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/
< dkipke[m]> Ryan Curtin I've been using the go bindings with good success. I've come across an issue with the HoeffdingTree function. It executes correctly sometimes and gets a segmentation fault other times. I've ad hoc tried different parameters to see if I could discern the issue, but that has not revealed anything. With verbose=true, the logs indicate that the function returns if it successfully classifies a sample, but fails
< dkipke[m]> if it does not. Any insights?
< dkipke[m]> Sorry, the previous message is the log excerpt. You can see it returns on the first call, but does errors on the second call. The [INFO] lines are the mlpack logs.
< rcurtin[m]> dkipke: awesome that the Go bindings are useful for you! do you think you can open an issue on Github? I can try and reproduce it 👍️
< dkipke[m]> sure. I'll do that.
< dkipke[m]> In the meantime, my use case involves labeled data that requires a fast, incremental classifier amenable to streaming data. Can you (or anyone) provide guidance on the preferred -- or most stable -- mlpack classifier function? I'm still getting up the curve on classifier models and can't really differentiate between one model or another.
zoq has quit [Ping timeout: 264 seconds]
zoq has joined #mlpack
< ShahAnwaarKhalid> Hey zoq! I was working on GAN class for PR #2777 as discussed last week. I have a few doubts:
< ShahAnwaarKhalid> 1. Initially, GAN class had access to private members of FFN class as it was declared as it's friend. However, in the updated version of FFN class, those lines of code were deleted. So, should I create getter and setter functions for the private variables of FFN class that are used by GAN class? Also, what about private functions like ResetDeterministic() which is used inside GAN class?
< ShahAnwaarKhalid> (edited) ... which is used ... => ... which are used ...
ImQ009 has joined #mlpack
< zoq> ShahAnwaarKhalid: Hello, I accidentally removed the parameter, you can re-add those if you like.
< zoq> ShahAnwaarKhalid: About the friend class, maybe it makes sense to inherit from the FFN class.
< rcurtin[m]> dkipke: I think that the Hoeffding Tree is a good choice---it's quite tunable, and tuning it can really affect performance; there is a big tradeoff between how fast it learns from data and how accurate it will be
< rcurtin[m]> I believe that linear SVMs can be trained incrementally too, but really the Hoeffding tree was specifically made for streaming situations so at least to me it seems like the most natural choice
< rcurtin[m]> I feel like there are some other incrementally trainable models available via bindings, but I can't think of them right now unfortunately
< dkipke[m]> ok. Thanks. That's helpful.
< rcurtin[m]> 👍️
< dkipke[m]> I just submitted the Hoeffding Tree issue on github.
< rcurtin[m]> thanks! I'll take a look when I have a chance
< AyushSingh[m]> I built mlpack using this branch - <https://github.com/mlpack/mlpack/pull/2822>. But, I am getting the error 'No such file or directory' upon running '#include <mlpack/core/data/BERT_tokenizer.hpp>'. Is there something else I should add too?
< zoq> AyushSingh[m]: When you build the branch?
< dkipke[m]> <rcurtin[m] "thanks! I'll take a look when I"> 👍
AtreyaMajumdarGi has joined #mlpack
< AtreyaMajumdarGi> Hey all! I'm Atreya, a pre-final year student of Electronics and Communication Engineering, with a passion for Deep Learning!
< AtreyaMajumdarGi> I'd love to start contributing to mlpack. I've mostly been working in Python till now but I'd like to start building my models in C++ now, so contributing to mlpack is the perfect opportunity for me as well :P
< AtreyaMajumdarGi> I have a few ideas about implementing some cutting edge DL models that have been published recently as well, but I'd like to start working on them later, after a few other contributions xD
< zoq> AtreyaMajumdarGi: Welcome, getting familair with the codebase first is a good idea.
< AtreyaMajumdarGi> I'm exploring the github repo right now! I found a few interesting issues as well that I'd like to work on soon
< AtreyaMajumdarGi> (edited) I'm exploring ... => Thank you! I'm exploring ...
< zoq> Sounds good, let us know if there is anything that needs clarification.
ImQ009 has quit [Quit: Leaving]
< ShahAnwaarKhalid> zoq: I was also initially thinking about inheriting from FFN class.And, if we do go ahead with that, we would need to make the private members of FFN class that are used by GAN as protected. But, I'm not sure if inheriting from FFN class would add any more benefit than just being able to access those members. So, maybe it's better to just write getter and setter functions? Please let me know your thoughts.