Maths Intuition Behind Support Vector Machine Part 2 | Machine Learning Data Science

[@572varuag]

I'm glad someone like you decided to make a video on this. I have found that many find SVM hard to grasp because they dive directly into the code without understanding the intuition behind it. This goes a long way in helping people out.

[@mahikhan5716]

a sensible tutor i have seen my life who always traces students pulse . there are lot of tutorials about svm in youtube but no other covered A to z as like as krish . appreciate u krish

[@victor75570]

I cannot begin to thank you enough for breaking down and simplifying the math behind the machine learning algorithms. Understanding the math under the hood is pertinent to tuning the hyperparameters. I love your videos and I'm always recommending aspiring data scientists to check out your channel.

[@YouTubelesss]

man how could you remember all this... I keep forgetting the concepts after few weeks and had to watch it back to get a grasp on it. A million thanks for you in sharing your precious knowledge with us.

[@myeschool2129]

My Dear Teacher From my heart, I salute you, cause you to work so hard for us, your students, to teach the things with so much clarity. Praying for you. ....Noushad Rahim, Kerala

[@abhijitbhandari621]

You really are too passionate about teaching sir. Sometimes you even are being breathless, that excitement of teaching..got no words. Hope you will upload videos on topics related to DNN as well. Proud to be learning from you

[@akshaykhavare5898]

The way you simplify things is really commendable. After reading lot of blogs and going through other resources finally landed here and it was worth it. Thank you Sir.

[@Reem.alhamimi]

At 4:43, W transpose will be a row vector. W itself is a column vector and you need to take a transpose of W to make it a row vector - that it [-1 0]. x is fine as its a column vector anyways. Your calculation of -4 for y=w(transpose)x will be correct for the row vector of W - that is transpose of the column vector W

[@datahat642]

Very informative video and simple to understand. A slight oversight error, as the X here is 2-d [x1,x2], W (without b) must be 2-d as well [w1,w2]... If we consider bias b, then X = [x1 x2 1] and W is [w1 w2 b]... and in such a case we shall have a plane instead of a line

[@nitinudgirkar]

At 4:43, I think the W transpose will be a row vector. W itself is a column vector and you need to take a transpose of W to make it a row vector - that it [-1 0]. x is fine as its a column vector anyways. Your calculation of -4 for y=w(transpose)x will be correct for the row vector of W - that is transpose of the column vector W

[@chandrimapramanick1111]

Thank you for the clearing this. While doing the math I was also confused.

[@prashantkumarvishwakarma8645]

Thanks Sir, For all of yours video agr aap nhi hote to kbhi v itna sikh nahi pate humare instute wale log sirf overview bata k chor diye but real knowledge to aapki video se mila..... Thank You So Much

[@kirushikeshdb1885]

The w matrix should be [1 1] because the line equation is x1 + x2 =0, also while computing the value of y, wT should be a dimension of 1x2 and X should be 2x1, so that you will get a single value.

[@JackSparrowBoat]

Can you pls elaborate?

[@prateeksingh3808]

Yes I feel the same

[@rvg296]

The Regularization Parameter C is basically how much we want to avoid misclassification of points. If C is very large (infinite) we get the perfect classification of training samples smaller margin is considered, but if C is very small(0) it will cause the optimizer to find the maximum margin classifier even though it misclassified some points. Hence we have a find a good value of C in between. The Gamma parameter defines how much influence a training example has. For example, if gamma is high only the nearest points from the margin are considered for calculating distances, but if gamma is low even farther points from the margin are also considered.

[@krishnaik06]

U r right...thanks amazing work..

[@VishalPatel-cd1hq]

hi Rohit Ci actually will be lagrange multiplier as here our optimiztion problem is constrain one we will generally denote lagrange multiplier as lambda(i).

[@VishalPatel-cd1hq]

In SVM we are trying to solve our max(min) problem but by satisfying KKT (Karush-Kuhn-Tucker) condition we are making as min(max) problem and try to optimize this problem through SMO(sequential minimal optimiztion) .

[@VishalPatel-cd1hq]

my bad actually there will be one more term which will be multiple with ci(w.T+b) that we will callas lagrange multiplier

[@coolzkabhijit]

Can you please explain what is the role of gamma? What happens if farther or closer points are considered? How gamma affects the output?

[@SumanBhartismn]

Abhi tk kahan the sir, I was searching a teacher like you in ML. Finally mission completed. Love from my side.

[@usamaahmad7191]

Thank you so much Sir, the way you were teaching and i was getting all of your points, my love for your method and dedication starts hiking, a lot of love , respect and salute from Pakistan... knowledge have no boundries...

[@sivareddynagireddy56]

I saw so many articles about svms, every one say directly distance formula simply maximize,but your simplification from strach is awesome sir !!!!

[@andrewwilliam2209]

Hey Krish, I just want to say that your explanations are superb. I am new to Machine Learning and I took an online course about it but it barely gets into the mathematics. I understand that to get good and serious at ML we need a solid mathematical understanding of the various models, so i appreciate these videos that go in depth. To be honest I watched it the first time and didn't completely get it, but I'm going to watch it again now!

[@SAURABHKUMAR-ql8wi]

It was wonderful session. I have gone first time for SVM and able to relate the mathematics very well. Thanks a lot for this session.

[@nirangannirangan7175]

Excellent sir.Crystal Clear explanation.

[@jbhsmeta]

Just Great !!! Wow!!! - It was a great experience. Eagerly waiting for the part 3 of SVM covering the kernel trick.

[@lisa-sf7no]

Thanks

[@vignanvennampally6852]

Watching One video of yours = reading 100 blogs. Thank you for saving our time!

[@wenqichen4151]

I can't wait to express my infinite appreciation for you, sir! This video is so so so intuitive and uses less advanced math!

[@DhruvSharma14]

Salutes to you dear teach! You are one of the best for all concepts related to Data Science!

[@GujratiInGulf]

THANK YOU !!!! I can not even explain how much you helped me, I was about to cry as i was not able to understand the math behind SVM and why we use Lagrangian function. I have exam after 10 days and your videos are really helping me in this time thanks once again and HAPPY GURUPURNIMA from bottom of my heart !!!

[@himtyagi9740]

Lots of efforts are given by you in creating this clip...You are an excellent teacher...Best wishes

[@NicJd01]

Bhari dada!!! Mast samjavlat!!👌👌

[@chekrasena]

Krishna Anna Thoppu🙏

[@rupeshsingh4012]

Very very impressive explanation.. Thanks a lot. Bhagwan aapko hmesa khus or swasth rakhe....

[@zinaibrahim]

Krish, all I can say is thank you! the best and most comprehensive SVM lecture I've seen (and I've seen many).

[@darshitsolanki7352]

Kernel have sigmoid s shape graph n linear and polynomial form I m from statistics degree u have grt knowledge dude keep it up

[@TriedTastedJourney]

Best video to understand the math behind SVM. Thanks a lot Sir!

[@afsarullashareef3567]

Krish , I'm really thankful to you .. may God bless you .

[@bhabeshmali3640]

You are a Gem Krish Naik.

[@mayursalunke1654]

by looing this video, I actually understand, How important math in real life! and also How logistics and support vector machines actually differ. Thank you Krish sir

[@rajraji417]

Thank you so much Bro and I like you so much bro and Your individuality is seen through everyday and every videos I will become a data scientist one day ...

[@adipurnomo5683]

Clear explained. Very recommended

[@gianlucalepiscopia3123]

soon the best teacher out there

[@saisharadhashivakumar1004]

Hi sir without you i would have not understood deeplearning this much thank you so much

[@deleolukoya4634]

Wow! God bless you for all the efforts u put together to make this known to us, in fact, you're passionate and affectionate about us ur students. More strength and grace unto you. From Nigeria

[@mathrisk]

I am a beginner in the subject and your video gave pretty good idea about the topic. Thanks

[@hafimaoubarry6967]

Very Informative video. GOD bless you

[@tomthomas1431]

very good explanation....easy to understand

[@sciWithSaj]

Clearly explained Nice work sir

[@brown_bread]

at 2:55, b is not equal to c, which is consider here as 'slope'. b is bias term which is not added to the feature vector when dealing with SVM.

[@arjundev4908]

@Krish Naik.. My respect towards you and your work has increased by multiple folds.. You are godsend..my saviour :) Thanks for your contribution!! ❤

[@ranabhavesh1191]

Awesome video everything got clear.. 🙏🙏

[@faizalmakhrus8645]

Superb! This explanation is difficult to find in youtube.

[@tukaramugile573]

Very good explanation. Thanks

[@malathiavinash984]

This video is so good. Thank you Krish!!!

[@emmanuelibrahim6427]

Excellent delivery!

[@lokeshrathi5500]

The b value that you plot on the graph should be at the y-axis, right? Time: -@8:57 please have a check and let me know. Thanks

[@KrishnaMishra-fl6pu]

It's x intercept and hence that is correct

[@hessamjamalkhah9781]

Thank you dear Krish, you nailed SVM for me, I totally understood the concept behind it, and I really appreciate that. wish you all the best

[@geetanshkalra8980]

Thank you sir! Your video has helped me to get an internship to a very good company! ❤️🔥💥 Please continue the same work! It really helps us! Thank you!🔥🔥🔥

[@cryptogaming8026]

Please Check that whether w1 is the slope of the hyper plane because if we consider the equation w1x1+w2x2+b=0, then the slope comes out to be -w2/w1 , so tehnically in the example you explained you took y axis to be your hyperplane

[@jevoncharles8680]

You are a GENIUS!

[@waqarsarwar7012]

i appreciate your effort and your way of teaching

[@tramytran1992]

thank u so muchh for ur teaching. wish the best things to you

[@dipanwitamanna9540]

Beautifully and easily explained. Helpful

[@vijethrai2747]

18:49 C value is not how many errors to consider. It is quite the opposite. If C value increases, then tendency to make mistakes decreases because loss will increase with increase in errors. And vice versa. Greater C will create overfit and lesser C will create underfit.

[@debahutimishra3348]

This is an awesome video on mathematics behind Linear SVM... Too Good...Keep it up

[@rvg296]

Krish, I guess you missed the squaring of |W|. Basically, maximizing (2/|W|) or (1/|W|) is essentially the same. This means we have to minimize |W|. Just for math convenience, we will write it as (1/2) (|W|)^2 because differentiating this w.rt to W will lead us to obtain |W|.

[@muzamilhussain9908]

And distance X2-x1 was not the indicated distance ...that is just horizontal distance for that distance we have to use pathogoras

[@AmitSharma-rj2rp]

@@muzamilhussain9908 yes I was wondering the same thing - how do you get horizontal distance from subtraction?

[@expertreviews1112]

very nice and lot of effort put in to explain... complex topic but really nicely explained...

[@vishaljhaveri7565]

Thank you, Krish Sir.

[@devasheeshvaid9057]

At 3:00 how is 'WT'=[-1, 0] . W extends from origin and is perpendicular to 'X'(i.e. the hyperplane) So, shouldn't 'WT' be [-1,-1] (or any other point perpendicular to 'X' i.e. the hyperplane) ?

[@ChiTamNguyen-d9d]

I agree, this should be [-1,-1]

[@hemangdhanani9434]

understand completely... thanks for simplifying

[@ProgrammingCradle]

Beautifully explained... Thank you Krish :)

[@sandipansarkar9211]

Great explanation Krish.Thanks

[@mangaenfrancais934]

Your are the best

[@optimusVideo]

You tought good i understood Quickly.

[@ManishKumar-qs1fm]

In one words awesome

[@sridharmakkapati6586]

Thanks for knowledge sharing

[@vigneshvicky6720]

Nice nice very nice❤

[@vijayalakshmi3968]

thank u much sir.... ur videos are more helpful for my course......well explained!

[@rengarajanraman8608]

Thanks for putting lot of efforts in explaining such complex concepts.

[@davidlee4293]

super good... you explain it so very well. thank you

[@datakube3053]

respect to u r work n efforts

[@carearayam]

Excellent explanation, my friend

[@eeshdeepsingh7030]

Thanks alot for this video sir❤️ waiting for svm part 3 sir!!!

[@SAN-te3rp]

I ve been searching who teach maths like this finally found 🙏🙏

[@codeforcoders69]

Me too

[@LanteLuthuli]

Legendary ... New subscriber!

[@venirajan2772]

Thank you so much for awesome explanation.keep on doing sir.

[@williamblanzeisky2524]

dude u r so fun to watch and u make SVM so much easier lol

[@RO_BOMan]

No words for this great work. Thanks you very much For making these concepts very very easy to understand. I would suggest you to arrange all of them in a particular order or give some sequential number for these videos, so that I will be easy to go through all the topic without any deviation. Thanks again, keep uploading more videos on different topics.

[@mohamedgaal5340]

Thanks bro. You really did your best to simplify things. I truly appreciate it.

[@akshaybagal2208]

great stuff and nice explanation!!!

[@shivadumnawar7741]

Great tutorial. Thank you so much sir.

[@ratulghosh3849]

Thanks for the explanation of intuition behind SVM.

[@midhunskani]

I was for this video. Thanks again. Liked and subbed

[@fancy4926]

at around 5:30, take point (4,4) for example, if the b=1, the product should be [-1, 1]T [4, 4] = A 2*2 matrix. then we cant say it is a positive or negative value? is there any error during writing the equation?

[@SaulWilliamss]

These videos are incredibly helpful, thank your very much for sharing your knowledge with us!

[@hemantdas9546]

Sir we need SVM regression

[@AbhishekKumar-gu3ny]

I have a doubt, at 10.43, Isn't the maximum distance between margins be X1 + X2? Thank you for this simplified lecture on SVM.

[@K-mk6pc]

Sir can you disclose how to learn these concepts, what reference materials You usually use so that we can start to learn in your way.

[@mwaurades]

I really appreciate the tutoring Sir , keep up the good work !!!

[@Neuraldata]

You are really an inspiration to many 👌

[@JaydeepSinghTindori]

Thanks for your good explanation on SVM. But I have a confusion that when you changed the representation at 20:29 from max(2/||w||) from algebraic representation to to calculus representation min(||w||/2), I think it should be min(||w||^2)/2 as the derivative will give you min(||w||) as we want to maximize the (1/||w||) or minimize (||w||).

[@krishnaik06]

Yes u r right...missed that one

[@geogeo14000]

@@krishnaik06 Hello and Thank you. How do we go from ||w|| to ||w||^2 ?

[@tabindabhat9078]

Wow. Great video. Thanks.

[@turkeshpote3239]

fantastic explanation

[@sarabjeetsingh5033]

Hi Krish @3:32 the matrix multiplication is incorrect. Consider a matrix A of order m X n ie it has m rows and n columns Consider a matrix B of order n X p ie it has n rows and p columns When you do matrix multiplication of A and B, the resultant matrix should of order m X p While in your case you multiplied a matrix of order 2 x 1 with 1 x 2, so the resultant matrix should be of order 2 x 2 but you have 1 x 1 order. Also, what you have written looks more like a determinant of a matrix ( integer value) than a matrix. Could you check on this?