<jonpsy[m]>
zoq: Just to confirm, would you be joining today 6pm ist?
<jonpsy[m]>
jjb: would you be free at that time?, let's meet at the usual spot (zoom link)
<trshtrsh[m]>
Can someone give a short introduction about how mlpack differs from tensorflow and maybe areas where it can be considered better?
<jonpsy[m]>
<jonpsy[m]> "zoq: Just to confirm, would you..." <- im waiting :), let me know if you cant attend
<trshtrsh[m]>
Ehh
<rcurtin[m]>
trsh (trsh): mlpack and TF kind of serve different needs. for one thing, mlpack has a lot more available in it than just neural networks---it has specific k-means implementations, nearest neighbor search implementations, random forest implementations, etc.; those can be awkward to write in TensorFlow (if they are even possible to, I'm not sure)
<rcurtin[m]>
I'd say mlpack has less overhead and is easier to deal with than TensorFlow (for instance, you don't need a session or anything like this, and the set of dependencies you need is lighter so it is probably easier to deploy a trained mlpack model), but as an mlpack developer, I suppose I have a bit of bias 😄
<rcurtin[m]>
at the moment, mlpack does not have GPU support, but that is in progress through the Bandicoot project. for some users, that's a big deal, so we hope to have it soon... but another difference between us and TensorFlow is that we don't have a giant team of full-time Google employees doing development :)
<trshtrsh[m]>
Tnx for the info..
<trshtrsh[m]>
:thumbsup:
<jonpsy[m]>
<trshtrsh[m]> "Can someone give a short..." <- C++: mlpack, Python: Pytorch. Tensorflow who?
<jonpsy[m]>
jk, tensorflow is good for edge AI but that's where its pros end
texasmusicinstru has quit [Ping timeout: 260 seconds]