Intro to Kernel Density Estimation

This video gives a brief, graphical introduction to kernel density estimation. Many plots are shown, all created using Python and the KDEpy library (github.com/tom.... A .pdf of the presentation may be found here: github.com/tom...
Contents
00:22 - What is kernel density estimation?
01:27 - Kernel functions
03:27 - Bandwidth
04:30 - Silverman's rule of thumb
05:19 - Improved Sheather Jones
06:10 - Weighting the data
07:30 - Bounded domains and reflections
09:18 - Kernel density estimation in higher dimensions
10:02 - The choice of norm
11:11 - Example of 2D kernel density estimation
12:36 - A fast algorithm using linear binning and convolution
15:30 - 2D linear binning
16:18 - KDEpy - software for kernel density estimation in Python
16:51 - References

[@n.sabriozturk6520]

Finally here I found a super video that explains briefly and clearly what Kernel Density Estimation is. Thank you so much.

[@webelod4999]

Thanks man. Glad the video was of help :)

[@jasonhe6947]

I love this tutorial, the pace, example, and visualization are just so great

[@XY-yg1ci]

so straightforward explanation. understand kernel in the first 2 mins

[@peterstanbridge3871]

Thank you so much for this presentation - first time I've been able to even begin to understand this at an overview level.

[@webelod4999]

Awesome! Thanks for leaving the nice comment :)

[@yunfenghu3786]

Thanks Tommy for this amazing video. I am a visual person and this video gives me a clear view of how density kernel works in 1D and 2D using graphs. Your visualization for norms in higher dimension was fantastic. I will use recommend it to my students in the future!

[@webelod4999]

Thanks! I appreciate it!

[@matematikce9490]

Clean, on the the point, good theory/practice ratio. Very much appreciated, thanks.

[@zyflying]

Really great intro, briefly and straight to the point

[@aaronlin8785]

Amazing video Tommy. I couldn't understand KD in a week of Uchicago lectures and you did it in about 45 seconds.

[@PianoMan333]

Great video. I found this topic rather abstract but this makes it a lot clearer. Thank you!

[@carl416]

Relatively clear exp, good. Visuals really make the difference.

[@okokpk123123]

Thank you for your presentation.It is really briefly and clearly.It really helps a lots.Hopes you can share more presentation!

[@webelod4999]

Thanks! The success (in terms of views) on this video inspires me to create more.

[@ali-kadar]

Thank you a ton for the very clear and concise explanation. I like that you go into some algorithmic details nearing the end of the video.

[@ukvaishnav]

Thanks for making this video. Its concise and quick guide to KDEs.

[@michaeljagdharry]

you are amazing, that was one the clearest explanations of a nonstandard statistical concept I have ever seen

[@webelod4999]

Thanks!

[@luismisanmartin98]

This video is absolutely precious! Thank you Tom for taking the time to create this

[@webelod4999]

Glad you liked it. So happy to get positive feedback, since it took some time to create.

[@timuryalta]

This deserves much more views!

[@nengjingding5942]

Finally found a video to get a rough but clear idea what KDE is. Highly recommend!

[@IroXtreme]

Great video, clear and concise - thanks!

[@pcenxyz1838]

Sir thanks for the explaination.Very well explained actually I came here with zero knowledge. Thanks for the explanation and I will definitely use KDEpy in my projects...thanks for saving the day

[@Scoutik997]

This is a very clear explanation of KDE, good job

[@nakko3017]

Thanks for the very clear explanation. ありがとうございます

[@webelod4999]

どういたしまして ! (I used Google Translate)

[@ummesalmamofficial7637]

Thank You Sir for explaining KDE in a simple way.

[@barnabyinteractive]

super well made couldnt ask for anything better lol

[@tgwashdc]

Short, sweet and perfect!

[@marcelsa5191]

Extremely good video! Well explained and nice graphics. Thank you and greetings from Oxford :)

[@webelod4999]

Many thanks!

[@RajeshSharma-bd5zo]

Beautifully explained!!

[@rajanalexander4949]

Clear visualisations, succinct and lucid explanations -- fantastic video. Thanks!

[@tymothylim6550]

Thank you very much for this video! It was very easy to understand (although this topic is still quite new to me). The use of graphs helps a lot with the explanations!

[@ZinzinsIA]

Very nice, even if i did not get the part about the linear binning and what it is exactly

[@ZinzinsIA]

And very nice for the library btw !

[@Brumor]

Great video, thanks!

[@aman.bansal]

Thank you for making this helpful video.

[@Ariel-px7hz]

Excellent video. Thank you!

[@samuelfischer5131]

This is awesome. Thank you for this overview!

[@JayPatel-et4vi]

Best video for KDE

[@user-pq6ed3zs5k]

Great visualizations

[@snehagaikwad2655]

Thank you so much for the video! It was easy to understand conceptually!

[@jeffreychong3467]

Watched about 10 videos, only this one clicked for KDE.

[@rhodesengr]

Thanks for this video. It makes the concept very clear. Other videos, not so much. I have an application where I would like to use 2D KDE on data sets that are set of point on an xy plane. My goal is to fit a 2D Gaussian to the data and then compare goodness of fit for different data sets. I believe I first need to generate a density function for the data and then fit the Gaussian to the density function. KDE looks like a good way to generate the density function. I would prefer to do this in Excel so an Excel plugin would be ideal. I am not really setup (or proficient) to do regular programming in Python, C, or whatever.

[@singlebinary]

Excellent video and clear explanation. Please keep making more!

[@NadavBenedek]

Great audio quality

[@webelod4999]

Thanks. For anyone curious, the microphone I use is Audio Technica AT2020 USB+

[@stephengargan3907]

super informative, nice job!

[@webelod4999]

Thank you!

[@jaantollander]

Great tutorial. Thank you!

[@qwqsimonade3580]

thanks for the dedicated video

[@martinwutke3386]

Thanks for this very good explanation. Will definitely look into your library. Best Wishes

[@webelod4999]

Glad it was helpful!

[@alejozen3457]

Great explanation. Thank you for the effort.

[@capricacity]

I wish I saw this before completing my PhD. This would have made the process "smoother" get what i mean? HAHA!!!

[@zenchiassassin283]

lol, congrats for your PhD too

[@ebrahimfeghhi1777]

Thank you, great explanations!

[@lifestoriesfromearth6271]

Thank You Tommy for this wonderful explanation. :-)

[@delinyahkoning6882]

What a nice video this is! Super clear.

[@bernardoamorim9182]

amazing tutorial, thank you very much for the video and the library :)

[@powerchucho007]

Thanks a lot. Great explanation!

[@svendavidsson]

Great explanation!

[@juandavidcaicedoms7686]

Thnks for this video! It’s a really good explanation, super helpful!

[@webelod4999]

Thanks man, I appreciate it!

[@gekkejunior3262]

Clear. Thank you a lot!

[@himanshudalai1028]

Thank you so much for the video. Loved it.

[@h-hugo]

Very nice lecture!

[@felipefavadelima]

Thanks for your video! Very well explained.

[@webelod4999]

Glad it was helpful!

[@chenghungchou9521]

Amazing easy to understand!!!!!!!!

[@pranavkumar9782]

Is it possible to sample from the KDE after fitting, either in sklearn or KDEpy, apart from the usual method of going to a point x_i and sampling from N(x_i, h) if the kernel is Gaussian in the KDE ?

[@webelod4999]

Not that I know of. You could use the Inversion method and the CDF of the returned PDF, but "the usual method" that you mention is equivalent to sampling from the PDF.

[@khubaibraza8446]

Thank you so much, Super clear explanation.

[@makimakiwii]

Very helpful. Thank you so much!

[@giuliofederico7638]

Perfect explanation

[@laxmanbisht2638]

precisely explained

[@Colegial24]

Excellent video! Extremely helpful!

[@Abafoteq-Ltd]

Wow..... wonderful. thank you so much. this was indeed very helpful.

[@webelod4999]

Glad it was helpful!

[@aparnamuralidhar5413]

Hello there. I tried using your KDE package for my work. Used FFT KDE. When i was trying to evaluate the model with some data-i got an error-'Every data point must be inside the grid" . could you elaborate on this,please?

[@webelod4999]

If you have a data point at 0, say, and you grid ranges from 1 to 5, then you will get this error. The data point is outside of the grid. Best to let KDEpy create the grid for you. It automatically sets up a reasonable grid.

[@diwakarns1600]

Thank you..I did not understand what a norm is, can you explain a bit more on that? Thank you!

[@webelod4999]

It's basically a measure of distance. A generalization of abs(x) in one dimension. See Wikipedia :)

[@rajm3496]

genius...happy that I found this :-)

[@raduiulia4034]

Amazing video!

[@bean217]

9:50 why is the sum only normalized by 1/(h^d) and not 1/(N * h^d) ?

[@nallakrishna8796]

finally, i found an amazing lecture on kernel density estimation thanks a lot . but i have one query how it can be used to find the anomaly detection. sir can u please make one lecture about this topic otherwise can u please recommand me some good references for KERENEL DENSITY ESTIMATION FOR ANOMALY DETECTION

[@nassehk]

What a great video. Thank you.

[@mahadeibnsalam6735]

Great content!

Sorry for the dumb question but why in the first formula X is subtracting Xi? What it does mean?

[@webelod4999]

If I have a function f(x), then subtracting 2 will shift the function. So f(x-2) shifts the function to the right by 2. When we subtract the data point x_i, we shift the kernel function so it lies "on top" of that data point.

@@webelod4999 Thank you very much 🙌🙌🙌

[@richardtarbell946]

This is king shit right here.

[@abdizinab7934]

Thanks you some much, please Can you sent me the programs of all those representations

[@michaelsongbai]

Nice tutorial! Thanks!

[@lilaberkani4376]

Thank you so much for your video, it helps me a looot

[@canmetan670]

Thanks man. Great video.

[@TheOfficialJeppezon]

Please make more videos!

[@MiroLogie]

Thanks for the video, what you used to do the plots BTW

[@webelod4999]

This is a nice extension/improvement! I considered looking at moves too, but determined that (1) getting and preprocessing the data and (2) potentially optimizing over both pokemon and moves would be too much work for a weekend project. If anyone wants to take this even further, I think your ideas are good. At the end of the day the most interesting thing might be to train a reinforcement learning algorithm (like alphago / alphazero et al), but that would be a lot of work!

[@teresaebernardo]

Does the size of the grid make a difference?

[@webelod4999]

Yes. The finer the grid, the better the results. In KDEpy the default is 1024 grid points.

[@zilezile4942]

Good evening everyone, 🔵 Discover now our books and training that we have produced for you on our site. : www.amikour.wordpress.com 🔵 Click here to go directly to our books and training. : amikour.wordpress.com/nos-formations/

[@realreactteseract6261]

Amazing, really!!!!

[@LeeLeeCode]

Thank you!

[@canernm]

Thanks for the video ! Quick question, are the kernel functions probability density functions? I know the fulfull their properties, but is that enough to make them PDFs? Thanks in advance.

[@webelod4999]

They are, yes. If they fulfill the properties, they are PDFs by definition.

[@shivshankarkeshari6604]

4.07- 4.14 how can I do similar in my py project?

[@juheesingh1157]

Very hepful video 😊

[@justforsynchtc]

Having to implement this and don't understand the "discrete convolution (possibly by fourier transform)". Any pointers?

[@webelod4999]

Look to wikipedia for information about discrete convolution.

[@zahrahsharif8431]

Hi, how would you interpret a kde if the x axis is probability and the y axis is density?

[@webelod4999]

As a prior distribution in Bayesian statistics.

[@dayy14]

Thanks a lottttt!!!

[@thomasalderson368]

Liked!

[@cendradevayanaputra7150]

do you have review of Density Estimation?

[@webelod4999]

kdepy.readthedocs.io/en/latest/literature.html

[@cendradevayanaputra7150]

@@webelod4999 thank you

[@jeanny2852]

what is the difference between x and xi?

[@webelod4999]

x is a continuous variable (the domain), while the x_i's are the observations in the sample.

[@Borzacchinni]

Do you happen to be from Norway?

[@MiroLogie]

What you used to plot the data ?

[@webelod4999]

matplotlib!

[@MiroLogie]

@@webelod4999 thank you

[@SLee-xj4jn]

Best

[@jg9193]

Wow...

[@43SunSon]

pika pika