04:14 < AakashkaushikGi4> > `zoq on Freenode` Aakash kaushik (Gitter): Yes, 1e-6 should work.

04:14 < AakashkaushikGi4> So i figured it out later but we need to replace it with the exact decimal points as there are in the value that's been compared suppose you are comparing loss to 0.23415 so it will be 1e-5.

04:25 < AakashkaushikGi4> Let me know what do you think about this ?

06:56 < KhizirSiddiquiGi> Hey everyone! I opened mlpack website after weeks now, it is awesome!!

06:57 < KhizirSiddiquiGi> Also, there is a self ref link "Read More" in Benchmark section of homepage. I am not sure where that should point to exactly, so mentioning it here. :)

07:51 < blakjak888> Hi, I installed MLPACK via VCPKG and I found that the ENSMALLEN version is quite out of date. It is loading 2.11.2 instead of 2.14.2 which means some features are missing. I decided to check the MLPACK version installed by VCPKG and it is also quite far behind. How often is that updated?

07:51 < blakjak888> ... and is there a workaround?

08:05 < AakashkaushikGi4> > `blakjak888 on Freenode` Hi, I installed MLPACK via VCPKG and I found that the ENSMALLEN version is quite out of date. It is loading 2.11.2 instead of 2.14.2 which means some features are missing. I decided to check the MLPACK version installed by VCPKG and it is also quite far behind. How often is that updated?

08:05 < AakashkaushikGi4> @rcurtin I think we should bump up the version to at least 2.13.0 when mlpack is installed that would also resolve this [issue](https://github.com/mlpack/examples/issues/117) in mlpack/examples

08:06 < AakashkaushikGi4> > `blakjak888 on Freenode` ... and is there a workaround?

08:06 < AakashkaushikGi4> can you check which version is installed by doing `vcpkg install ensmallen:x64-windows`

08:08 < blakjak888> 2.11.2 build 2

08:12 < AakashkaushikGi4> I think the other option you can try is to build it from source https://ensmallen.org/

08:16 < blakjak888> I went to the VCPKG option as I had a lot of trouble getting the latest packages to build with VC++. I have actually built MLPACK in the past with no issues but it was about one or two years ago but the new simpler process seems to have more issues.

08:16 < blakjak888> VCPKG works but it is loading old versions

08:17 < blakjak888> I did a manual workaround to update the CONTROL and portfile.cmake to take the latest release version from github and it seems to work for ensmallen

08:17 < blakjak888> However, I am not sure it will work for MLPACK.

08:18 < AakashkaushikGi4> I think someone else with more access and knowledge might be able to help you.

09:27 < blakjak888> "However, I am not sure it will work for MLPACK."

09:27 < blakjak888> I tried but it doesn't work - at least there needs to be some changes to the cmake files I think since it seems there is a patch file.

09:57 < zoq> rcurtin: Right, haven't looked into the implementation, but what I do is to set MaxIterations() = 1 start with the last clusters and assignments and everything in a loop, to get the results for the steps, since there is no callback. But I guess that works as well.

09:58 < shrit[m]> Aakash kaushik (Gitter): Seems good to me, if you would like to open an issue please that will be awesome

09:59 < AakashkaushikGi4> Sure i will do that, also i somehow can't tag you.

10:00 < shrit[m]> Aakash kaushik (Gitter): That is strange to me, everyone can tag me. (Do you mean on github or here in the chat?)

10:00 < AakashkaushikGi4> here in the chat.

10:00 < AakashkaushikGi4> @shrit comes up with nothing on the list

10:00 < AakashkaushikGi4> okay i guess i can

10:00 < shrit[m]> try just shrit and see what is does

10:01 < shrit[m]> You are on gitter if I am right,

10:01 < AakashkaushikGi4> > @shrit comes up with nothing on the list

10:01 < AakashkaushikGi4> this one is you right.

10:01 < AakashkaushikGi4> yes i am on gitter

10:03 < shrit[m]> From my side If you just mention my name, it will make the line color red to I can be sure that this discussion involves me

10:05 < AakashkaushikGi4> I guess you don't come up in the list when i do @ so i have to write your complete name so it does refers to you then.

10:05 < AakashkaushikGi4> also i should create this issue in the mlpack/mlpack right ?

10:10 < shrit[m]> Depends on what you would like to do, If you would like to write tutorials, we have a tutorials sections in mlpack/mlpack. Otherwise, if you would like to write tutorials in for existing examples you can open it in mlpack/examples I believe it is better to open it in mlpack/examples. If we can build a set of solid tutorials their that will allows us to reference all of these tutorials on mlpac website later on.

10:12 < AakashkaushikGi4> Yeah that will be great, i will open a issue in mlpack/example shortly. :)

12:48 < KhizirSiddiquiGi> Hello everyone!

12:48 < KhizirSiddiquiGi> I came across AffinityPropagation recently, it works with graph representation to account for non-metric dissimilarities. Scikit-Learn has it but not any proper C++ implementation. Will it be good to have it here? I was thinking to implement it.

13:13 < rcurtin> KhizirSiddiquiGi: I don't see a problem with adding it, but we have to be careful in the design of the graph representation---we don't currently have any graph algorithms in C++, and so we need to carefully think about how we want users to represent graphs

14:07 < AakashkaushikGi4> Hey @shrit, I have created the issue: https://github.com/mlpack/examples/issues/120 take a look when you get the time.

16:48 < KhizirSiddiquiGi> @rcurtin the input graph can simply be the similarity matrix of the nodes or the node feature vectors (then we can find the similarity matrix internally). Further steps involved are matrix operations only, not much of a worry then.

17:39 < abernauer[m]> rcurtin: Is writing examples or tutorials using the R bindings an area of need?

17:46 < rcurtin> abernauer[m]: yeah! that would be great---maybe some of the existing examples can be adapted to R notebooks

17:46 < rcurtin> we are still in the process of submitting to CRAN though

17:56 < abernauer[m]> Ok cool yeah I would priratize working on those adapted exapmles first before doing novel examples.

17:56 < abernauer[m]> *prioritize

17:56 < abernauer[m]> *examples

17:57 < rcurtin> awesome, sounds good

17:59 < abernauer[m]> Wrote a blog post and got a previous project running on my blog. Takes a while to load. https://andrewbernauer.com/2020/09/18/revisiting-an-old-project-and-embedding-a-shiny-app/

18:07 < zoq> abernauer[m]: Would love to see R notebooks as well.

18:31 < abernauer[m]> zoq: Yeah the default in the R community for data science is R markdown documents which render to html, pdf, etc. My site is bulit with those docs. If the defualt for the examples repo is Juptyer notebooks I will use those of course.

18:36 < zoq> abernauer[m]: I mainly use Juptyer notebooks because it eliminates the step of setting everything up, which means you can start right away.

18:39 < zoq> abernauer[m]: Also I can plot something or show an image, which is not something I can do in C++ that easily, without leaving my environment.

18:47 < abernauer[m]> zoq: Yeah I agree with you Juypter notebooks are amazing for ease of use. Also I generally, think Python is a better language than R. It's more extensible and has better design. Also OOP in Python is better designed and way more straight forward.

21:14 < AdityaKandekarGi> Hi, I am Aditya and I am new to open-sourcing if anyone is open to help, it will be great. I am an engineering undergrad student at IIT Guwahati and familiar with C++, Python and frameworks like TensorFlow and PyTorch. Thanks!

21:15 < shrit[m]> Welcome to mlpack Aditya Kandekar (Gitter) here is a starting point : https://www.mlpack.org/community.html

21:17 < shrit[m]> rcurtin: I really need to find time for the gold linker, it take up to 4 minutes just to link some neural network code that I have wrote

22:23 < rcurtin> shrit[m]: I think it should be an easy change, I believe you can just specify to use gold as a CMake configuration variable (like CMAKE_CXX_FLAGS or CMAKE_LINKER_FLAGS, I can't remember)

22:46 < shrit[m]> rcurtin: Agreed, tomorrow I will have some time to make a pull request, I just want to finish cereal 😒

22:46 < rcurtin> yeah, I know the feeling :)