What is kernel density estimation? And how to build a KDE plot in Python? | Seaborn KDEplot

[@KimberlyFessel]

If you enjoyed 😄, please subscribe and check out my full "Introduction to Seaborn" playlist: kzbin.info/aero/PLtPIclEQf-3cG31dxSMZ8KTcDG7zYng1j

[@laxmanbisht2638]

Great! I want to do KDE pot for a road network? Have your written this sort of code?

[@KimberlyFessel]

@@laxmanbisht2638 I haven't done KDE plots for networks, but that sounds interesting

[@snelethuhashibi8254]

Hi Kimberley, Do you know how I can label the KDE contours?

[@eatbreathedatascience9593]

Hi Kimberly, great video. Could tell me what you used to produce the slider for the bandwidth bw ? Thanks very much. !

[@Tntpker]

Arguably the most underrated video on KZbin.

[@KimberlyFessel]

So glad you enjoyed it! This is the first public video I made for KZbin.

[@Gebev]

Thank you very much for making the effort to visualise how the distributions add up. That made all the difference.

[@jon7187]

The de-jargon-ification of terms is reason enough to watch. The visuals and step by step walkthroughs are the reason I subscribed!

[@KimberlyFessel]

Awesome -- thanks!

[@msarmad5217]

Wow. I am speechless. Haven't seen content of such quality on KZbin for a long time. Stumbled upon it and Learned a-lot today. Kudus Kimberly!

[@KimberlyFessel]

Oh wow -- thank you! So glad you enjoyed the video and learned from it -- cheers!

[@rajatgupta7344]

I wasted 2-3 days just clearing this. I got your video really to the point, explaining each and everything I need. Really appreciate yours affords. Thank you

[@KimberlyFessel]

Very glad to hear my video helped!

[@aadilvalconi9972]

The only thing I was confused about was how to interpret or read the bivariate kdeplot. 'You can thing about ring as higher coming closer and lower being in the background" Thank you for that explanation.

[@yvessantosborges3491]

I am form BR 🇧🇷. I liked the simplicity of the video and the material used - a lot of quality and accuracy: from basic to advanced. CONGRATULATIONS

[@KimberlyFessel]

Hello 👋🏼 and thanks for stopping by. So glad to hear you enjoyed the video!

[@shubhamjain3028]

The content is pure gold. Please don't stop making more videos. Your channel will be on top one day I am sure

[@KimberlyFessel]

Thanks so much and glad you enjoyed the video!

[@22niloc]

Excellent - Very clear and extremely comprehensible. I"mcurrently studying DataViz with seaborn and your tutorial has just made sense of KDE.

[@Louis-Li-r5b]

Dude, this video is so sick. I have been so confused about the concept and the addition of Gaussian distributions explains it well for people to understand in a split second. I love it :)

[@KimberlyFessel]

So glad to hear that! I remember making that animation and thinking, “yeah, this makes total sense!” 😁

[@zphyr3737]

There is hidden great genius in this simple explanation of this complicated topic. I love it

[@yuxrazafar8532]

Thank you so much! It helped me alot in understanding the underlying concept. I couldn't find other tutorials for kde that teaches so on point!

[@KimberlyFessel]

So glad that helped you - cheers! 😀

[@rutwikhiwalkar9583]

I was struggling with visualizations and just came across the best series on YT, coincidence? I think not. Btw the series is amazing!

[@KimberlyFessel]

Excellent -- so glad to hear this series has been helpful to you!

[@geekyprogrammer4831]

Perfect explanation of Kernel Density Function. I wish you have more subscribers in the upcoming future.

[@KimberlyFessel]

Thank you very much 👍So glad my explanation was clear!

[@thomasdehee9626]

Thanks you SO MUCH I bumped into this concept in a Python course and was looking for a concise explaination about KDE and was getting really anxious as anything I was finding was overly complicated You have been really clear and concise, amazing ! Suscribed

[@jayantdikshit4181]

One of the best series on Seaborn. Thanks for making such a great content!!!

[@KimberlyFessel]

Excellent -- glad you are enjoying it!

[@nyxhemera]

Presentation of KDE is really smooth and clear. Thx

[@KimberlyFessel]

Excellent - glad you found it helpful!

[@Esmeralda-ut1br]

I'm here to click the like button in every videos of this serie. Hehe these are very clear and concise. Like, like , like

[@chizzlemo3094]

the last bit, 2 separate color palettes, you didn't explain!! please add a note. Very useful video though, thank you

[@adamatkins8496]

such an underrated channel. thank you for uploading this quality content for free.

[@KimberlyFessel]

Why thank you! Glad you enjoyed the video 😄

[@raulitorc]

This helped me to better understand and apply the KDE concept to my statistical work, Kim. Thanks a lot

[@balajikannan7393]

Whoa...by far the crisp and concise explanation on KDE. I enjoyed learning and replicating it. Thanks.👍

[@KimberlyFessel]

Awesome -- glad you enjoyed and were able to replicate! 😄

[@ethanspitips5903]

Great video on getting started with KDE plots in seaborn! Super easy to follow along, understand, and replicate the plots in a notebook. Nice job :)

[@KimberlyFessel]

Thanks much! Great to hear you could easily follow along 😄

[@anagabrielamoralesona2579]

This video was clear, concise, and easy to follow and understand. Thanks! :)

[@KimberlyFessel]

So good to hear - glad it helped!

[@chicochinochao5493]

just started my python journey, very good tutorial on KDE plot

[@KimberlyFessel]

Glad you enjoyed it!

[@pedroluisalcazar9285]

Great Video!!!!! thanks, your explanations and the resources you use to make your concepts clear are great!!! so nice to have people like you on You Tube, Thanks, and best regards from Bolivia

[@pedroluisalcazar9285]

One more question please, what program you use to make that bandwith graph where you move the bandwith and see how the graph changes?

[@diegoac090]

Helpful for my thesis, blessings to you

[@KimberlyFessel]

So glad to hear that -- best regards for your thesis!

[@arfanwicaksono8590]

really cool, the explanation is really easy to understand

[@KimberlyFessel]

So glad to hear it was helpful!

[@thejahcoop]

Finally, good videos on seaborn. Thank you!

[@KimberlyFessel]

You're welcome! Glad they are useful!

[@shubhamgawali8030]

what a Great explanation it was, I fall in love of knoweldge, subscribed channel, lot of respect from india

[@shubhamgawali8030]

I just have a one question for mpg dataset sns.histogram(cars.horsepower, kde=true) is it equal to the plot we obtained from sns.kdeplot(cars.horsepower)

[@KimberlyFessel]

Thanks very much for subscribing! And yes, you should get the same KDE plot; however, the sns.histplot(cars.horsepower, kde=True) will also give you a histogram under the KDE plot. The histplot will also likely truncate/cut the edges of your KDE plot.

[@sriratna9211]

You don’t have idea how you helped me, thanks for the good work !

[@KimberlyFessel]

So good to hear! Very happy to help 😀

[@charmendro]

Very well explained! Learned alot about KDE as a whole as well

[@KimberlyFessel]

Oh so good to hear that - KDE is pretty cool!

[@juanpi.hojita]

This video saved my semester, thank you so much!

[@KimberlyFessel]

So good to hear. Glad it was helpful and good luck with the rest of your semester!

[@crbMKE]

Thanks! This helped me understand how the FSeq2 software works!

[@Raiden_Amani]

Thank you so much. Was really looking for something more effective than a simple gaussian.

[@thesenate8268]

Great explanation and underrated channel.

[@KimberlyFessel]

Thanks much! 😄

[@deldridg]

Great tut and thank you. I might have a go at generating some multivariate KDE plots of K-means outputs to see how they might add some interpretative value. Many thanks from Sydney - Dave

[@jongkargrinang8012]

Your all videos are very helpful. Thank you very much.

[@puhozavrik]

So happy I found your channel! Amazing content, thank you

[@KimberlyFessel]

So happy you found my channel as well - welcome! 👋

[@paulaaimar5555]

Amazing video! You explain some complex concepts very well, thank you so much, keep making this videos please! You are the best!

[@KimberlyFessel]

Thanks very much! So glad to hear you enjoyed the video, and I definitely will keep making more 😄

[@VikasKM]

Super simple and easy.. to the point. Must watch

[@KimberlyFessel]

Awesome -- glad you liked it!

[@espua72]

good job on explaining this in such simple, easy to understand way. keep up the good work

[@KimberlyFessel]

Thank you! Will do!

[@Persona_Incognito]

This is a superb presentation! Thanks so much for posting it!

[@KimberlyFessel]

Oh, thank you! Glad to hear you enjoyed it. 😄

[@storyinmyhead]

clear and concise explanation, thanks kimberly!

[@KimberlyFessel]

You're very welcome -- glad it was helpful!

[@adityasahu96]

Such a simple and perfect explanation. Thank you

[@KimberlyFessel]

Awesome -- glad to hear you enjoyed the explanation!

[@guganesan.ilavarasan]

This is actually amazing!

[@wolf-lvccm]

Hi Kimberly, thanks for the great video. I just subbed your channel. A quick question: Any tips on how you made possible what's going on at 4:28 and onwards? It reminds me of filtering in Tableau, but I have no idea how to do it with Python.

[@simbol5638]

Thank you, now I have a grab of what KDE actually do.

[@KimberlyFessel]

Excellent! Glad to hear that 😁

[@sabarivishnujayanthanjaikr4653]

Perfect explanation. Thanks, Kimberly!

[@KimberlyFessel]

Excellent -- you're welcome!

[@ankitbioinfo]

very nice. I learnt something new today from your video. Great!

[@KimberlyFessel]

Excellent - always a good day to learn something new! 😄 Cheers!

[@colemanvollrath7180]

Excellent video! Thanks so much Kimberly!

[@qwqsimonade3580]

how can plot like 8:25 to put different color on the same plot? Thank you

[@ling6701]

Thanks, very good explanations, I needed this.

[@supamdeepbains5172]

Thank you so much. You made it look very easy and clear :) good job!

[@KimberlyFessel]

So happy to hear that my video helped make the KDE more clear -- thanks!

[@andreazecchi812]

Kimberly, you are Simply the best! Greetings from Rome, in Italy👋

[@KimberlyFessel]

Thank you very much, Andrea! 👋

[@cristianalzateurrea1948]

Thank you!!! The best video of KDE plot :)

[@KimberlyFessel]

Thank you! Very glad you enjoyed it

[@ST-rq8gk]

What an awesome video, thank you!

[@KimberlyFessel]

So glad you enjoyed it -- cheers!

[@katharinastck]

Thanks!

[@KimberlyFessel]

Amazing - thanks to you as well! Glad you liked the video 😁

[@nassarzinho]

Great video! I would like to use the kde over a map, e.g., given a set of lat and lon, how to plot the kde using the map mask? Do you have any video teaching about it?

[@leidycastro3629]

I have the same question :/

[@londonmackerel7462]

I am enjoying this, thank you, but I wonder if some of the code is now a little out of date. For example to plot a bivaraite distribution I had to use (eventually) sns.kdeplot(data=cars,x='horsepower',y='mpg'); sns.kdeplot(cars.horsepower, cars.mpg) just did not work. Strangely, sns.kdeplot(cars.horsepower) {the one dimensional case} worked. Any insight here would be appreciated.

[@saurabh3614]

Nice explanation thanks! , how did you added a gaussian kernel for each a data point? 1:22?

[@KimberlyFessel]

Thanks! That animation is a series of individual matplotlib figures that I created basically with a for loop, adding up normal curves with standard deviation one and each point as the mean. Once I had all individual plots, I just used PowerPoint with a fade transition between each slide and recorded my screen. You can check out the code for the matplotlib figures on my GitHub page: github.com/kimfetti/Videos/blob/master/Seaborn/02_KDEplot.ipynb

[@tahfim37]

That was really helpful, great work.

[@KimberlyFessel]

Very glad to hear it was helpful -- thanks!

[@MouradBENKADOUR]

The `bw` parameter is deprecated in favor of `bw_method` and `bw_adjust`. Using 5 for `bw_method`

[@thejahcoop]

How would you plot multiple KDE on the sample plots? Specifically, if I have columns as different conditions and each row a measurement of that condition.

[@KimberlyFessel]

The newest version of the kdeplot (Seaborn 0.11.0) allows for the hue argument to split out data into different color groups based on a categorical variable. You can also just plot two kdeplots on the same figure: sns.kdeplot(X) immediately followed by sns.kdeplot(Y) in the same Jupyter cell or before doing plt.show(). That would just layer one plot on top of the other -- assuming you have similarly scaled values.

[@piiopolive7084]

This is an awesome video, thank you so much for sharing.

[@KimberlyFessel]

Why, thank you! Very glad you enjoyed it. 😄

[@elastvd7503]

Great seaborn tutorial , thanks ❤

[@yashparanjape9091]

Superb video! Can you please guide me on how to plot density plots from an ongoing video, instead of pre-fed data?

[@KimberlyFessel]

Thanks very much! Yes, I have started using some external data in a few of my recent videos. One thing you could definitely try is loading your data into a pandas dataframe and then doing a KDE plot for your column(s) of choice. Just keep in mind that if you have a lot of data you might need to sample it. I will also consider making a follow up video for this one! ☺

[@yashparanjape9091]

@@KimberlyFessel Thanks! Yes, that would be great!

[@bubblebath2892]

How do you interpret these plots is the most important part for me ...can you please include that in your videos !!

[@KimberlyFessel]

Great suggestion! Perhaps I can make some videos about how to diagnose the information seen in the plots 🤔

[@imdadood5705]

Another banger!!!! Thank you for this!!

[@KimberlyFessel]

Most welcome! Always striving to make more #bangerz! 😄

[@fahadreda3060]

Nice video, very informative , Thanks Kimberly

[@KimberlyFessel]

Thank you!

[@paulomur4]

When using KDE to create maps, am I creating a heatmap or a hotspot map? I think KDE is linked to heatmaps and not hotspots...is it correct? Thanks

[@amanvishnoi2721]

Thank You very much for this wonderful video

[@vladimirrav]

Outstanding presentation!!!

[@KimberlyFessel]

Thanks so much -- glad you liked it!

[@user-wr4yl7tx3w]

this is really well explained

[@Daijyobanai]

Could someone explain to a n00b how we are using "cars = sns.load_dataset('mpg').dropna()" ? *dropna* is a Pandas method but we have not imported Pandas, only seaborn, so how does it work to chain dropna onto seaborn.load_dataset() ? it shouldn't be available? My understanding is that dropna has to be available to load_dataset, but dropna is not in the seaborn API, only in pandas API.

[@chibutza9092]

Great video! Very helpful.

[@EveGabyGT]

Hi, how can I get the KDE function values? No, the initial data, but instead the function created from it. X being the data and the function being f(x)=KDE plotted

[@rakshitjoshi823]

Hey, can you provide any help with kdeMultivariate in statsmodles library. I can't seem to plot the graph after evaluating the gaussian kernel. Thanks

[@usmansaeed678]

Can you please make a video explaining in detail about Kernel density estimation? Thank you.

[@KimberlyFessel]

I will consider it!

[@MeetPatel-ps1hh]

It is possible to get the PDF's estimated equation?

[@KimberlyFessel]

Unfortunately, no, not with Seaborn. The KDE gives you a numerical estimation of the PDF but not an analytic expression. The theory is based on adding up Gaussians, however. So I suppose if you worked out the bandwidth, you could have an analytical expression with one Gaussian term for each datapoint, given that you use a Gaussian kernel.

[@MeetPatel-ps1hh]

@@KimberlyFessel thank you!

[@azhar145]

Best explanation. Thank You.

[@abdullahbasaran2538]

THANK YOU FOR YOUR VIDEO How can I find bandwith gif 4:45?

[@KimberlyFessel]

You're welcome! I made that bandwidth video with a Jupyter Widget using the package ipywidgets (ipywidgets.readthedocs.io/en/latest/). You can find my code here: github.com/kimfetti/Videos/blob/master/Seaborn/02_KDEplot.ipynb -- that widget is in cells #47-49. Then I just took a Quicktime screen recording of myself interacting with the widget to overlay on my video. Thinking I will likely do a video about ipywidgets in the future!

[@varunjayananandakumar1663]

how to solve the bivariate part i'm getting type error as only 0 or 1 positional argument can be passed onto kde plot???pls help

[@Kuzmar48]

Hi Kimberly , Nice Video very well explained, just a quick question since im new to visualization, can do we integrate these KDE plots to a folium map?

[@ivanrafaelvidaurrethofehrn9118]

Hi, many thanks for the videos! You have one new subscriber for sure. I have a question, is it possible to have in one graph the histogram of the hole population of the data and the kde of one variable of the same data on the same graph?

[@sartyakimanna9986]

Could you please guide me on how you build the code of bandwidth scale, as when you are increasing and decreasing it by mouse the graph is changing and it is looking cool!!

[@KimberlyFessel]

Glad you enjoyed that! That's called a Jupyter Widget and I created it using the package ipywidgets (ipywidgets.readthedocs.io/en/latest/). You can find my code here: github.com/kimfetti/Videos/blob/master/Seaborn/02_KDEplot.ipynb -- that widget is in cells #47-49. Then I just took a Quicktime screen recording of myself interacting with the widget to overlay on my video. Maybe I will create a tutorial video about making widgets in the future!

[@josephkarpinski9586]

Great set of Python videos! An idea, if you haven’t already done it. Consider doing videos on running Python under VSCode and VSCode Insiders. Think your viewers would find it of interest. Best Regards... 👨‍💻👨‍💻👨‍💻

[@KimberlyFessel]

Thanks very much for the suggestion! Haven't done anything like that yet, so I'll consider it. Cheers! 😄

[@SVannyatAth]

Thank you so much! , may I know if we can change legend position of kde plot?

[@KimberlyFessel]

The move_legend function may be useful for this: seaborn.pydata.org/generated/seaborn.move_legend.html

[@cw9249]

can you explain how to get the optimal bandwidth? that seems to be crucial to get the "correct" number of clusters when trying to cluster data

[@KimberlyFessel]

Absolutely. Finding the appropriate bandwidth definitely depends on your situation. Sometimes a smoother curve makes more sense; sometimes you need to see all those "wiggles." The methods I mentioned in the video (Scott and Silverman) attempt to give you the optimal bandwidth - but overall both tend to yield bandwidths that are too large to see the finer details of your data (more information here: stats.stackexchange.com/questions/90656/kernel-bandwidth-scotts-vs-silvermans-rules).

[@FunkToFunkiest]

You 're great Kimberly!!

[@KimberlyFessel]

Thanks so much! Cheers 😀

[@cyrilh7625]

Thank you very much! Very useful video:)

[@KimberlyFessel]

You're welcome -- glad it was helpful!

[@cristinaseg7625]

Really useful video, thanks :)

[@KimberlyFessel]

Most welcome! Glad it helped 😄

[@ChristianGardner]

Would you make an updated video? Some of the functions are slightly different now I believe

[@KimberlyFessel]

Iknowright? The code change so quickly! There was a big update (seaborn version 0.11.0) a few months after I uploaded this one. Will consider creating an updated video!

[@ChristianGardner]

@@KimberlyFessel even if not, I ended up figuring it out .. but may help some people! Great videos!

[@jihan6780]

Thank you! Great video.

[@ManishKumar-rz9ub]

Thanks for the video. how you have created those animation in beginning?

[@KimberlyFessel]

That's some good old-fashioned mathematics and Matplotlib. 😄I basically create Gaussian kernels myself and add them together. You can check out the code here: github.com/kimfetti/Videos/blob/master/Seaborn/02_KDEplot.ipynb

[@adityadubey9565]

Nice explanation!! But can you please explain how we got that last plot... with two different colors?

[@KimberlyFessel]

Sure thing - I basically just split up the 4 and 8 cylinder cars and did separate KDEplots for each of them. You can find the code for that visual on my GitHub page under the "Bivariate KDEplot by Category" section (github.com/kimfetti/Videos/blob/master/Seaborn/02_KDEplot.ipynb) In a newer version of Seaborn (0.11.0), the shade_lowest argument has now been deprecated, but you can just remove that part of the code or switch over to the new syntax "thresh".

[@gouthamansaravanan7692]

Hi Kimberly, Thank you very much!!

[@AlvessSavio]

Kimberly, do you know any material where I could find info about the interpretation of density? My graphs have values over 1.0 and i'm kinda confused

[@KimberlyFessel]

Totally! This is definitely tricky. The KDE estimates the continuous probability density function (pdf) of the data, and it is definitely possible for a pdf to go above one. Instead of thinking of the density like a probability for a discrete set of values, we need to integrate the pdf to come up with probabilities. So the integral, or area under the curve, tells us the probability that our variable will be within a specific range of values. The area under the curve is required to integrate to one, but the curve itself can go above one -- especially if the curve has a particularly small width. (Just try multiplying the cars' horsepower by a really small value like 0.0001 to see those density values get really large! 😄)

[@KimberlyFessel]

The notes section of this resource describes this as well: amsi.org.au/ESA_Senior_Years/SeniorTopic4/4e/4e_2content_3.html#:~:text=A%20pf%20gives%20a%20probability,curve%20reflects%20the%20relative%20probability.

[@AlvessSavio]

@@KimberlyFessel Thank you so much. This helped a lot

[@harshagarwal8170]

thank you so much...... superb explanation

[@KimberlyFessel]

So glad you enjoyed it!