As a senior data science student, I want to enter the job market with as much knowledge as possible. Easy-to-follow videos like this make that goal so much easier. Thank you!
Thank you for this amazing video! But I have a question. At the beginning, the question was defined as "What is Population Density". But, does not KDE give us the density of a spesific data point instead of the whole population as estimated? Because the result is found as using a data point which does not appear in the results. Therefore, we actually try to understand the density of a spesific point instead of population. Do I get it wrong or was the question generalized?
Sublime. This topic just came up in a data analytics course I'm taking (it wasn't a central theme of the lesson, but I hate not knowing the details sometimes) and this programme is a perfect complement to that. Like others have said, your style is intuitive but not over simplified. In general, I feel like you're striking a great balance between ease of understanding and mathematical rigour.
You integrate cause you're working with continuous functions. It is already normalized since the squared difference could be at most 1. We also want a good estimsted distribution to perform well on other samples from the true distribution. That's why we take the expected error on various samples
Great explanation! Gaussian KDE is great for bimodal and skewed distributions. One downside with gaussian KDE is difficulty accurately modeling distributions with high excess kurtosis.
Thanks for the good explanation about KDE method. could you please make a video about prediction intervals PI that sometimes uses the KDE method? thanks!
Very clear flow of explanation, thank you. I'm thinking that it would be useful to design a hypothesis test for the chosen setup to back up the idea of the final density and so to get an extra information along with the vertical position of the chosen point as of how much proof we have for the final result that is allowed by the number and positions of the known fixed points. More research would be nice.
Clear explanation and easy to follow, thank you! Silly observation: "Integrate over all possible weights of fish. All the way from negative infinity to positive infinity": I'm no ichthyologist or fisherman but I feel negative weight fish ain't an option.
Great content, and very clearly explained. May I just suggest starting from "white sheet" or almost? it doesn't need to be written or drawn incredibly well but the full sheets feel pretty overwhelming
Question, when conducting MC and sampling, can you use a KDE as a valid PDF as opposed to assuming a distribution (e.g normal, log normal, etc.)? Also, could this be considered kind of like a 1-d k means?