How DTW (Dynamic Time Warping) algorithm works

[@darkmurphyy]

I loved the last slides! I was able to understand the algorithm in less than 5 minutes, something I wasn't able to do after reading for 30 minutes a poor explanation about it. Thank you for your video!

[@tkorting]

Thanks a lot for your positive feedback.
Please like and share the video and subscribe to my channel.
Regards

[@tynolayco895]

absolutly

[@Metaloid-wv4kz]

agree

[@heartbeatpoetry7704]

Your videos are very clear and cover things better than much longer videos. Please make more!

[@tkorting]

+heartbeatpoetry many thanks for your positive feedback. Please like and share the video and subscribe to my channel. What you said is exactly what I intend to do in this channel. Create short videos with the core of what should be explained. Best regards

[@xiangzhang8901]

The most clear tutorial on DTW I ever seen. Thank you.

[@ppalutube]

Wonderfully explained! Complicated formulae turned into simple joy 😁

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@AdzSONLINE]

Been sat here for a few hours trying to get my head around these journal articles and getting nowhere. Thank you

[@tkorting]

Many thanks for your feedback
Please subscribe to my channel
Regards

[@amanpreetchander7386]

Very well explained. I really appreciate that you have mentioned the references.

[@_jojo11]

Such a clear, concise explanation. This video helped me understand the concept in 5 minutes, after having spent hours reading up convoluted blogs and still not grasping it.
Thank you so much

[@Noel_13]

Excellently explained. Great job!

[@GISDoctor]

OMG! I was looking for material on this and look at where I landed!!!! Thanks Thales!

[@tkorting]

Thanks! Please don't forget to subscribe to my channel ;)
Um abraço!

[@theecat3689]

wow been going through papers and couldnt understand anything until this video. Thank you so much!!

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@galva1083]

Excellent explanation, and I love the details in the example. Thank you!

[@tkorting]

Thanks for your positive feedback.
Please subscribe to my channel, and like and share the video with your peers.
Regards

[@amvsekai]

Pretty cool.. nice and easy explanation ❤

[@heteromodal]

Thank you for this! So clear and concise!

[@tapaskar1]

Wow !! Thank you so much for the explanation. I understood the algorithm perfectly because of those examples and I think it covered all the test cases I had in mind haha. Keep it going !!

[@tkorting]

Thanks for the comment, please subscribe to my channel

[@vietnguyenhoang8671]

Very understandable, have a nice day. Thank you!

[@andreoulis1996]

Your example helped tremendously, thank you very much.

[@tkorting]

Good to know.
Please subscribe to my channel
Regards

[@JohananJoysingh]

The visuals were amazing.. :) Example was really useful.. !

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@tzhern]

very clear explaination

[@AB-bp9fi]

Great job. First author from references is a person from Japan, no surprise, they like sudoku 😁.

[@muathalmoslem5803]

It is a clear explanation. Thank you for the great video

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Like and share!
Regards

[@escrituratalks]

Great Explanation, this has made it easier to understand

[@tkorting]

Thanks for the feedback.
Please like and share the video and subscribe to my channel.
Regards

[@harikumarg2470]

Great explanation. The figures helped to understand easily. Thank you so much

[@tkorting]

Thanks for your feedback
Please like/share/subscribe
Regards

[@TejasEkawade]

Brilliant explanation!!

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@gabriellara7456]

excelente explicação e referências ao final do vídeo. obrigado Thales.

[@madhu1987ful]

Excellent step by step explanation

[@Arglass123]

It's a great explanation, thank you. An explanation of the min() component would make it better.

[@mariorubio9809]

Very good explanation, thank you!

[@tkorting]

Thanks a lot. Please subscribe

[@foobar1672]

Good explanation. Thank you!

[@tkorting]

thanks for your feedback
please like/share the video and subscribe to my channel
regards

[@haroldsu1696]

Thank you, Sir

[@mssabaay]

Thank you so much for making things easy to understand.

[@tkorting]

Thanks for your comment.
Please, subscribe to my channel!
Regards

[@juanotavalo]

What a great explanation thank for this video!!

[@andreas16426]

Excellent presentation! Thank you so much.

[@tkorting]

Thanks for this positive feedback
Please subscribe to my channel
Regards

[@zzzwhite9997]

Great job! Thank you!

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@Csartreweghhgf]

Nice example!

[@tkorting]

Many thanks for your feedback
Please subscribe to my channel
Regards

[@havalsadiq3655]

Very informative post, this is what I was looking for long time. Thanks alot

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@robertocorti4859]

Nice explanation! Thank you so much!!

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@ravibohmanntridapalli7236]

bom demais, o sotaque deixa o video melhor ainda kkkkkk

[@martinschidozieezuma6741]

Beautiful. Thanks

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@ankurankur9843]

That's a great explanation. Thanks very much..

[@tkorting]

Thanks a lot for your comment. Please like/share/subscribe.
Regards

[@myhandle364]

Great explanation. Very easy to understand. Thanks.

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@alanrgo1]

Amazing explanation. Obrigado!

[@schrodingerscat3912]

beautiful explanation. thank you

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@taotaotan5671]

Thanks. This reminds me of pairwise sequence alignment...

[@tkorting]

Thanks for the feedback, please subscribe to my channel

Great work! Thank you

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@hamedzarei4197]

Dear Thales,
Thanks for your teaching and sharing. I applied the algorithm using the A and B matrices. But I found that there is some mismatch array in the filled matrix. for example, the corresponding distant value for the A8=5 and B7=4 is (5-4)+min(15,19,14)=15, while in the filled matrix is 14. I was wondering if you could let me know which is the correct one?

[@supriyarao666]

Well explained....Thank You.

[@tkorting]

Many thanks for your feedback
Please subscribe to my channel
Regards

[@thesoulofsea6840]

Very clear, thanks so much

[@tkorting]

+The Soul of Sea thanks for your feedback. Please like and share the video and subscribe to my channel. best regards

[@Lisandro1jimenez]

Excellent! Very clear.

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@konstantinosbourantas7999]

Thank you for sharing!

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@bozhang6455]

very nice explanation!

[@tkorting]

Thanks for your comment.
Please like/share/subscribe.
Regards

[@OliverLonghi]

Bem legal. Parabéns pelo cuidado em produzir o vídeo com tantos detalhes.

[@tkorting]

muito obrigado!
por favor, inscreva-se no canal.
um abraço

[@nemdenemam9753]

Thank you very much. It is an amazing explanation.

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@michaelbaudin]

Just great! Thank you.

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@George-lt6jy]

in the backtracking part, how do you know that you will find the optimal path with only that local search algorithm?

[@mohammedabdallatif22]

Awesome, thank you Sir for the clear explanation

[@vicwang7582]

Thanks a lot for the clear explanation!

[@tkorting]

+Vic Wang thanks a lot for your feedback. Please like and share the video and subscribe to my channel.
Regards

[@krishnasashank6116]

Great video!
Thanks a lot!

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@fathahimuthedath9778]

Thanks. Well explained. Is there any algorithm that can compare pattern of matrices.

[@js913]

Fantastic!! Thank you so much

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@quantummath]

Great job Thales!
Greetings from Hamburg.

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@suhitaghosh4381]

thanks for the precise video!

[@tkorting]

Thanks for commenting.
Please like and share the video and subscribe to my channel.
Regards

[@nabajitdey]

Thank you for this explanation :) You saved my night!

[@samSamsem]

Every thing was clear up until the minute 3:10. I have a question from that part. Why is did i and j became those numbers? meaning the numbers for i became [1, 3, 4, 9, 8, 2, 1, 5, 7, 3] and the same way for j where did we get those numbers from? those are indexes of time points for each of the time series right? why aren't we using all the indexes? and why aren't they in order?

[@saideepshetty785]

I loved your explanation! Thank you for the video :D

[@tkorting]

Thanks for your feedback, please subscribe to my channel.
Regards

[@xiaowenyu6562]

Very very helpful! Thank you!

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@MN-sc9qs]

Great job!

[@tkorting]

thanks for your feedback
please like/share/subscribe
Regards

[@accelerator345]

thank you so much

[@tkorting]

Thanks for the feedback, please subscribe to my channel

[@byoungdookong2755]

amazing!! I understood about DTW through this video.

[@chaoxi8966]

clearly explained thanks

[@AndyLee-xq8wq]

so the "DTW distance" between two curves is the final sum-up?

[@rizalybeniga5935]

Hi. Thank you so much for this crystal clear explanation about DTW. I am just wondering on how to determine if this two series are same. I wonder if you can suggest some solution on how to find out if this two series are really same or not. Thank you so much for this video, I hope you can give me some suggestion to my problem. Thank you and May God bless you

[@tkorting]

Thanks for your comment. Please subscribe to my channel, share/like the video.
Two series are the same when the DTW distance is equal to zero.
For the case that you need to find equal series, you can also use a simpler distance, such as Euclidean for example.
Regards

[@ElChe-Ko]

What if there is a lower value of 15 in the same row when you start the identification of the warping path?

[@raziel846]

Thank u so much for this explanation!
I have a doubt in the last step, which term i must choose if two or more values are the same?

[@amirhosseinzeraati4641]

Hello sir, may I ask about the A_i - B_j part? Isn't it about the elements of matrix? not the indices! but in the video in first example it seems you subtract indices (j from i). Am I correct, or I missed something?

[@kadimdev]

you have mistake in the cost matrix representing
The cost matrix cells are filled one column at a time from the bottom to up, from left to right

[@slimanehamouda2583]

thank you infinitly

[@checkpeck]

excellant video...and when to use DTW ? 1. Finding similarity of two signals? 2. Comparing two signals? what happens if there are more than 2 signals?

[@tkorting]

Thanks for your comment.
Please like/share/subscribe.
Yes, DTW is used to compare signals. For example in sound recognition, apps like Shazam. Or to compute the similarity between two time series. It you have more than 2 signals, you can create an average signal, or compare one-to-one. These are only 2 examples.
Regards

[@bertal201199]

thank you good sir

[@schrodingerscat3912]

nice!!

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@yewenli]

thanks, it's very clear!

[@srazavitousi]

Could not explain better!

[@zhinadr534]

Thank you!

[@tkorting]

Thanks for your comment.
Please like/share/subscribe.
Regards

[@slomo4056]

perfect, thanks

[@tkorting]

Thanks for your feedback
Please subscribe to my channel
Regards

[@ahmetcengizhandirican8240]

Hi Thales
Thank you for this very clear explanation of DTW. But I wonder something. What does tell us the final chain of numbers we found? How are we going to interpert the result or understand if a macth found between the twp series?

[@tkorting]

thanks for your feedback and question.
Please like/share the video and subscribe to my channel.
The final result is a distance measure between two curves. So it is up to the analyst (user) to identify a proper threshold to characterize two curves as similar or not.
Regards

[@ahmetcengizhandirican8240]

Thanks

[@C.string-A]

I love this video so much. Thank you so much for your clear explanation. So what I understand is , DTW provides distance between two different signals by matching up similar local points (like a distance of peak-peak, and valley- valley of signals). and it is .. more of clasification methods like knn, rather than feature? Am I right?
BTW, I have a question, so the DTW measures distance between two time sequential signals. For example, Can it be applied to decide whether waveform A is similar to B or C? by getting distance vector from using DTW?
Once again, Thank you so much for the great video

[@tkorting]

Thanks for your feedback
Please like/share/subscribe.
Regarding your question, yes, DTW can measure the distance between a reference wave and other signals. I heard that (not sure) Shazam app (to discover music) uses this measure in their algorithms.
Regards

[@bnproduction9298]

@Thales Sehn Körting What if two time series are at different scale ? for example, one time series with y value ranging between 0-10 and another with y range 10000-20000.

[@asifmohammed033]

Beautiful !!

[@tkorting]

Thanks a lot.
Please like and share the video and subscribe to my channel.
Regards

[@omar.a.mahmud3297]

?Great explanation, thank you, what about multi- dimensions series

[@tkorting]

Thanks for your feedback.
Please like and share the video and subscribe to my channel.
In the case of more dimensions you can adapt the algorithm for the part that computes the difference between two points to feed the distance matrix, inserting the second, third distances as well.
Regards

[@omar.a.mahmud3297]

@@tkorting ,the metrics used to compete the distances will be Euclidean distance?

[@SeanAubin]

There's multiple approaches to this problem. See stats.stackexchange.com/q/163753/29949

[@omar.a.mahmud3297]

thank you @sean Aubin

[@ControlTheGuh]

So it's a way to compute the autocorrelation? Best alignment means best lag?

[@dr.mashaelalluhaybi5245]

Many thanks!!

[@tkorting]

Thanks for the comment, please subscribe to my channel

[@ZeMMacaes]

Muito obrigado, Thales
Só para ter certeza, o valor da distância entre as duas séries seria o valor na célula de cima da direita depois de preencher a matriz?

[@kaiohenrique]

Excelente vídeo!

[@tkorting]

Muito obrigado pelo comentário. Inscreva-se no canal.
Abraços

[@dr.mashaelalluhaybi5245]

how can I apply it if I want to measure the distances between more than 2 series (145 series)??!!

[@tkorting]

Thanks for your question. Firstly I would like to invite you to like and share the video and subscribe to my channel.
When you have more than 2 series, you should think on what is your application. For example, you can create a rank of the more similar series, and for this you can measure pair by pair all of your series and store the rank. Or if you want to measure what is the most similar curve to some reference that I have, you should compute also pair by pair and find the most similar curve. These are some suggestions, if you want to discuss more please reply.
Best regards

[@dofunback]

Nice video friend!! :) great job!!

[@tkorting]

Thanks for the comment, please subscribe to my channel

[@justwiredme]

Thank you

[@cw9249]

thank you for the explanation. how does the "window constraint" fit into this? what happens when u adjust it?

[@oldcowbb]

i don't get the sum part, isn't the upper right corner already the sum of the minimum cost alignment?

[@thetimesofcelebrities3157]

where were u? awesome.

[@danielcalencar]

Video top demais, vou aplicar na minha pesquisa isso dai. Abraço!

[@tkorting]

Abraço, inscreva-se no canal

[@ds5375]

Once the alignment vector is computed. How would one go about displacing the alignments to generate integrated data-points? E.g. if we assume A and B are the same signal, recorded with different instruments how would one perform a type of "batch correction" to shift time series A onto time series B using the aligned points.

[@tkorting]

Thanks for your question, please subscribe to my channel
I did not understand your question. What do you mean by correction?
DTW is used to compute if two series are similar (for example two instruments, same chords).
Regards