Every Distance in Data Science (Almost 100K Subs!)

0:00 Intro
2:19 Euclidean Distance
5:47 Manhattan Distance
9:14 Minkowski Distance
12:49 Chebyshev Distance
15:40 Cosine Distance
19:35 Hamming Distance
20:17 Haversine Distance

[init_yeah]

Honestly you're a treasure in this age of misinformation

[trapItaliana]

i cant believe you have a video for everything

[Kortemaki]

Great video, lots of good content. You do a great job of making the math and intuitions easy for anyone to digest!! I did click in looking for information on Poincaré distance for concave/hyperbolic surfaces and was bummed that it wasn't covered, but I understand it's more like your honorable mention of Haversine/Vincenty for convex surfaces.

I would recommend making a part 2 with some other interesting distances especially for sets! There are lots of possible examples for point-set distances, set-set distances, or even point-point distances with respect to a distribution (e.g. Mahalanobis). Some good examples that do come up regularly in the literature: perpendicular distance (SVMs), Jaccard distance (CV - image segmentation), Mahalanobis distance (Variational inference/ELBO), and Kolmogorov-Smirnov distance (comparing probability distributions).

[peterhopkinson]

100k is so well deserved, your content is so easy to understand and learn.

[ben_jammin]

I'm liking your video. WRT Euclidean distance, when you say it "doesn't make too much sense", it would have been nice to provide a little insight here as to why that is. Haven't finished watching, but thanks for the content!

Edit: Fun and lively. Engaging, with powerful explanations of things. A-Grade teaching :-)!!! Also, nice pace and very calm :).

[videotrash]

Congrats man! Appreciate all the work you've put into your videos and they've helped me out quite often.

[Set_Get]

thank you for bringing up this topic.
i hope you will have time to add/extemnd numerical examples for the less-used distances.
the fact that you break topics into 10 to 20 mins videos makes them more enjoyable.

[pieter]

16:26 I respect that you review and edit!

[Caesarr]

Hamming Distance is useful whenever there's sequence data, like in NLP applications (spell-checking) or bio-informatics (DNA), though it's usually better to use the more powerful Levenshtein distance.

[mikekertser]

mahalanobis distance?
I'm wondering about a choice of right metrics to be invariant to a small distortions, measurement error or scaling

[apterixbr]

one more very intuitive video, thank you! maybe another one in the future about the steps (the intuition) to create your own distance model. ps: let´s go to 500k! you are helping a lot of people to really understand the math using intuition instead of just read what is happening in some powerpoint. we need more people like you teaching us :)

[TheElementFive]

Hamming distance is foundational to K-Modes Clustering, a neat and underrepresented algo. Also, congrats on 100k!

[kirudang]

Congrats, you’re always my savior 😊
Thank you so much for all valuable lessons 🎉

[bryceasay]

That's the best explanation of chebyshev I have ever seen

[sakib.]

Surprised the KL Divergence wasn't mentioned

[ffhgffg]

Thank you for sharing. I actually encountered a case where chebychve distance would come in handy. Too bad that I didn’t know it earlier l

[Break_down]

Excellent content!! Would love to hear how you’d explain Gaussian processes and Gaussian process regression

[paulhofmann]

You forgot the distance of all distances the Kolmogorov complexity ie K(x|y). It’s incomputable but can be very well approximated for many use cases.

[k.alipardhan]

Congrats on 100k.
Would be cool to hear about the Distances which we don't hear about, like obscure ones in biology or engineering (or Computer vision but we have pyTorch docs for that)

Quick shoutout to Jaccard Similarity, which is the Intersection over Union

[rahulshah]

12:08 I believe all Minkowski norms are convex so I don't think it is concave (I could be wrong though but visually it looks convex to me).