatically download ensmallen and stb locally, so even if you can't install them it shouldn't be an issue
shikharj[m] has quit [Ping timeout: 240 seconds]
jeffin143[m] has quit [Ping timeout: 240 seconds]
Nakul[m] has quit [Ping timeout: 240 seconds]
heisenbug has joined #mlpack
< heisenbug>
Heyyy...
< heisenbug>
I want to contribute to MLPack in GSoC 2020
< heisenbug>
I was thinking that we should add some more models to the model zo
< heisenbug>
zoo*
heisenbug has quit [Remote host closed the connection]
< PrinceGuptaGitte>
Hi @zoq , Thanks for reviewing my PR #2184 , I've made the changes you've suggested.
< PrinceGuptaGitte>
Also I apologize it took so long for such minor things.
Yihan has quit [Ping timeout: 260 seconds]
Debajit has joined #mlpack
Shiko has quit [Ping timeout: 260 seconds]
Debajit has quit [Ping timeout: 260 seconds]
< PrinceGuptaGitte>
Hi @kartikdutt18, I am working on triplet margin loss. I'm confused about how I should implement the back propagation. Should i find partial diff. of cost w.r.t to positive sample encoding, negative sample encoding or anchor encoding or should I take average of all of them. Can you please give some input so I can proceed further
< PrinceGuptaGitte>
(edited) ... Should i find ... => ... Should I find ...
< chopper_inbound[>
Prince Gupta: Considering this, you need to differentiate wrt. A
< kartikdutt18Gitt>
Hi @prince776, I think we should implement partial derivative wrt anchor input.
< kartikdutt18Gitt>
I think in that case derivative will be same derivative of MSE with input anchor and positive input minus derivative of MSE of anchor and negative input, If I'm not wrong.
< kartikdutt18Gitt>
*as
< PrinceGuptaGitte>
@kartikdutt18 @chopper_inbound thanks for the help.
< PrinceGuptaGitte>
I have one last query, when i looked at code of other loss functions like mean_squared_error, it uses arma::accu() function, which takes summ of all elemtns of vec/matrix/cube
< PrinceGuptaGitte>
(edited) ... help.
< PrinceGuptaGitte>
I have one last query, when i looked at code of other loss functions like mean_squared_error, it uses arma::accu() function, which takes summ of all elemtns of vec/matrix/cube => ... help.
< kartikdutt18Gitt>
If we do it wrt to positive or negative input we will loose other terms. And we generally differentiate wrt input
< PrinceGuptaGitte>
Everything's clicking again and now I can easily compare it to logistic regression, thanks a lot
< kartikdutt18Gitt>
Great.
< PrinceGuptaGitte>
Partial Derivative w.r.t anchor is coming to be 2(Negative - Positive)
< PrinceGuptaGitte>
Is it correct?
< kartikdutt18Gitt>
Yes, I think anchor input got cancelled out right?
< PrinceGuptaGitte>
Yes
< kartikdutt18Gitt>
Great, I think this is correct.
Ibtihaj has joined #mlpack
< Ibtihaj>
SVM in mlpack supports kernels ?
Ibtihaj has quit [Remote host closed the connection]
< sailor[m]>
<outmanipulateGit "Hey everyone, I'm Mani Tyagi, a "> did you overcome this error?
< sailor[m]>
<sailor[m] "did you overcome this error?"> same error happened for me as well.
< sailor[m]>
> <undefined>
< sailor[m]>
(edited) same error happened for me as ... => im facing the same issue as ...
< sailor[m]>
<sailor[m] "same error happened for me as we"> "sudo apt-get install -y libbfd-dev" solved the problem for me.
< PrinceGuptaGitte>
Hi @kartikdutt18 , I've added the Triplet Margin Loss. Since it's slightly different from other functions(it takes 3 inputs and not 2) I'm unsure if my implementation in correct. Please review it once you get time. PR #2208 . Thanks
< kartikdutt18Gitt>
Hi @prince776, I will take a look. Thanks for all your work.
< jeffin143[m]>
lbtihaj : no currently we only have linear svm , and no kernels
ImQ009 has quit [Read error: Connection reset by peer]