Correction: 4:16 KNN should have 10 correct and 14 incorrect. NOTE: There has been a debate if we should call the "testing dataset" a "testing dataset" or "validation dataset". In my opinion, this depends on the size of your dataset. We'd all like to have a large dataset that we can divide into three parts: Training, Validation and Testing, but that doesn't always happen in the real world. Support StatQuest by buying my book The StatQuest Illustrated Guide to Machine Learning or a Study Guide or Merch!!! statquest.org/statquest-store/
Can you clarify what is this "Correct" and "Incorrect" indicating after each testing using different blocks of data?..what is the interpretation when correct:4 ? :( Unable to get it.. :(
@@ahanapal4055 The machine learning methods that I am comparing in this video are classifying observations. Since we are training the methods, we know how the observations should be classified in advance. Thus, if the method makes the correct classification, then it is "correct". If the method makes the incorrect classification, then it is "incorrect". Does that make sense?
Dear sir/ Dear Josh, Your StatQuest series is brilliant to say the least. The internet is these days flooded with ML tutorials that teach how to run algorithms such as logistic regression or KNN using softwares, or with the lengthy incomprehensible mathematics that explains those algorithms. Yours is one of the rare materials that explains the philosophy! Philosophy, that is the deal for humans, not just feeding numbers and generating more numbers using a machine. Thanks a lot for giving me clarity on how exactly to use cross validation, and for clearing some of the nagging doubts from my tiny,less intelligent brain .
yes, that relates to me very much. I'm now in a Data Science bootcamp, and they just explain the maths behind each of algorithms incomprehensibly. they said that the point is just know the little math, because on the field, we just import the sklearn library, and try every model, every algorithm, which one gives the best prediction.... after listening to their statements like that, it makes me wondering. "hmm, i'm afraid that they are probably true, that there is no point at all to learn the math behind these ML Algorithms, because they just import module, choose each of existing algorithm, and done"
I think he is got the world's best teaching skills. Trust me learning ML is not easy unless you are interested. Even if you are not at least you will not feel sleepy in his lectures.
I can read books and listen to professors for hours about a subject like this and still not understand it... then I watch a 6 minute video and it is crystal clear. Thank you StatQuest!!!!!!!!
What's the difference between a machine learning method and machine learning model? Is a model applying a method to a specific dataset therefore modeling how it behaves? Will you make a statquest about what is a model??? That's a triple question mark bam! Love your videos! Thank you!
@@jcourn1 A Machine Learning Method is a way of teaching a machine using the data-driven approach. A Machine Learning Algorithm is a set of rules or a list of steps or a procedure to teach the machine using that methodology. A Machine Learning model is what we have received after applying the algorithm on a certain dataset to teach our machine. It represents what was learned by machine using the algorithm. Hope that helps 🙂🙂🙂
I hardly ever comment on RU-vid videos, but I just wanted to say that this has been a TREMENDOUS help and I absolutely loved the breakdown, logic, humor, and visuals. Thank you for for making this brilliant video!
Couldn’t agree more. After going through machine learning course materials on virtually every educational platform like coursera, simplilearn, EdX from top universities and companies from Harvard to Google, I think none of them remotely reaches the clarity and no-bushitness here. BAM!!!!
Who also liked the video upon hearing the short musical interlude at the beginning?! Your voice is very soothing. I’m preparing for an exam in about 2hours and needed to understand this concept. Thanks a lot! Every single info in this video came in my exams! I wrote with understanding!!! Thanks!
probably some of the best explained stats videos i've seen on youtube. thank you josh for constantly providing us with material that we can actually understand 0:)
If I pass my machine learning exam next week it will literally be all thanks to you. Either my book is completely unreadable or I'm stupid, but your videos make so much sense and I finally feel like I actually get the stuff you're talking about. Thank u!!
When I am gonna make my videos, you'll be my inspiration. The way you take us through the video is like a guide taking us through a guided meditation. Edit : and at the end it make us feel satisfied and delightful.
I have a ML quiz on monday and was so worried about not grasping these concepts in time - your videos are super clear and helpful and genuinely enjoyable to watch! Thank you StatQuest with Josh Starmer
there are a lot of teachers that have knowledge , but of them 80 percent dont know how to teach , 10 percent knows but dont care, 8 percent really care but are not succinct with their methods but 2 percent knows how to teach clearly and precisely in layman terms , they can teach anyone with their style , You are in that 2 percent category . Respect >>>>>.
I loved the Tiny Bam, Sir! You Patience to go slow tell us that you have a low bias; meaning, it's easy for non-native English folks to understand the concepts clearly. Keep up the good work. I will stalk your channel and like all the videos you have every made by the end of this week. Thank you Again.
Why can't most other lecturers on this world teach like you, why can't MY lecturers teach like you, im crying now :(((( if I have to learn Stats/AI/DL/... every single day for the rest of my life, but if it's you who taught us, it's well worth it.
Oh my sweet lord! I couldn't have ever imagined that someone can teach data science concepts soooooooo interestingly and easily. I never ever comment!! But you made me do this first time in my life
Clearly explained, great video! Maybe you skip this on purpose due to its complexity, but there is a small caveat. At the end you mention 'parameter tuning' using cv, these 'parameters' are called hyperparameters, different as model parameters. In order to do so, you need to further split the data into train/validation/test set, and only use train/validation part for tuning, while still having the test set for a final estimation of model performance.
please , I have question regarding cv for ridge regression , I will try different (lamnda) in each fold for example (10 different values for lamnda ) with ten folds or should I try each (lamnda) I need to test with all 10 fold and compare in the final between them
thanks bro. you help me a lot. now I understand what is testing, training, cross validation, bias and other lingos. i read many articles, but I don't understand a thing when they use this kind of words. thank you very much. from this, I also know what, why, how training and testing thing. thanks a lot. idk what to say anymore.
Thank you very much for this video, Josh! The use of visuals to explain cross validation really helps! I learnt a lot through this video about the fundamental basis behind cross validation as well as the extreme case of Leave-One-Out!
Hey Josh! This is the first time I'm watching your videos and I love the way you teach: pausing for a second before saying the next sentence. It gives time for the listener to digest what you said before! Love it!
These videos do such an amazing job summarizing concepts that my professor has spend hours trying to explain. I was pulling my hair in frustration at his teaching until I encountered your videos. These videos are like a breath of fresh air to my knowledge and understanding of data science. A huge thanks to you Josh Starmer! Keep up the amazing work!
The amount of BS they try to get you to wade through when explaining concepts E.g. Instead of starting with a massive equation and the formal explaination, a simple intuitive explainatiom, then relate that to the formal process
Your teaching style is just awesome. You explained everything in simple words and great English accent which is easily understandable. You got a new subscriber
I find this crazy that before and after every (very expensive) class now I'm looking up the same info here.... I'm a top-down learner though and my class seems to be built around bottom up learners. Thank you soooo much - yes I'll get a hoodie! #statquestforlyfe
Best concept descriptions I have found yet. Explained over-fitting in a better way that my textbook or course have. Hoping for a linear algebra course! Thanks!
I have completed Applied Machine Learning course from a University in US. The concepts I learned there are being reinforced after watching your Video Josh. Thank you so much for putting out these videos.
What you do is simply amazing!!!!! Thank you!!!! Just a tiny question: when you divide your data into blocks which you use as training set, do you use each different block for a different algorithm, or do you use the same training data to train different algorithms? Thank you again!
If you split your data into blocks 1, 2, and 3, then you would train all of your models on blocks 1 and 2 and test with 3. Then you would train all of your models on blocks 1 and 3 and test with 2 and then you would train all of your models on blocks 2 and 3 test with 1. bam.
I use a tenfold cross-validation method in the ridge and lasso regression implementation in my master thesis on SONAR/RADAR imaging. At that time I read a lot about Cross-validation to grasp the concept. Today your video help me to brush up the concept again. Thanks a lot. and feel bad that time I did not found this channel.
Firstly i like to thank you for explaining these concepts in such a crystal clear manner , this is one of the best video i ever witnessed. second, i request you to please make some video on backpropagation and some tedious concepts of M.L. once again thank you.
Your videos are very helpful, much practical and simple way to explain concepts. I learned more in your videos than my grad lecture notes. Thank you so much!
Dude I came to understand the difference between Cross Validation and Leave one out, instead I found that i completly missunderstood cross validation. Happy that i had a big breakthrough, i decided to watch the video to the end. And DOUBLE BAM in one sentence you explained what leave on out is. -> Subscribed!
1:18 ML methods, Logistic regression, K nearest neighbours, Support vector machines. Cross validation allows us to compare different ML methods and get a sense of how well they will work in practise. We need two things to do with the data collected. i) estimate the parameters for machine learning method(training the machine learning method) ii) test the machine learning method(evaluation of the model) 4 fold cross validation,leave one out cross validation, 10 fold cross validation(commonly used), tuning parameter
Just as I was getting seriously over my head with K Fold CV for a Numerai model... Lo and behold! My favorite statistical troubadour, Josh, appears to light the way. Bam to every which way you can validate it!
These videos are so helpful for me. One thing I'm running into though is understanding cross validation for time series data. When to apply a gap to the folds, when to use an expanding versus sliding window, etc. There isn't much quality info out there easily explaining the process. Might be a good future video idea!