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/
eadwu has joined #mlpack
eadwu has quit [Remote host closed the connection]
prudhvi-hack has joined #mlpack
prudhvi-hack has quit [Remote host closed the connection]
sipun has quit [Ping timeout: 272 seconds]
favre49 has joined #mlpack
< favre49>
It gets so active here when GSoC starts :)
< favre49>
For those interested in NeuroEvolution, I'd suggest reading the PR for NEAT, I remember I had made it so that it could more easily be extended into HyperNEAT and other variants
< favre49>
I was also thinking, for those interested in the project on CMA-ES, there is issue #70 on ensmallen
< favre49>
I made PR #88 to address some of those concerns, but I never got around to finishing it.
< Nakul[m]>
@favre49 long day..no see.
< favre49>
Also in the realm of neuro-evolution, our current implementations could be even more heavily templated with different mutation or selection schemes (though I'd defer to @zoq about whether it is worth it)
< favre49>
Nakul[m] Yeah I've been very very busy with college and other projects/competitions :)
favre49 has quit [Remote host closed the connection]
prudhvi-hack has joined #mlpack
prudhvi-hack has quit [Remote host closed the connection]
arvnd has joined #mlpack
travis-ci has joined #mlpack
< travis-ci>
birm/ensmallen#1 (backup-master - 08994c5 : Marcus Edel): The build has errored.
< ShikharJaiswalGi>
@zoq @rcurtin What do you guys think about adding a disclaimer in the Ideas section mentioning aspirants to reach out to mentors through public channels and not their personal mails?
< zoq>
ShikharJaiswalGi: Well, I think you added that last year?
< ShikharJaiswalGi>
Did I? Hmm, interesting.
< zoq>
ShikharJaiswalGi: "Also, it is probably not a great idea to contact mentors directly (this is why their emails are not given here); [...]"
< ShikharJaiswalGi>
Hmm, then I guess we're doing all we could from our side to encourage private channel talk.
< zoq>
Hm, I encourage people to use the public channels.
< ShikharJaiswalGi>
Ah yes, public, not private. Gitter has a an edit option luckily.
< zoq>
Perhaps, we can use another color as well.
< ShikharJaiswalGi>
I didn't know GitHub markdown had that option.
< zoq>
I guess HTML inline works, but not sure.
< ShikharJaiswalGi>
Nice, I'll try it out then.
< ShikharJaiswalGi>
Apparently, Markdown doesn't support HTML based color rendering the text.
Ritik has quit [Remote host closed the connection]
< SriramSKGitter[m>
```diff
< SriramSKGitter[m>
- What about a hack like this?
< SriramSKGitter[m>
```
< SriramSKGitter[m>
Hmm that didn't come out right over here. Anyway, wrapping the text within ```diff and - highlights it in red, albeit there is a hyphen at the start of the sentence.
< zoq>
SriramSKGitter[m: I think that could work
< SriramSKGitter[m>
Glad to hear it : )
< rcurtin>
we could remove the mentor names from the projects entirely, but, I'm not sure that's a great idea since then everyone will be asking "who's the mentor"?
< Nakul[m]>
<rcurtin "we could remove the mentor names"> Hahahaha
< zoq>
Looks like the diff solution would put everything into a single line.
< SriramSKGitter[m>
@zoq : Definitely a hacky solution
< SriramSKGitter[m>
Honestly, it's in bold already and pretty prominent. I don't think highlighting it will help much.
< zoq>
you are probably right
< rcurtin>
yeah, I think there are some people who don't manage to read or find the right pages to read and there's not too much we can do about it :)
< rcurtin>
it's *way* better than the first year when we didn't have *anything* set up for students... the only thing we had was the mailing list, and it wasn't easy to get involved writing code
< rcurtin>
so the first year or really the first three years, until we got a nicer process for onboarding, most of March and April was spent responding to emails on the mailing list
< rcurtin>
I used to set aside 60-90 minutes every morning to handle the emails that had come in... much better to have some pages we can link to quickly now, and better-described projects that show interested folks where to look
< SriramSKGitter[m>
It's really well organized now! I found just about everything I wanted to know without any help :D
< rcurtin>
that's great to hear, it means we did a decent job :)
< rcurtin>
(that said, if there is anything you think we can improve... do let us know!)
ImQ009 has joined #mlpack
< Param-29Gitter[m>
Hey @rcurtin regarding #2169 do you want me to implement similar code with information_gain? then you can review both of them once you are free?
< rcurtin>
I wouldn't take the time to do it for information gain until the gini gain code has been adequately reviewed
< Param-29Gitter[m>
had a similar thought. That's why didn't do it yet :)
< rcurtin>
yep, sounds good---sorry that I'm moving slow on these. I have to limit the time I put into reviewing PRs, otherwise I don't actually manage to write any mlpack code :(
< Param-29Gitter[m>
I understand I'll try to find other parts in the code which I can rewrite to improve performance. Any specific algorithms which you expect to be improved in Profiling for parallelism?
< Param-29Gitter[m>
or any specific research papers that would help me in understanding what we need to do in that project?
< rcurtin>
totally up to you; it would generally be better to parallelize more widely used algorithms (logistic regression, random forests, k-means, etc.), but it may be that those have already been optimized pretty heavily so it may not be easy
Mindjolt has joined #mlpack
< Param-29Gitter[m>
Ya... I found a paper but algorithms mentioned in it have already been optimized using OpenMP. Ill do profiling of some other algorithms and try finding some other research papers which could be helpful. Also I tried working on KNN, results are not that satisfactory yet :( . I'll probably spend some more time on it and open a PR if i find something interesting ;)
< rcurtin>
Param-29Gitter[m: sounds good, but do keep in mind that OpenBLAS is itself parallelized, so any algorithm that calls out to OpenBLAS for linear algebra functions may already be effectively parallelized
< rcurtin>
but to see this you'd need to compile against OpenBLAS, and not ATLAS or the reference LAPACK/BLAS implementations
< SaraanshTandonGi>
Is it because of the type of the file
< SaraanshTandonGi>
?
< SaraanshTandonGi>
I'm trying to run the following simple code:
< PrinceGuptaGitte>
Hi, I was thinking about GSOC project(once again), and I had an idea which doesn't exactly fit any Idea propsed in the idea-list, but as it suggests we can provide some other ideas too.
Mindjolt has quit [Ping timeout: 260 seconds]
< PrinceGuptaGitte>
It would be great to hear some opinions on "Hyperparameter optimization tool" in MLPack for Ann
< SriramSKGitter[m>
Do you mean something like hyperas? That sounds interesting
< PrinceGuptaGitte>
Yes
< PrinceGuptaGitte>
It'll have a lot of things to implement like grid search, random search, coarse to fine sampling.
k3nz0 has joined #mlpack
< jeffin143[m]>
We do have some of the algorithm implemented
< jeffin143[m]>
But we have to implement some more definitely
< PrinceGuptaGitte>
@jeffin143 I can't seem to find where they are, can you tell me.
< PrinceGuptaGitte>
Guess I will work on it but not for GSOC project
< SriramSKGitter[m>
Happy to help :)
ArunavShandeelya has joined #mlpack
geek-2002Gitter[ has joined #mlpack
Nipunwaas[m] has joined #mlpack
AryamanBhagatGit has joined #mlpack
Valliappan_CAGit has joined #mlpack
EL-SHREIFGitter[ has joined #mlpack
letrunghieu[m] has joined #mlpack
sailor[m] has joined #mlpack
HimanshuPathak[m has joined #mlpack
TanayMehtaGitter has joined #mlpack
AniThoGitter[m] has joined #mlpack
johnsoncarl[m] has joined #mlpack
aadarsh-asthanaG has joined #mlpack
tanvi has joined #mlpack
tanvi has left #mlpack []
tanvi has joined #mlpack
TaapasAgrawalGit has joined #mlpack
< PrinceGuptaGitte>
Hi @zoq @ShikharJ , can I ask if implementation of YOLO algorithm for object detection is a good enough idea for GSOC 2020 regarding the idea *Application of ANN Algorithms Implemented in mlpack* in Idea-list. @zoq I know that it seems interesting but I was confused if it's enough for GSOC. Can you please give me your input. I've been working on theoretical part, now I want to move ahead.
< tanvi>
Hello everyone. I am a beginner to start as an open source contributor so can anybody help me to start?
< zoq>
PrinceGuptaGitte: Yes, it's an interesting and big enough project for GSoC.
< tanvi>
I want to contribute in mlpack how can I start from here?
< zoq>
favre49: Hm, don't think we can publish older proposals without any consent, and I guess we would probably see simialr proposals, which I think it's what we are looking for.
favre49 has joined #mlpack
< favre49>
zoq That's true, i guess a lot of people would see it as a copyable format rather than just an example
favre49 has quit [Remote host closed the connection]
Nakul has joined #mlpack
< Nakul[m]>
Here are some of my seniors proposal.If you guys need any help related to that.
< Nakul[m]>
Making proposal is totally personalize think so don't just copy paste do research about work.
Pranshu54 has quit [Remote host closed the connection]
< Nakul[m]>
Let me know if I am wrong
< saksham189Gitter>
Hey @zoq I wanted your opinion regarding the adaptive pooling mean and max layers PR ( https://github.com/mlpack/mlpack/pull/2195 ). Do you think we should implement the layer as a wrapper over the original pooling layer since most of the code is exactly the same once the stride and kernel parameters have been calculated?
< rcurtin>
Nakul[m]: agreed, every proposal is specific to the person and the project. there's no strict template or guideline to follow, and since so many people work directly with the community before submitting an application, we often have a pretty clear picture of the applicant already
< rcurtin>
which should relieve a bit of pressure; we already know who you are, so the application doesn't have to be *perfect* :)
< rcurtin>
EL-SHREIFGitter[: not sure why there's not a name there, but someone wrote it in the list so there likely is a mentor. it was probably an oversight
< rcurtin>
in any case, feel free to discuss the project here
< jeffin143[m]>
Toshal Agrawal (Gitter) / ToshalAgrawal[m] is the mentor for that project @rcurtin:matrix.org ,EL-SHREIF (Gitter)
< Nakul[m]>
rcurtin: birm added some comments #49. have a look when you got time.
< Nakul[m]>
And also is there any tool or any way to test build (due to change made in cmake) on all different OS. currently using Virtual Machine for that.
< Nakul[m]>
> rcurtin: birm added some comments #49. have a look when you got time.