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/
petris has joined #mlpack
MystikNinja has joined #mlpack
MystikNinja has left #mlpack []
csoni has quit [Ping timeout: 256 seconds]
s1998 has quit [Ping timeout: 240 seconds]
ckeshavabs has joined #mlpack
govg has joined #mlpack
s1998 has joined #mlpack
< ckeshavabs>
zoq: yesterday I had asked about implementation of reward clipping method. I just wanted to clarify about it's implementation. Some of the blogs just classify all positive rewards as +1 and the negative ones as -1. They live the zero rewards as it is.
< ckeshavabs>
They retain(leave**) the zero rewards as it is.
csoni has joined #mlpack
< ckeshavabs>
Is this the correct form of implementation of reward clipping?
csoni has quit [Read error: Connection reset by peer]
ckeshavabs has quit [Quit: Page closed]
manthan has joined #mlpack
travis-ci has joined #mlpack
< travis-ci>
ShikharJ/mlpack#115 (master - 793fe4a : Marcus Edel): The build has errored.
< Nisha_>
Hi @zoq, please review my draft after the suggested changes :) Waiting for the feedback :)
nikhilgoel1997 has joined #mlpack
s1998 has quit [Ping timeout: 260 seconds]
s1998 has joined #mlpack
ricklly has joined #mlpack
ricklly has quit [Client Quit]
csoni has joined #mlpack
csoni has quit [Ping timeout: 240 seconds]
s1998 has quit [Ping timeout: 240 seconds]
s1998 has joined #mlpack
s1998 has quit [Ping timeout: 276 seconds]
s1998 has joined #mlpack
s1998 has quit [Ping timeout: 256 seconds]
s1998 has joined #mlpack
manthan has quit [Ping timeout: 260 seconds]
manthan has joined #mlpack
hassanmahmood has quit [Ping timeout: 264 seconds]
< manthan>
zoq : i have updated the proposal according to your suggestions, could you please have a look once you find time. Thanks. API structure for TokeniserEncoder has been updated so kindly review that please.
nikhilgoel1997 has quit [Quit: Connection closed for inactivity]
thepro has joined #mlpack
< thepro>
hey
< thepro>
can anyone tell me if the decision tree implementation available in mlpack is susceptible to outliers as in case of C4.5?
sujith has joined #mlpack
s1998 has quit [Read error: Connection reset by peer]
s1998 has joined #mlpack
sujith has quit [Ping timeout: 260 seconds]
csoni has joined #mlpack
csoni has quit [Ping timeout: 240 seconds]
s1998 has quit [Read error: Connection reset by peer]
s1998 has joined #mlpack
nikhilgoel1997 has joined #mlpack
sumedhghaisas2 has joined #mlpack
sumedhghaisas has quit [Ping timeout: 240 seconds]
_sunny has joined #mlpack
_sunny has left #mlpack []
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
manthan has quit [Ping timeout: 260 seconds]
< zoq>
ckeshavabs: This is a variant of reward clipping, you could also say you only clip the positive rewards.
bhuwnesh has joined #mlpack
< rcurtin>
thepro: yes, there is no special handling of outliers in the mlpack decision tree implementation, so I would expect it to act the same as C4.5
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
< thepro>
rcurtin:Can we implement C5.0 for mlpack (it is covered by a patent.)?
< rcurtin>
thepro: I don't know about the legal status of C5.0 but I do think that some research would need to be done first to look into that
< rcurtin>
zoq: any problem if I merge #1313? ("miscellaneous test fixes") I added a few more changes since you approved it; they are all fairly minor
sumedhghaisas has quit [Read error: Connection reset by peer]
< zoq>
rcurtin: Have seen the changes, looks ready for me.
< rcurtin>
thepro: that paper appears to be just a comparison of ID3, C4.5, and C5.0; to me it does not say anything about the legal status of C5.0 and whether it is patent protected
< rcurtin>
zoq: ok, I'll go ahead and merge it
< rcurtin>
I'd like to restart the matrix builds, but I feel like we should have a fix for the batch normalization test first; maybe we should merge #1327 now and let Manthan provide an update if he'd prefer to solve it differently?
< zoq>
rcurtin: I've asked him on the issue what he likes to do, but I think merging this now works for me, if you like to test the matrix build.
< rcurtin>
sure, I think maybe we should do that, just to make sure that no other tests are failing
< rcurtin>
let me update the PR
< zoq>
rcurtin: Okay, I can update the PR and add the comments now or later, your call :)
< rcurtin>
if you'd like to add the comments, I think that would be useful and then we can merge it
< zoq>
rcurtin: okay
< rcurtin>
my guess (but I am not sure) is that Manthan will be fine with the changes, so it seems better to me to just do it all now
< nikhilgoel1997>
@rcurtin: @zoq it will be really helpful if you could review my proposal on profiling for parallelization and suggest the changes so I can improve it
< rcurtin>
nikhilgoel1997: it's not likely that I'll have time to do that
< rcurtin>
thepro: I am not a legal expert, and I think there is risk of C5.0 having patents associated with it (so I suggest maybe you do a patent search), so I think it might be better to avoid it
< rcurtin>
nikhilgoel1997: there is an application guide on the mlpack github wiki, I'd suggest taking a look at that
s1998 has joined #mlpack
s1998 has quit [Client Quit]
< nikhilgoel1997>
Yes, I've followed the application guide. I submitted the proposal a few days ago and didn't get feedback so I was a bit concerned.
< nikhilgoel1997>
Also, will it be fine if I implement the changes in one of the algorithms mentioned and submit a PR just to understand the exact way in which the results should be submitted?
< rcurtin>
nikhilgoel1997: many proposals won't get feedback, it's simply not possible---last year we received 100+ applications
< nikhilgoel1997>
I'm sorry I didn't realize that and I understand
< rcurtin>
no problem
csoni has joined #mlpack
govg has quit [Ping timeout: 264 seconds]
< rcurtin>
zoq: the comments in #1327 look good to me, if you want to go ahead and merge feel free, then I'll restart the matrix build and we'll see what happens
manthan has joined #mlpack
csoni has quit [Read error: Connection reset by peer]
ImQ009 has joined #mlpack
bhuwnesh has quit [Quit: Connection closed for inactivity]
sumedhghaisas has quit [Ping timeout: 265 seconds]
sumedhghaisas has joined #mlpack
thepro has quit [Ping timeout: 240 seconds]
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
thepro has joined #mlpack
sumedhghaisas2 has joined #mlpack
sumedhghaisas has quit [Ping timeout: 256 seconds]
< zoq>
rcurtin: Does the http://masterblaster.mlpack.org/view/docker/job/docker%20mlpack%20nightly%20build/ use the latest master branch? Looks like the gradient check failed on some machines, "critical check CheckGradient(function) <= 1e-3 failed [0.0024685914704367208 > 0.001]" but the new tests uses 1e-4 so, not sure it's the latest version.
< rcurtin>
yeah, I just realized that I forgot to run the git checkout job first
sumedhghaisas2 has quit [Read error: Connection reset by peer]
nikhilgoel1997 has quit [Quit: Connection closed for inactivity]
sumedhghaisas2 has joined #mlpack
sumedhghaisas3 has joined #mlpack
sumedhghaisas2 has quit [Ping timeout: 276 seconds]
yashsharan has joined #mlpack
yashsharan has quit [Ping timeout: 260 seconds]
< Atharva>
sumedhghaisas3: on gsoc ideas page, there has been mention of of inplementing denoising VAEs, in the linked paper, there is only mention of denoising autoencoders (not variational), what exactly is expected there?
Nisha__ has joined #mlpack
< Nisha__>
Hi Zoq, i was wondering if you had a chance to review my proposal after the changes.
< zoq>
Nisha__: No, note we are doing our best to give as much feedback as we can, but there are too many application, so it's possible that we can't review an application twice.
thepro has quit [Quit: Leaving]
ricklly_ has joined #mlpack
donjin_master has joined #mlpack
< Nisha__>
Okay, no problem.. I understand completely :)
< donjin_master>
Hello everyone . How many environments do we have to create for implementing RL in atari games over this summer..
< zoq>
donjin_master: Not sure I get the point, we already have some env's for testing, and we can use https://github.com/zoq/gym_tcp_api which allows us to use eveything gym provides.
Nisha__ has quit [Quit: Page closed]
< donjin_master>
But these envs are for claasic control they are not atari games environment
< zoq>
donjin_master: Correct, as I said, we can use the wrapper, if you like to run test against Atari.
< donjin_master>
Ok thabks
< donjin_master>
Thanks i got it
< zoq>
donjin_master: Here to help.
< donjin_master>
Zoq i have still some doubts regarding my proposal
< sumedhghaisas3>
@Atharva: Denoising autoencoder is achieved with VAEs in that paper.
< sumedhghaisas3>
we will implement VAE and may test its Denoising capabilities
< sumedhghaisas3>
although the priority will be to achieve without noise reconstruction
< sumedhghaisas3>
the implementation of Denoising and without is basically same. in Denoising we provide input with little noise added and train it to reconstruct the image back
< donjin_master>
zoq where should i get the dataset for my CNN in DQN algorith
< donjin_master>
algorithm
< zoq>
donjin_master: You could extract it from the environment itself.
< donjin_master>
I am sorry for asking that do we only have to implemet a dqn algorithm over a period of three months
< donjin_master>
If something else we have to do then please tell me.
< zoq>
donjin_master: You should pick the number of methods you think is reasonable. The ideas pages lists some methods, you are free to propose anything in addition or else.
< donjin_master>
Means i can give the proposal for 2 methods if i am comfortable with that..
< Atharva>
sumedhghaisas3: got it, thanks!
< donjin_master>
Atharva
< zoq>
donjin_master: Not sure what you mean, note a student is expected to work full time (40 hours) over the "summer". If you think implementing DQN takes like one month or more, you should list what you like to do the rest of the time.
< donjin_master>
Thanks zoq
donjin_master has quit [Quit: Page closed]
apoorv has joined #mlpack
apoorv has quit [Client Quit]
donjin_master has joined #mlpack
< donjin_master>
Zoq i have a doubt why don't mlpack use gitter for their communication
< rcurtin>
donjin_master: why should we?
< rcurtin>
or, more specifically, what advantages does gitter offer over IRC that would benefit us?
donjin_master has quit [Ping timeout: 260 seconds]
csoni has joined #mlpack
donjin_master has joined #mlpack
ckeshavabs has joined #mlpack
< donjin_master>
In my opinion in gitter we can see the person profile and github profile and by using irc no one have their originality (thats my thinking)
< rcurtin>
in this particular case, basically everybody's IRC nicks match their github profile names
< donjin_master>
But in gitter everyone is in group and they have logs according to date wise which take more time to search any message
< donjin_master>
I mean irc have logs..
< donjin_master>
Which is mostly filled with the joined and quit statement
< rcurtin>
you can use any client you like for IRC, and you can filter out the join and quit statements
< donjin_master>
How i can filter join statements
< rcurtin>
you are using the web client that I think is not very configurable
< rcurtin>
so you would probably want to find another client, read its documentation, and figure out how you want to configure it
< donjin_master>
I got it how to filter join and quit statement
< donjin_master>
rcurtin can i ask you a question as a fresher
< donjin_master>
Newcommer to gsoc
< rcurtin>
donjin_master: yes, I can try to answer the question
< donjin_master>
What can we do in community bonding period because everyone is discussing their project ..
< rcurtin>
I don't understand the question
< donjin_master>
There is a gap of 1 month for community bonding
< donjin_master>
In that gap what we will have to do ?
< rcurtin>
right, so accepted students aren't required to do anything at all, but the month-long period is useful for getting in touch with a mentor, getting to know them, meeting people (I guess over IRC) in the organization, and becoming more comfortable with the codebase
< donjin_master>
Thanks rcurtin
< rcurtin>
sure
< donjin_master>
rcurtin can we have a sample proposal in our website
< donjin_master>
Because i am not getting what to share in my proposal
< rcurtin>
there is an application guide in the wiki and a lot of resources online for how to prepare a proposal for GSoC; I'd suggest reading those
< donjin_master>
I have gone through the application guide given by you.
< rcurtin>
there is no way we can provide a sample proposal because what we are looking for changes every year
< rcurtin>
it is up to the student to provide a high-quality proposal based on what they think is important, and it's very important that the student is able to figure out how to put together a good proposal themselves
< donjin_master>
Yes you are right its the work of student
< donjin_master>
I will try to figure it out by myself
< rcurtin>
the reason that is important is because students should be able to accomplish their projects without needing supervision at every step from the mentor---the mentors typically have only a handful of hours available in the week to help
< donjin_master>
Only one student is selected for one project or more than that
< rcurtin>
that depends on the project
< rcurtin>
and on the strength of the proposals
< donjin_master>
Do u think to give a timeline for the project is good or must?
< rcurtin>
I think that all of the questions you are asking can be answered in the application guide I linked to earlier
< donjin_master>
Sorry for asking
< ckeshavabs>
zoq: Hi, prioritised experience replay doesn't seem to be implemented for Reinforcement Learning. So would it be better to implement it as a separate method/class and integrate it with the existing DQN implementation during the summer?
< rcurtin>
it's ok, I am happy to help with questions that can't be answered elsewhere, but please try and answer questions yourself if you can (it saves time)
< donjin_master>
Ok i will take care from next time
donjin_master has quit [Quit: Page closed]
sumedhghaisas3 has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
ckeshavabs has quit [Quit: Page closed]
< manthan>
rcurtin : could you please review my proposal once when you have time, i have updated the proposal according to zoq's sugestions.
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas has joined #mlpack
manthan has quit [Ping timeout: 260 seconds]
sumedhghaisas2 has joined #mlpack
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas2 has quit [Read error: Connection reset by peer]