Great video Luis. I think this is a very clear and concise explanation of Gaussian Mixture Models. One thing you could do to improve your videos is to focus on the audio quality. The volume levels varied considerably between parts, as did the sound of the vocals (echo, room tone), and the instrumental covered your voice for the first minute or so. Using a consistent recording setup with some curtains or sound absorbing foam will help to keep the reflections down and give a consistent sound. Using a compressor on your audio track and adjusting your levels both during recording and during mixing to get consistent audio levels will help to keep the volume consistent. A quick A/B listen of each part compared to each other will also help to tell if things are inconsistent and need to be adjusted. I think your illustrations are excellent and the explanations are very clear. You seem to achieve a great balance between giving simplified explanations while still providing correct and accurate explanations. Great job! I look forward to the next video. And I still plan to read over your grokking machine learning book too!
Thank you VERY MUCH. i have been struggling with understanding the GMM for two whole days and no book or video could explain it very intuitively like you have done, i truly appreciate it
After wasting time on all other videos I had lost hope to find a good video on this topic. Amazing use of visuals to support the explanation. You nailed it. Subscribed your channel too :)
Again a very informative video...Can You please make playlist explaining all machine learning algorithms and then deep learning? I know you are busy person...it's just that I and many people like myself really learn from your videos and if they are in order it's really easy to implement and become knowledgeable. Thanks for all your time and great videos.
Thank you! :) I have everything organized by topic here: serrano.academy , otherwise, you can also look at the channel page: ru-vid.com, where a bunch of playlists appear more organized. I hope that helps. Happy learning!
Thank you, and thanks for the suggestion! Definitely been looking at attention/transformers. In the meantime, check out this material by a friend of mine jalammar.github.io/illustrated-transformer/
Thanks, glad you liked it! Check this video out, it’s one I made on reinforcement learning! ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-SgC6AZss478.html
Your description is well explained, with clear visuals and with a good intuitive explanation of the subject. I encourage you to spend a little money on production values, better consistent sound quality, a better less intrusive intro music and these will move you into be there with Statquest and even ThreeBlueOneBrown. Good luck.
Questions: How to use fraction point to create new Gaussians, Example lets say we are in 2D with x1=2, y=8 and we find this point belongs to a Gaussian with 60% next how to use What should I do x1 = 2 * .6, y1 = 8 * .6 sort of ?? Please provide clarity on Hypothesis.
fabulous explanation. Most of the authors try explaining the subject in machine learning mathematically using jargon and symbols which become too hard to understand.
Thanks Luis! very good explanation. Here you assumed that you have two clusters to being with. In many real-life cases (for example: biological datasets), we do not know how many clusters are there. In those cases, I guess we have the number of clusters itself as a parameter, and we have to play with it till we get the right number of clusters, right? If yes then how can we be sure that we got the right number of clusters if we do not know the ground truth? Do we have to employ some kind of nonparametric model for such a case? What's the justification for assuming that each cluster can be modeled by Gaussian distribution?
Thanks Lukesh, great question! There are several methods that can be used to figure out the ideal number of clusters, although most are heuristical methods. A very common one is the elbow method. It is explained here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-QXOkPvFM6NU.html
