@rachel No, your voice is not breaking now.
1 Like
where we can find the kmeans notebook?
1 Like
Jeremy will post it after class
2 Likes
Gaussian distribution is symmetric. Did Jeremy draw only the positive part?
Yes, Jeremy only drew half.
1 Like
Yes and I think it is because distances are positive
3 Likes
I get how batching helps with locality and cache, but I do not quite follow how does it help otherwise? Especially with respect to accelerating for loop?
Some of the math functions are coming from Torch, and others are coming from the math python library… what is the difference? When you use the python math library does that mean that the GPU is not being used? Can you talk about that more?
2 Likes
Last class he mentioned that he would explain when and why to use keras versus pytorch. If you only had brain space for one (in the same way some have only brain space for vi or emacs) which would you pick?
6 Likes
One request for future audio streams if possible. There are occasional max volume pops in the audio track from Jeremy’s mic that are really hard on the ears. I’m guessing this isn’t solvable right now but if you could look into some sort of filter to prevent those in the stream my ears would really appreciate that.
2 Likes
1 Like
lol 
I have pops every 2-5m and does bother ears
2 Likes
was looking for some intuitive motivation for using mean shift clusters. found this lecture note online with some examples thought was helpful as well.
http://vision.stanford.edu/teaching/cs131_fall1314_nope/lectures/lecture13_kmeans_cs131.pdf
11 Likes
This is a totally random question and you don’t have to answer this as its kind of off topic: How do you think the Google acquisition will effect Kaggle (what changes might happen in the future) -> do you have any insight into this?
@dradientgescent @Even Remote listeners, let us know if it improves now that Jeremy has turned the volume down
Is Dynamic Quantum Clustering (http://arxiv.org/abs/1310.2700 and the GPU version http://dx.doi.org/10.1016/j.jcp.2012.08.048) a viable alternative to k-means? It does not require specifying the number of clusters, which seems bogus, and seems an improvement on the mean shift algorithm.
At the moment, if i’m not wrong, you can export pytorch models to tensorflow right? Can you serve the models you built using Pytorch like how tensorflow serving is? I tried to read it up on their forums but there seems to be no way to do it as of now.
Looks like popping might be related to CPU over use with Logitech camera. Found on open broadcaster bug reports.
1 Like