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/
rf_sust2018 has joined #mlpack
nikhilgoel1997 has joined #mlpack
ricklly__ has joined #mlpack
ricklly__ has quit [Client Quit]
ricklly__ has joined #mlpack
rf_sust2018 has quit [Quit: Leaving.]
ricklly__ has quit [Quit: Konversation terminated!]
nikhilgoel1997 has quit [Quit: Connection closed for inactivity]
surya has joined #mlpack
< surya>
I need some help understanding how BPTT works in the optimizers
govg has quit [Ping timeout: 256 seconds]
govg has joined #mlpack
hassan_ has joined #mlpack
govg has quit [Ping timeout: 240 seconds]
< hassan_>
Hello People. I showed interest in "Essential Deep Learning Modules" project on the mailing list.
< hassan_>
I digged deep in the project, specifically "Stacked Generative Adversarial Networks" and "Bidirectional Recurrent Neural Networks".
< hassan_>
There are a few questions that I need to ask to the mentors of project, Marcus Edel and Mikhail Lozhnikov.
ricklly_ has joined #mlpack
hassan_ has quit [Ping timeout: 264 seconds]
navneet has quit [Quit: Connection closed for inactivity]
surya has quit [Ping timeout: 260 seconds]
govg has joined #mlpack
surya has joined #mlpack
sumedhghaisas has quit [Ping timeout: 240 seconds]
sumedhghaisas has joined #mlpack
hassan_ has joined #mlpack
hassan_ is now known as hassanmahmood
surya has quit [Ping timeout: 256 seconds]
surya has joined #mlpack
luffy1996 has quit [Quit: Connection closed for inactivity]
surya has quit [Read error: Connection reset by peer]
manthan has joined #mlpack
< Atharva>
zoq: Is there a way we can get the activations of certain hidden layers which we want in the ffn/rnn classes?
donjin_master has joined #mlpack
< donjin_master>
hello everyone zoq i have an query that for one project more than one person can be selected or not?? like in reinforcement learning project should i have to submit the propsal of one algorithm or according to my capability if the same person has given proposal for that then should i have a chance to get selected..
donjin_master has quit [Client Quit]
surya has joined #mlpack
surya has joined #mlpack
surya has quit [Read error: Connection reset by peer]
surya has joined #mlpack
killer_bee[m] has quit [Ping timeout: 276 seconds]
killer_bee[m] has joined #mlpack
csoni has joined #mlpack
csoni has quit [Read error: Connection reset by peer]
vivekp has quit [Ping timeout: 264 seconds]
vivekp has joined #mlpack
surya2 has joined #mlpack
surya has quit [Read error: Connection reset by peer]
surya has joined #mlpack
surya2 has quit [Ping timeout: 248 seconds]
surya2 has joined #mlpack
ckeshavabs has joined #mlpack
surya has quit [Ping timeout: 240 seconds]
ckeshavabs has quit [Quit: Page closed]
< manthan>
atharva i think, looping for the network, just like how looping for evaluating the network or backprop is done, can get you the activations.
ckeshavabs has joined #mlpack
< ckeshavabs>
zoq: thanks for the link on Alias layer. I will look into it and get back to you
< ckeshavabs>
Also, can you help me find the header files/definitions of certain classes in the codebase? Like the UpdaterType is used extensively in q_learning.hpp file, but I wanted to see it's implementation. I googled and could find ReplayType and NetworkType definitions, but I wanted a more general way to find files.
hassanmahmood has quit [Ping timeout: 240 seconds]
ckeshavabs has joined #mlpack
< Atharva>
manthan: thanks!
< ckeshavabs>
Also, I am interested in implementing reward clipping and gradient clipping for the q_learning_impl.cpp file. So, is it better write them as a seperate class? Also, these could be useful for other purposes than RL also? So, where would it be appropriate to place it (which directory)?
surya2 has joined #mlpack
surya has quit [Ping timeout: 276 seconds]
ckeshavabs has quit [Quit: Page closed]
manthan has quit [Ping timeout: 260 seconds]
Nisha_ has quit [Quit: Page closed]
hassanmahmood has joined #mlpack
ckeshavabs has joined #mlpack
< ckeshavabs>
Hi, I had asked a question about finding the relevant files at about 09:21, but I figured that I could just follow the #includes in .cpp files. I couldn't figure it out then, because I was only looking through header files
dandelion_ has joined #mlpack
dandelion_ has quit [Client Quit]
ckeshavabs has quit [Quit: Page closed]
surya2 has quit [Ping timeout: 264 seconds]
sumedhghaisas2 has joined #mlpack
sumedhghaisas has quit [Ping timeout: 264 seconds]
sujith has joined #mlpack
< zoq>
hassanmahmood: Hello, please post the queries here, once we have a chance we will get back.
< zoq>
donjin_master: The number of methods should be reasonable, we only have a couple of weeks and we will have to write tests and documentation for each.
< zoq>
ckeshavabs: Gradient clipping is already implemented, reward clipping could be implemented in the same way.
surya has quit [Ping timeout: 240 seconds]
sumedhghaisas has joined #mlpack
sumedhghaisas2 has quit [Read error: Connection reset by peer]
sumedhghaisas2 has joined #mlpack
sumedhghaisas has quit [Ping timeout: 256 seconds]
sumedhghaisas2 has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
sumedhghaisas2 has joined #mlpack
sumedhghaisas has quit [Read error: Connection reset by peer]
haritha1313 has joined #mlpack
< haritha1313>
@rcurtin: @zoq: Hi, I have uploaded a project proposal draft for idea 'Alternatives to neighborhood based collaborative filtering'. Could you please have a look and suggest changes whenever you have time. Thanks.
haritha1313 has quit [Ping timeout: 260 seconds]
haritha1313 has joined #mlpack
< haritha1313>
@sumedhghaisas2
haritha1313 has quit [Ping timeout: 260 seconds]
< rcurtin>
haritha1313: looks good, my only question is, how will this integrate with the existing `CF` class and cf binding? ideally I'd prefer to not also have a separate `NCF` class and ncf binding
ImQ009 has joined #mlpack
hassanmahmood has quit [Ping timeout: 265 seconds]
hassanmahmood has joined #mlpack
sujith has quit [Ping timeout: 260 seconds]
rf_sust2018 has joined #mlpack
haritha1313 has joined #mlpack
< haritha1313>
@rcurtin: Thanks for going through the draft. The reason I kept NCF as a different class is because it has almost different data member/member function requirements. True, it would be logically better to have it within CF itself.
< haritha1313>
A few alternatives I could think of are: 1. Have NCF as a seperate class within mlpack/methods/cf which can then be accessed within cf_main to create objects.
< haritha1313>
2. Derive NCF class from CF class. 3. Merge both classes and overload functions like Train(), ComputeRecommendations() etc. (but this could result in unwanted data members depending upon algorithm chosen).
< haritha1313>
I am not sure of how clean the structure would be in any of these cases, please do let me know what you think.
haritha1313 has quit [Ping timeout: 260 seconds]
< Atharva>
sumedhghaisas2: on the gsoc ideas page, it has been mentioned to add regularizing functionality for the VAE, what exactly is expected? The paper it points doesn’t say much about it.
haritha1313 has joined #mlpack
manish7294 has joined #mlpack
< manish7294>
rcurtin: I have made some changes in proposal. If you get to find some time, then please comment your opinion.
< rcurtin>
haritha1313: I'd suggest possibly the use of templates to make the CF class more generic; for instance, I might consider something kind of like
< rcurtin>
CF<DecompositionStrategy> where DecompositionStrategy can be NCF, AMF, or whatever else
< rcurtin>
I think this will require some amount of refactoring of the CF code, but I think it would result in a better class in the end
< rcurtin>
manish7294: I don't have time today, but I will try and look in the next couple of days
< manish7294>
rcurtin: Sure, No hurry
< haritha1313>
@rcurtin: I did consider this approach.
< haritha1313>
I was concerned with the FFN related data members. If I make them a part of the CF class, they won't be useful if decomposition strategy is set to NMF or something else.
< haritha1313>
But this is what I meant by merging the classes in the third alternative I suggested.
< rcurtin>
haritha1313: right, but if you do this with templates I think you can avoid having decomposition-specific members be a part of the CF class itself
haritha1313_ has joined #mlpack
haritha1313 has quit [Ping timeout: 260 seconds]
< haritha1313_>
@rcurtin: True, it will need some refactoring in CF, but it will be worth doing :) . Thanks for helping out.
< haritha1313_>
I will make modifications to accomodate this in the proposal. Please do let me know if there are any other suggestions regarding the draft.
haritha1313_ has quit [Ping timeout: 260 seconds]
s1998 has joined #mlpack
ricklly_ has quit [Quit: Konversation terminated!]
manish7294 has quit [Ping timeout: 265 seconds]
< sumedhghaisas2>
@Atharva: Hi Atharva
< sumedhghaisas2>
the regularizing functionality is mentioned in the tutorial paper that I have added.
< sumedhghaisas2>
let me see if I can find the link again
< sumedhghaisas2>
I would prioritize over conditional distribution before regularization. :)
rajeshdm9 has quit [Quit: Page closed]
sumedhghaisas2 has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
rf_sust2018 has quit [Quit: Leaving.]
< Atharva>
sumedhghaisas: Sorry, I am a little confused, do you mean to say that I should prioritize conditional distribution and then look at regularisation?
< sumedhghaisas>
@Atharva: Yes :)
< Atharva>
Okay, and is it okay if I leave the implementation of some of the things to be figured out later? my proposal is getting very long and also some things feel like they cannot be decided before we actually start working.
< Atharva>
sumedhghaisas:
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
Arshdeep has joined #mlpack
Arshdeep has quit [Remote host closed the connection]
sumedhghaisas2 has joined #mlpack
sumedhghaisas2 has quit [Read error: Connection reset by peer]
sumedhghaisas has quit [Ping timeout: 246 seconds]
sumedhghaisas2 has joined #mlpack
sumedhghaisas has joined #mlpack
Arshdeep has joined #mlpack
Arshdeep has quit [Remote host closed the connection]
sumedhghaisas2 has quit [Ping timeout: 256 seconds]
sumedhghaisas has quit [Ping timeout: 256 seconds]
sumedhghaisas has joined #mlpack
< Atharva>
sumedhghaisas: in conditional distribution, we concatenate the labels to the input data. But what if the input data is 2d(say an image) and we want to use convolutional layers, how should we concatenate?
rf_sust2018 has joined #mlpack
rf_sust2018 has left #mlpack []
csoni has joined #mlpack
hassanmahmood has quit [Ping timeout: 248 seconds]
sumedhghaisas has quit [Ping timeout: 276 seconds]
sumedhghaisas has joined #mlpack
< sumedhghaisas>
@Atharva: the condition can be treated in many ways, the simplest is to process labels through linear layer to get it in the image shape, for Mnist it's 28 * 28 then add it to the image as channel
< Atharva>
Oh, okay, that’s really cool
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
sumedhghaisas2 has joined #mlpack
sumedhghaisas3 has joined #mlpack
mystery has joined #mlpack
< mystery>
hey
sumedhghaisas has quit [Ping timeout: 246 seconds]
< mystery>
i wanted to know more about the project
< mystery>
like where to start from
< mystery>
i am a new commer to this project
< mystery>
maybe someone could help me with that
sumedhghaisas2 has quit [Ping timeout: 246 seconds]