I have a great story about "a model that predicts worse than random". Many years ago when I first became involved in developing ML tools at work, one of my seniors noticed that the tool my team was working on was predicting things worse than random, which we found very surprising, because we had tested it before and it worked well. In fact, he scolded us for making such a terrible model, saying, "You would be better off if you guess the opposite of what your model was guessing!" Sure enough, we discovered that some refactoring had led to an extra negative sign being introduced that had flipped the direction of our model. Removing the negative got us back to a very accurate prediction!
Wish you were my teacher when I was in school/college :/ Nevertheless, I'm grateful that I am getting to learn from u via youtube... Big thank you... Stay blessed
I am disappointed in all the professors at my university who proudly overcomplicate things. After watching your videos, all the math and advanced topics make sense and become very obvious. This channel and your videos deserve to be called "must-watch," just like Jeremy Howard's and Andrew Ng's courses. I wish your full courses were available on a platform more affordable than Udacity😅
I am confused about what is referred to as "model". When we say that "the greater AUC the better the model" it seems to me like it's a bit of a misnomer since AUC characterises the whole set of the models (which we can obtain by translating the line from one extreme to another) but not a distinct one. A point on the curve, from the other hand, seems like a legitimate representation of "a model". It seems like these two terms are used interchangebly in the video and this is a root of my confusion. Other then that, it's a great visual!
Gracias Jorge! Yes, this one! www.udacity.com/course/deep-learning-pytorch--ud188 And more to come soon, I'll announce them in the channel! (also I'm sorry for the very late reply) :)
@@SerranoAcademy I didn't know your deep learning course was free! 😮 Every course I found in Udacity through search was around $1,000. While I can't afford that price point, I still want to support your work. Besides buying your book and subscribing, are there any other ways?
Very informative video, just one question can this be used as metric for multilabel classification (considering using machine learning algorithms not neural-nets)? I mean what can we get from roc-auc curve about multilabels and if not what will good metric to look for?
Hola Jairo! Este todavia no, pero pronto lo pongo. Por ahora en este canal hay algunos en espanol, pero mi plan es ponerlos en un canal separado. Por aca hare los anuncios!
Luis...its my request to make series of videos of quantum computing bay sian learning and reinforcement learning..I am ready to pay for it..it is much needed to society ...the godly person like you can make the life of people better Trust me you are here for giving like god ...please think of my request ..
I have a great story about "a model that predicts worse than random". Many years ago when I first became involved in developing ML tools at work, one of my seniors noticed that the tool my team was working on was predicting things worse than random, which we found very surprising, because we had tested it before and it worked well. In fact, he scolded us for making such a terrible model, saying, "You would be better off if you guess the opposite of what your model was guessing!" Sure enough, we discovered that some refactoring had led to an extra negative sign being introduced that had flipped the direction of our model. Removing the negative got us back to a very accurate prediction!
Just want to leave a comment so that more people could learn from your amazing videos! Many thanks for the wonderful and fun creation!!!
probably the best explanation- simple and effective illustrations, thank you very much!
You made it so easy, could never understand this concept well before
Absolutely brilliant! You're the one guy I dont mind spending a lot of money on your courses for
The final prediction rotation on RoC curve is really inspiring
Amazed with your teaching style! Kudos Dr. Serrano
by far the best video on ROC curve.
Wish you were my teacher when I was in school/college :/ Nevertheless, I'm grateful that I am getting to learn from u via youtube... Big thank you... Stay blessed
Very clear explanation!
Very clear. I come to realize the different curve is due to different threshold.And some threshold doesn't hurt FPR in some cases as you mentioned.
Your explanations are the best!!🔥🔥🔥
Awesome! This video is simply a gift; thanks for sharing this.
Great and Amazing explanation like usual Serrano
Great lecture. Thanks.
Thanks, you helped me clear my mind up.
Thank you for your wonderful videos. It was great!
best intution given on this topic ever !!
Amazing video ❤
Best on the globe 💯💯💯🙏🏻🙏🏻🙏🏻
Insightful, thanks a lot!
thank you so much one of the best videos
Nicely explained. Would have been great if I had your video 30 years ago😅
fantastic, thank you very much!
thanks for the appropriate illustrations!
I am disappointed in all the professors at my university who proudly overcomplicate things. After watching your videos, all the math and advanced topics make sense and become very obvious. This channel and your videos deserve to be called "must-watch," just like Jeremy Howard's and Andrew Ng's courses. I wish your full courses were available on a platform more affordable than Udacity😅
I am confused about what is referred to as "model". When we say that "the greater AUC the better the model" it seems to me like it's a bit of a misnomer since AUC characterises the whole set of the models (which we can obtain by translating the line from one extreme to another) but not a distinct one. A point on the curve, from the other hand, seems like a legitimate representation of "a model". It seems like these two terms are used interchangebly in the video and this is a root of my confusion. Other then that, it's a great visual!
Hey, I wonder how could we decide in a similar situation when the number of features is not just x & y but say for example 3 features or even more?
Thanks Luis. This is the simplest and best ever tutorial in ML I have come across. do you have more courses? udemy? or other platform?
Gracias Jorge! Yes, this one! www.udacity.com/course/deep-learning-pytorch--ud188
And more to come soon, I'll announce them in the channel!
(also I'm sorry for the very late reply) :)
@@SerranoAcademy I didn't know your deep learning course was free! 😮 Every course I found in Udacity through search was around $1,000. While I can't afford that price point, I still want to support your work. Besides buying your book and subscribing, are there any other ways?
This is brilliant
Did you mean "so that it won't have any false negatives at all" @2:19? By the way , great video and explanation style!
Exactly. It confused me too, made me watch the whole thing again, and question everything I know about hypothesis testing. haha!
You're amazing!
Great video thanks
How can you make it so simple to learn?! when I need to revisit some concept I always search for: "Something that I need to learn Luis Serrano" :)
Thank you Rodrigo, that's such a nice thing to hear! :)
I come to say exactly the same thing! We need a video from Luis to tell us how he understands everything in simple terms like that!
Thank you :)
Hi, the manning website does not work. I am unable to buy the book :(
Hi Ashay, sorry about that. Did you try this one?
www.manning.com/books/grokking-machine-learning
Let me know if that still doesn’t work. Thank you!
Very informative video, just one question can this be used as metric for multilabel classification (considering using machine learning algorithms not neural-nets)? I mean what can we get from roc-auc curve about multilabels and if not what will good metric to look for?
wow, thank you
Thank you so much
Perfect!
¿hay este vídeo en español?
Hola Jairo! Este todavia no, pero pronto lo pongo. Por ahora en este canal hay algunos en espanol, pero mi plan es ponerlos en un canal separado. Por aca hare los anuncios!
Mmmm the temptation to draw a curvy overfit line through the data
awesome
You are god
Luis...its my request to make series of videos of quantum computing bay sian learning and reinforcement learning..I am ready to pay for it..it is much needed to society ...the godly person like you can make the life of people better
Trust me you are here for giving like god ...please think of my request ..
@@maj46978 Yesssssss 💯
Great explanation. Brilliant.
Great video! Thank you so much