<jonpsy[m]>
zoq: hey, im totally clueless how to use ```friend``` in our use case...
<jonpsy[m]>
i thought of inheritance, but then it will also inherit GAN as it `friend` which would b awkard
CountryNerd has joined #mlpack
<zoq[m]>
<jonpsy[m]> "i thought of inheritance, but..." <- Only temporary, the ongoing refactoring will change that.
<jonpsy[m]>
Out of curiousity, what will happen to GAN after refactoring?
<jonpsy[m]>
im assuming all of them would be marked as virtual, in our inheritance I'm deleting most functions
<jonpsy[m]>
* most functions using ```delete```
<zoq[m]>
<jonpsy[m]> "im assuming all of them would be..." <- We where hoping to actually use the base FFN class without adding any extra functionality to support GAN's.
<zoq[m]>
Not sure yet that will work out, because currently the main focus is on getting the different layers ready as well as the main class.
<zoq[m]>
Like the RNN class will mainly use the FFN implementation.
<shrit[m]>
I do not know what we have removed from mlpack recently, but the CLI binaries size is going lower and lwoer
<shrit[m]>
s/lwoer/lower/
<shrit[m]>
The mlpack_knn now is at 2.2 MB
<shrit[m]>
Last thing I was able to get is 3.4MB
<rcurtin[m]>
Could be the boost::any removal?
<jeffin143[m]>
That's good :)
<shrit[m]>
Maybe
<jeffin143[m]>
Only if we could know how did that happen
<rcurtin[m]>
Most of those will fit on individual 5.25" floppies that I have sitting around 😄
<rcurtin[m]>
maybe there is hope for my 386 machine learning project after all 😂
<shrit[m]>
Yeah definitely
<jeffin143[m]>
The removal of boost:any is not mentioned in history may be it's good to mention it
<shrit[m]>
I know I am going to open an PR for this
<shrit[m]>
sometime this day
<shrit[m]>
maybe tomorrow
<jeffin143[m]>
Did the ceremony happen ?? When Ryan prepares a ppt of all the projects and improvements done over a gsoc period
<jeffin143[m]>
It's just over a year , when last version of mlpack was released
<shrit[m]>
Also this is the master, so it does include boost spririt. I am waiting for heisenbuug (Gopi M Tatiraju) to finish the last comments so I can merge his pr
<shrit[m]>
maybe we can get even a better results without boost spirit
<jeffin143[m]>
Yes under 2mb may be
<shrit[m]>
I do think all of these model can fit a micro controller such as a RPI pico
<shrit[m]>
but not all arduino boards
<rcurtin[m]>
jeffin143: I guess I did that once or twice, but it was super time consuming and I'm underwater with work at the moment :( if someone else wanted to they should feel free!
<rcurtin[m]>
I know it's been a long time since the last release, but I really think before mlpack 4 we should finish the various refactorings that are in progress
CountryNerd has quit [Read error: Connection reset by peer]
<jeffin143[m]>
We have come a long way from the last year
<jeffin143[m]>
Haven't had the chance to look at all the changes
<rcurtin[m]>
yeah definitely! still lots to do until Boost is gone though :(
<jeffin143[m]>
Boost maths and spirit are on the way 🚀
<shrit[m]>
I was surprised, it has been a lot of time that I did not profile these binaries, I would like to have a look at these symbols to see what is happening.
<shrit[m]>
However, even we did a great improvement over the last year considering the installation and binaries size, I never checked the memory usage, I do not know how much RAM is being used by these binaries during runtime
<shrit[m]>
if someone has a great simple tool to recommend, I will use it
<rcurtin[m]>
I just watch `free` or use `htop` but that's not a very scientific solution 😄
<shrit[m]>
yeah, but it does not scale 😄
<rcurtin[m]>
it's true 😄
<shrit[m]>
We need something to monitor the stack in real time, and see what is happening
<zoq[m]>
I used massif in the past.
<shrit[m]>
is it only for the heap?
<zoq[m]>
I think so.
<jonpsy[m]>
<zoq[m]> "We where hoping to actually..." <- Hm, so i was saying that I'd delete functions like Evaluate, EvaluateWithGradient, Backward because they'd be inherited. Hope that's fine?
<jonpsy[m]>
Deleting would help make sure that the wrong network isn't selected
<jonpsy[m]>
Deleting would help make sure that methods which aren't useful to `MORLNetwork` can't be invoked