rcurtin_irc changed the topic of #mlpack to: mlpack: a scalable machine learning library (https://www.mlpack.org/) -- channel logs: https://libera.irclog.whitequark.org/mlpack -- NOTE: messages sent here might not be seen by bridged users on matrix, gitter, or slack
AryannathAryanna has left #mlpack [#mlpack]
<jonpsy[m]> hey, hows it going :)
<rcurtin[m]> pretty good! knee-deep in OpenCL linear algebra code at the moment ... (for bandicoot)
<jonpsy[m]> Oh, you're still going strong on that one
<rcurtin[m]> my head hurts from the function names... dormbr dormqr dgesvd sgesvd dgelsd
<jonpsy[m]> rcurtin[m]: Good ol' C
<rcurtin[m]> hope things are well for you also... and at least involve less confusing names 😃😃
<jonpsy[m]> Ah, I miss C/C++ truly
<jonpsy[m]> been doin a lot of web development. PSQL and Javascript
<rcurtin[m]> nice, enjoying it? or would you rather be doing something else?
<jonpsy[m]> Idk, I don't feel that passion I had when i was doin gsoc
<rcurtin[m]> :(
<rcurtin[m]> well, don't let it consume you completely. I made the mistake over the past couple years of doing things that I was getting less and less passionate about... then, once I left RelationalAI I discovered all of a sudden that the passion came back; I had thought it was kind of gone forever
<jonpsy[m]> Its so fun, reading research paper and implementing papers. Has that kick in it
<rcurtin[m]> agreed, I feel alive again! when I was at RelationalAI it was just too much to try and work on stuff like this after a full day of working on their codebase... there was nothing left
<rcurtin[m]> now I stay up late at night writing C++... what a change! I have to be careful to remember to take some time off and go do other things too, to avoid burnout :)
<jonpsy[m]> Defo, when we enjoy our work time passes by quickly
<jonpsy[m]> What has been the progress in bandicoot since we last talked?
<jonpsy[m]> any major breakthrough?
<rcurtin[m]> yeah, I would say so---I figured out how to port OpenCL implementations of decompositions like SVDs and eigendecompositions from MAGMA/clMAGMA
<rcurtin[m]> it's a tedious process; I discovered that Conrad had already done it for the Cholesky decomposition, so mostly I am imitating what he did there
<rcurtin[m]> but this is why I'm digging around dealing with functions like dormbr or whatever
<rcurtin[m]> to me this is a big breakthrough, because I had been really worried about how to provide these more complicated decompositions to the OpenCL backend... I definitely did not want to implement them manually (that would take a very long time)
<rcurtin[m]> there's still a lot to do even once I get more of this ported, but now I can at least see the path to implement everything that's needed for a first release of bandicoot, which is super exciting, and maybe is the reason I am staying up late at night working on it 😃
<jonpsy[m]> I completely understand
<jonpsy[m]> basically you've got an important piece of the block out of the way
<rcurtin[m]> exactly
<rcurtin[m]> or... well... it will be out the way once I finish porting all these functions 😃😃 I'm getting closer though
<jonpsy[m]> haaha, must be super exciting
<jonpsy[m]> I remember being that excited when my algo's result started to match the papers
<rcurtin[m]> yeah, it feels like you are getting close and just spurs you on further! :)
<jonpsy[m]> EXACTLY
C0ntraria has joined #mlpack
C0ntraria has quit [Quit: Leaving]
<RohanKumarSingh[> Hello, I am Rohan Kumar Singh. My interests lies in Deep learning like Reinforcement Learning and NLP. Currently, I am exploring the Reinforcement learning course from Deepmind x UCL 2021. I have basic knowledge of C++ and intermediate in python. Can someone please guide me in Reinforcement learning and what steps should I follow to understand the RL codebase of mlpack?
C0ntraria has joined #mlpack
C0ntraria has quit [Quit: Leaving]
C0ntraria has joined #mlpack
<zoq[m]> Hello, a good starting point is the RL tutorial https://github.com/mlpack/mlpack/blob/4.0.1/doc/tutorials/reinforcement_learning.md
C0ntraria has quit [Quit: Leaving]
<zoq[m]> Also, check out the codebase, to get an idea how the different pieces work together.