You have the skill to simplify a complex topic which can be understood by everyone. Please continue your great work. This world needs more teachers like you.
Great video for people like me who are beginners and don't want to go deep in the Statistics part of it but a simple explanation for data science. 🧡 from India.
I loved this video. I've heard about "reducing the coefficient values" in so many other places, but you explained the 'why' behind this better than any of the others that I saw.
You made it easy to understand. But where do you get the alpha and slope? From the testing data set? Then the testing data set becomes the training data set.
Really a pristine work, in explaining the ideas behind the concept. I found it really useful for having an overview look before dealing with all the math behind. Thanks
Really appreciate the tutorial, just one query, Does regularisation always reduce the slope? I mean i think it's possible for the test dataset to have more slope than training set.
Regularisation minimises the sum of squared errors while also minimising the sum of squared magnitudes of the coefficients. This pushes the ridge coefficients closer to zero. But yes, if the penalty term is too small, the slope may resemble that of OLS.
Thank you for your short video. But I did not understand why we should minimize the slope. It is just a possibility and depends on test data. You may increase the slope to get minimum residuals.
It just feels like a fancy way to include your testing set into your training set, essentially making 100% of your data a trainingset. What is the difference between those?
The explaination is good, but I think that your example could be better. Having 3 points in the training set and 5 points in testing set is not a good practise. Also your 3 training points will give the same line every time, so again: not the best example
The example is perfect, it is for illustration, and textbooks use the same amount for training data points, it’s better to emphasize the idea of more testing data points to show the mainstream and pattern of the data, in reality, the dataset you use will never be as much as the samples it was testing or seen on. The 3 similar training data points are the same reason why the problem occurs, and the ideal mechanism for solving it is to deviate your model from it.
You have the skill to simplify a complex topic which can be understood by everyone. Please continue your great work. This world needs more teachers like you.
I've watched dozens of videos on regularization and your explanation is perfect! thanks!
Wow! It took me several rewinds to understand that from my professor and I got it in 3 mins with the way you explained and visualized it! Thank you!
the best explanation for the ridge regression I have ever listen
Great video for people like me who are beginners and don't want to go deep in the Statistics part of it but a simple explanation for data science. 🧡 from India.
Thank you for explaining bias and variance and not just moving forward without the explanation!!
I loved this video. I've heard about "reducing the coefficient values" in so many other places, but you explained the 'why' behind this better than any of the others that I saw.
You made it easy to understand. But where do you get the alpha and slope? From the testing data set? Then the testing data set becomes the training data set.
You explained it in simple way and with a short video. very effective
Really a pristine work, in explaining the ideas behind the concept. I found it really useful for having an overview look before dealing with all the math behind. Thanks
Wow , your explaination are too good, it's my first time seeing your video and i'm really satisfied
The best explanation I've heard on ridge regression. Straightforward and precise! Thank you very much!
Very well explained, finally i got it! Many thanks.
Perfect explanation!!
You explained it in simple way and with a short video.
Thanks, keep the good work
Very good explanation. Thank you. It gives me the idea of ridge regression.
Very helpful! Thank you Professor!
Great explanation! Thank you!
Thank you for the quick and easy to understand tutorial
Glad it was helpful!
Thank you! This is a very helpful explanation and visualization of ridge regression.
You're very welcome!
This is a great video, thank you!
Really appreciate the tutorial, just one query, Does regularisation always reduce the slope? I mean i think it's possible for the test dataset to have more slope than training set.
Black hole here... Looking for this answer...
Regularisation minimises the sum of squared errors while also minimising the sum of squared magnitudes of the coefficients. This pushes the ridge coefficients closer to zero. But yes, if the penalty term is too small, the slope may resemble that of OLS.
So it is highly unlikely for regularisation to increase the slope than that of OLS.
The only and first video that allowed me to understand this shit. Thanks!!
Good Intuition. Contradicting in the slides whether ridge regression increase/decrease for bias and variance.
Thank you sir, it's so simple!
Thank you Sir! the great explanation made the concept seem so easy!
really appreciate your effort thanks for help!
Wonderful explanation. Thank you.
I like how you explained that well in a 7 min video.
Awesome Explanation. thanks!
Glad you enjoyed it! thanks!
great crystal clear
Amazing explanation, thanks ryan
My pleasure!
Good Explanation....
Thank you for your short video. But I did not understand why we should minimize the slope. It is just a possibility and depends on test data. You may increase the slope to get minimum residuals.
Minimizing or maximizing is decided after looking at the total errors. If maximizing increases the error then we will go to minimizing the slope.
amazing explanation!
Suggestion: You explained very well Ridge & Lasso Regression, make also one for Elastic Net!
what if model needs high sensitivity to dependent variable ?
In ridge regression alpha never be 0 . ☺️ Easy and clear explanation
Thank you for this *great explanation*
excellent concept explanation.. thank you
Hi Ryan, Can you please do a video on Elastic Net Regression?
Great video and great english as well, you gained a new sub
great intro !
I'm glad you like it
How does increasing Lambda trem reduces the slope. We are multiplying Lambda with Slope right, which is constant?
Sir how do we know that during regularization we have to increase or decrease the slope.
thank you for the video. do you speak Farsi ?
It just feels like a fancy way to include your testing set into your training set, essentially making 100% of your data a trainingset. What is the difference between those?
Thank you sir🙏🙏
👏👏👏👏👏👏 well explained!
Thank you!
5:01 door opens
But how ridge works if the variance decrease with a steeper slope?
Education is about pedagogy. Who teaches. Here's a good one.
Isn’t alpha actually lambda?
The explaination is good, but I think that your example could be better. Having 3 points in the training set and 5 points in testing set is not a good practise. Also your 3 training points will give the same line every time, so again: not the best example
your such a hater😢
The example is perfect, it is for illustration, and textbooks use the same amount for training data points, it’s better to emphasize the idea of more testing data points to show the mainstream and pattern of the data, in reality, the dataset you use will never be as much as the samples it was testing or seen on.
The 3 similar training data points are the same reason why the problem occurs, and the ideal mechanism for solving it is to deviate your model from it.