Thanks to you is not enough. You have done a lot to share knowledge with people. Great people like you are the people who make difference. Please continue to be generous and kind.
Sir many many many thanks to you. I really have no words to thank you for making this video. I have looking for these three methods since a very long time on internet but I have not found it anywhere and you made it. I'm extremely happy and extremely thankful to you. You are the best Professor in the world. Respect.
Thank you for clear and simple explanations. After you use Elastic to predict, how do you interpret your results? Please provide more explanation ex. prediction on train is 4.11 and on test is 6.15. What does this actually tell us?
Those are root mean square error (RMSE) values that tell us how accurate our model is. Lower values of RMSE indicates better prediction performance by the model.
Thank you for this - question: how would you code to group your factor variables so that elastic net throws out or keeps all levels of the factor variable?
Hello. I want to implement the ridge regression method on a small dataset. but I want to get it by solving the model manually (by hand). How can I do it? I will be glad if you can help.
Hi, thanks for nice video tutoial. one of the best. I wanted to go over your tutorial. Unfortunately, I also being stopped at Ridge. I got the same problem over there: > ridge
Thank you for this nice explanatory video. However, I do have one question. In the end, you end up with some variables having the highest coefficients, i.e. the ones explaining the outcome data the best. But how do you determine a cut-off ? In your case "nox" and "rm" have higher coefficients than most making them the easy choice. But what if you have some that are closer together etc. Do you not use variables that have coefficients lower than 1 (higher than -1), or how do you actually pick the right variables, and the correct amount none the less, instead of ending up with 10 variables for your model.
can we have an example of multiple factor level categories in the model and see how elastic net works. will it keep insignificant factor levels or drop them off as they are insignificant
thank you for your video, very help, i've got a question, how about SEE for penalized spline regression? what is the different with ridge regression? thanks in advance
Thanks for a fantastic video. I have run this tutorial without issue with Boston data set. I am replicating this with my own current data set of 10 variables with 226 observations. When running lasso and elastic net I keep getting this error "In nominalTrainWorkflow(x = x, y = y, wts = weights, info = trainInfo, : There were missing values in resampled performance measures." However I have no missing data within the data set. Any thoughts? Thank you.
Thank you for the video, Dr. Rai. Does anyone know how to make the font size on the label larger in the plot? It is extremely small. I searched online and tried for a while, but cannot get it done. Also, how about adding legend to the plot?
@@bkrai Thank you for your reply, Dr. Rai. I tried to use cex, but none of the methods change the size of the label on the lines. (I only know of cex.axis, cex.lab, cex.main, and cex.sub, and none of them work in this case.) - Anthony
@@bkrai Also, in the plots (log lambda vs. Coeff and Fraction Deviance Explained vs. Coeff), they showed that variable 6 is better than variable 5 as Variable 6 is not overfitted. What are Var(5) and Var(6)? Are Var(5), nox, and Var(6), rm? But based on the importance order, nox is much more important than rm, so did I get the number /variables order messed up?
Thanks for the video Prof. Rai. I have a couple of questions: What is finalModel, You have set lambda between 0.001 and 1. However I see large log lambda values in the plot. How to explain this? Thanks
Thank you so much for explaining in such simple and efficient way. Does Ridge/Lasso/Elastic Net fall into the dummy variable tarp ? I have a data set where I have created dummy variables for categorical variables, and then aggregated them. Can I use all of them in these methods?
Thank you very much Dr. Rai. I posted my question in the link below. I am leaving a record here in case anyone has the same question. stats.stackexchange.com/questions/502356/lasso-regression-with-aic-or-bic-as-model-selection-criterion
This is an excellent explanation sir, hats off to you. sir I have a query, could you please tell me how can I use the ridge, lasso and elastic net regression for unbalanced panel data... looking for your help.. kindly tell me the source also for the panel data ridge, lasso and elastic net regression...
Sir my data is in the form of Yit= a +bXit+ eit where i is the number of firm and t is the time period. i have collected the data for n firms over period 0f Time t. i mean i have collected the data for each firm for t period. now my problem is i have several explanatory variables therefore i want to use normalisation for variable selection and inference for imbalanced panel data kindly help me. @@bkrai
Thank you for a great tutorial, Dr. Bharatendra Rai! I wonder how you handled the categorical variable. Did you put into the LASSO model as numeric or factor?
Hi Sir, really a very helpful interpretations of outcomes for lasso ridge and elasticnet. I have a doubt, can we apply ridge, lasso and elasticnet on classification problem??
Hi sir, Thanks a lot for replying... I am practicing on churn analysis data with logistic regression. I request you to please guide us with a video demonstrating the customer life time value analysis on R and how and where it is used
Hello, I am trying to run your algorithm but when I tried to run the linear model (7:53) it fails, showing the following message: predictions failed for Fold01.Rep1: intercept=TRUE Error : $ operator is invalid for atomic vectors. How can I solve this? Thank you! PS: Nevermind...I could solve this by myself: it is just reset R. I worked now. Anyway, thanks!
Hello sir For my dataset i am getting same lower RMSE and same higher Rsquared value for lasso and elastic net methods...how can i choose better algo from these two?
Thank you for your excellent and informative lecture. I want to ask one question that I am facing in recently. I am trying quantile regression with LASSO penalty. Is it possible to use the “train” function by changing method and tuneGrid ? If not then what will be the solution ? would you please make video on quantile regression with LASSO penalty.
Thanks thousands for useful video . I have a question about "Error in plot.new() : figure margins too large " after running pairs.panels(Data[c(-2, -55)]) . I have 55 factors and 57686 observation . it can handle this up to limited number of factors?
Sir, If independent variables are highly correlated in classification problems than also we should use these types of procedures or we can ignore that. Kindly help
Sir I want to ask about the sequence, when in ridge regression you set the sequence as 0.0001 to 1 with length 5, for this sequence I got the lambda= 0.00001, when I changed the sequence to 0.001 to 1, my lambda also changed. Can you please tell me what lambda should I select?
Sir.. you did not performed any standardization on data before running the alogrithm. Is standardization required or not before running this techniques.
Dear Bharatendra, thank you a lot for the bright explanation! I would like to ask you how can I get p-values for coefficients from elastic net? Thank you in advance!
it's not simple, and is a somewhat new area of research. usually you don't worry about it, after all the only reason you see the variables is because the elastic net procedure decided they should be kept, remember?
Thanks a lot for the tutorial Sir. Really helpful.Just a small question though. Is there a code that i can use to see the P value of the regression equation also also along with the R squared and RMSE values?
sorry sir am asking lot of questions, in my case am having lot of independent variables that are factors and would like to crunch them in to few to make it simple. in my case target variable is "yes" or "no".glmnet function in R says it also supports binomial.
20:40 you say the purple variable (#6) is more important than blue (#5), but this always disagrees with the variable importance plots, which say nox (variable 5) is most important with rm (variable 6) the second most important. Did you mean to say the blue line was more important? You also say this at16:58
Okay thanks. Great video by the way there's a lot of great stuff in here in addition to what's in the title so thanks for making it, you're also the only person I've seen who speeds up the code writing portions instead of putting us to sleep to the sound of the keyboard slowly typing code so that was a smart touch too :)
Thanks for the comments! You can find some useful links below: Introductory R Videos: goo.gl/NZ55SJ Machine Learning videos: goo.gl/WHHqWP Deep Learning with TensorFlow: goo.gl/5VtSuC Image Analysis & Classification: goo.gl/Md3fMi Text mining: goo.gl/7FJGmd Data Visualization: goo.gl/Q7Q2A8
Sir I have got this problem and i have checked my each value i do not have any missing value plz help Error in na.fail.default(list(LTDA = c(0.26683158, 0.37309602, 0.40553323, : missing values in object
If you are already using the R code that I provided in the description area, then I'll suggest review the code as it may be a small error. Also check if you have missing data.
@@bkrai Sir the code you provided has some missing data. I tried to fill by looking at your video but it doesn't work. Can you share the exact code in this video? Thank you
Thanks to you is not enough. You have done a lot to share knowledge with people. Great people like you are the people who make difference. Please continue to be generous and kind.
Thanks for your comments!
Sir many many many thanks to you. I really have no words to thank you for making this video. I have looking for these three methods since a very long time on internet but I have not found it anywhere and you made it. I'm extremely happy and extremely thankful to you. You are the best Professor in the world. Respect.
Thanks for your comments!
i guess it is kinda off topic but do anyone know of a good website to stream newly released tv shows online?
@Khalid Alonso lately I have been using flixzone. Just search on google for it =)
@Winston Edison definitely, have been watching on FlixZone for months myself :)
@Winston Edison Thanks, signed up and it seems like they got a lot of movies there :) I appreciate it!
You are explained it clearly on Ride, Lasso and Elastic Net Regressions. and your teaching styles are awesome. Thanks Sir.
Thanks for comments!
This video is so helpful, I now have a glimpse of how to go about comparing Ridge, Lasso, and Elastic Net. Thank you so much Sir.
Thanks for your comments!
Dear Sir, in your videos u make our life so simple and save no of hr's.
Good to know it helps save time!
You just solved my of biggest problems in my article, THANK YOU!
You are welcome!
Please help at 23:43, what can I do so that xyplot() compares different models eg LASSO vs Ridge. Thanks!
Thanks for the video. For Lasso and Elastic Net, shouldn't we scale the data before running the model?
That's always a good idea.
Dr. Rai
After I run the following code
ridge
You can try that and see if it helps to improve the model.
Isn't value of lambda supposed to be between 0 and 1 if you're using the values in sequence?
Thank you for clear and simple explanations. After you use Elastic to predict, how do you interpret your results? Please provide more explanation ex. prediction on train is 4.11 and on test is 6.15. What does this actually tell us?
Those are root mean square error (RMSE) values that tell us how accurate our model is. Lower values of RMSE indicates better prediction performance by the model.
Thank you for this - question: how would you code to group your factor variables so that elastic net throws out or keeps all levels of the factor variable?
Hello. I want to implement the ridge regression method on a small dataset. but I want to get it by solving the model manually (by hand). How can I do it? I will be glad if you can help.
Hi, thanks for nice video tutoial. one of the best. I wanted to go over your tutorial. Unfortunately, I also being stopped at Ridge. I got the same problem over there: > ridge
getting the same error... is there a solutution?
I don't see any error. Probably you can delete code, and rewrite as sometimes it is something small that we may miss.
Thank you for this nice explanatory video. However, I do have one question. In the end, you end up with some variables having the highest coefficients, i.e. the ones explaining the outcome data the best. But how do you determine a cut-off ? In your case "nox" and "rm" have higher coefficients than most making them the easy choice. But what if you have some that are closer together etc. Do you not use variables that have coefficients lower than 1 (higher than -1), or how do you actually pick the right variables, and the correct amount none the less, instead of ending up with 10 variables for your model.
please reply on this sir
We don't need to pick variable as feature selection happens automatically here.
I saw your comment today.
You're the real MVP! thanks alot for sharing this
You are welcome!
can we have an example of multiple factor level categories in the model and see how elastic net works. will it keep insignificant factor levels or drop them off as they are insignificant
thank you for your video, very help, i've got a question, how about SEE for penalized spline regression? what is the different with ridge regression?
thanks in advance
Thanks for a fantastic video. I have run this tutorial without issue with Boston data set. I am replicating this with my own current data set of 10 variables with 226 observations. When running lasso and elastic net I keep getting this error "In nominalTrainWorkflow(x = x, y = y, wts = weights, info = trainInfo, :
There were missing values in resampled performance measures." However I have no missing data within the data set. Any thoughts? Thank you.
I saw this today, probably you already solved this problem.
Thank you for the video, Dr. Rai.
Does anyone know how to make the font size on the label larger in the plot? It is extremely small. I searched online and tried for a while, but cannot get it done. Also, how about adding legend to the plot?
To change font size you can use 'cex'.
@@bkrai
Thank you for your reply, Dr. Rai. I tried to use cex, but none of the methods change the size of the label on the lines. (I only know of cex.axis, cex.lab, cex.main, and cex.sub, and none of them work in this case.)
- Anthony
@@bkrai
Also, in the plots (log lambda vs. Coeff and Fraction Deviance Explained vs. Coeff), they showed that variable 6 is better than variable 5 as Variable 6 is not overfitted. What are Var(5) and Var(6)? Are Var(5), nox, and Var(6), rm? But based on the importance order, nox is much more important than rm, so did I get the number /variables order messed up?
You are a legend, this was super helpful.
Thanks for comments!
Thanks for the video Prof. Rai. I have a couple of questions: What is finalModel, You have set lambda between 0.001 and 1. However I see large log lambda values in the plot. How to explain this? Thanks
u would definitely work so hard for all this, hats off to you.
Thanks for comments!
Great music... :) What a great video. I just learn so much from you. Thank you.
Thanks for comments!
Thank you thank you so much . Can I ask is it the same testing for for logistic model ?
In logistic model response is a factor variable. So these two situations are different.
Thank you so much for explaining in such simple and efficient way. Does Ridge/Lasso/Elastic Net fall into the dummy variable tarp ? I have a data set where I have created dummy variables for categorical variables, and then aggregated them. Can I use all of them in these methods?
Yes, that should work fine.
Thank you Dr. Rai. This was helpful. Do you know how to alter the selection criterion (e.g., to aic, bic, sbc, etc.) by any chance?
See if this helps: www.methodology.psu.edu/resources/AIC-vs-BIC/
Thank you very much Dr. Rai. I posted my question in the link below. I am leaving a record here in case anyone has the same question.
stats.stackexchange.com/questions/502356/lasso-regression-with-aic-or-bic-as-model-selection-criterion
You are welcome!
This is an excellent explanation sir, hats off to you. sir I have a query, could you please tell me how can I use the ridge, lasso and elastic net regression for unbalanced panel data... looking for your help.. kindly tell me the source also for the panel data ridge, lasso and elastic net regression...
For class imbalance problem, you may find following useful:
ruclips.net/video/Ho2Klvzjegg/видео.html
Sir my data is in the form of Yit= a +bXit+ eit
where i is the number of firm and t is the time period.
i have collected the data for n firms over period 0f Time t. i mean i have collected the data for
each firm for t period. now my problem is i have several explanatory variables therefore i want to
use normalisation for variable selection and inference for imbalanced panel data kindly help me.
@@bkrai
Thank you for a great tutorial, Dr. Bharatendra Rai! I wonder how you handled the categorical variable. Did you put into the LASSO model as numeric or factor?
Note that chas is a factor variable in this example.
excellent work! cud you plz make a video to check for assumptions of linear regressions
thanks for the suggestion, I've added it to my list.
Hi Sir, really a very helpful interpretations of outcomes for lasso ridge and elasticnet. I have a doubt, can we apply ridge, lasso and elasticnet on classification problem??
Yes, they can be used for classification problems too.
Hi sir, Thanks a lot for replying... I am practicing on churn analysis data with logistic regression. I request you to please guide us with a video demonstrating the customer life time value analysis on R and how and where it is used
I've added this to list of future videos.
Thank u sir..
Hello, I am trying to run your algorithm but when I tried to run the linear model (7:53) it fails, showing the following message: predictions failed for Fold01.Rep1: intercept=TRUE Error : $ operator is invalid for atomic vectors. How can I solve this? Thank you! PS: Nevermind...I could solve this by myself: it is just reset R. I worked now. Anyway, thanks!
Thanks for the update!
Impresseive explanations Sir, I liked it too much....
Thanks for comments!
Hello sir
For my dataset i am getting same lower RMSE and same higher Rsquared value for lasso and elastic net methods...how can i choose better algo from these two?
You can choose whichever is simpler.
Thank you for your excellent and informative lecture. I want to ask one question that I am facing in recently. I am trying quantile regression with LASSO penalty. Is it possible to use the “train” function by changing method and tuneGrid ? If not then what will be the solution ? would you please make video on quantile regression with LASSO penalty.
Thanks, I've added this to my list.
Thanks for the great tutorial! Would anyone know why coef(ridge) returns many coefficients for each predictor? Thanks
Thanks thousands for useful video . I have a question about "Error in plot.new() : figure margins too large " after running pairs.panels(Data[c(-2, -55)]) . I have 55 factors and 57686 observation . it can handle this up to limited number of factors?
You can try with smaller number of variables. With 55 variables, even if plot is made, it will be difficult to see patterns.
Sir, If independent variables are highly correlated in classification problems than also we should use these types of procedures or we can ignore that. Kindly help
For classification problems, try this:
ruclips.net/video/hCLKMiZBTrU/видео.html
@@bkrai Thank you sir. Your videos helped me a lot in learning ML and R coding. Thanks a lot sir again.
You are welcome!
Amazing tutorial. Thank you very much, I just subscribed to your channel.
Thanks and welcome!
@@bkrai I am a graduate student currently doing a research project with sg-LASSO and your video really helped me. Thanks again.
Thanks for comments!
Sir, can you do a tutorial for using glmnet package for doing logistic regression.
Thanks, I've added it to my list of future videos. Meanwhile you may also refer to this:
ruclips.net/video/AVx7Wc1CQ7Y/видео.html
Thankyou for sharing this knowledge sir.
I am having some errror.Follwing is the code and error
airglasso
Please helpout sir
I see you have used "
There is a typo in the code.
@@bkrai Its the typo while copypasting code over here sir.Even after that correction i getting same error
Share your code so that I can take a look
Sir I want to ask about the sequence, when in ridge regression you set the sequence as 0.0001 to 1 with length 5, for this sequence I got the lambda= 0.00001, when I changed the sequence to 0.001 to 1, my lambda also changed.
Can you please tell me what lambda should I select?
How do you find the final equation of the regression ? and the confusion matrix ?
Line-95 gives intercept and coefficients for the model. In regression problems, confusion matrix is not required. You can use RMSE or R-square.
Sir, i do have a question can this method apply to gene expression dataset for feature selection?
I'll have to look at the data to make suggestion.
@@bkrai the data usually contains small sample and hundred to thousands of gene variable Sir.
You can follow the steps in the video and let me know if there is any issue. You can also try:
ruclips.net/video/VEBax2WMbEA/видео.html
@@bkrai I will do. tqvm Sir 😊
Welcome!
Very Nice. Are the lambdas in SS's not different?
excellent sir..pls continue posting videos
Thanks for comments!
Sir.. you did not performed any standardization on data before running the alogrithm. Is standardization required or not before running this techniques.
Not required here but even if you do it, should be fine.
so good!!!! the video helps a lot! my first comment on RUclips. thank you very much!
Thanks for comments!
Excellent video.
Thanks for comments!
thanks so much sir. when we have the target variable as classification(0 or 1),will the same approach work?
For target variables as factor, you can use logistic regression.
thanks sir
sir, i have lot for independent factor variables. is there a model to crunch them?
What if my lambda = 1? Is it an issue with my model and CV?
Very informative. Thank you.
You are welcome!
Dear sir,
Please on below error i am getting while getting same code-
set.seed(1234)
ridge
Looks like your data has missing values which is causing this error.
Is there any way you could do a video on caret ensamble?
Thanks for the suggestion, I've added this to my list.
@@bkrai Thank you sir.
What if I wanted the programm to define the best valeu for lambda? I mean, without me putting a sequence in the code
Sir, Will you make sequential feature selection algorithm in r...
Thanks for the suggestion, I've added this to my list. For feature selection you can also refer to this:
ruclips.net/video/VEBax2WMbEA/видео.html
sir kindly tell me if we don,t want partitioning of data can we omit that part why we use train word in each model
It is always a good idea to partition data to avoid over-fitting. "train" is the function used for developing a model, that's why we need to use it.
thanks for ur reply but when i run ur program i got lots of errors
sir waiting for your video on multi level regression by using lasso and rigid regression
I would suggest first try same data that is used in the video. That will make sure codes have no error.
In this code, the train function is regressing against all data which doesnt allow for any feature engineering. can anyone clarify ?
The methodology takes care of that.
Dear Bharatendra, thank you a lot for the bright explanation! I would like to ask you how can I get p-values for coefficients from elastic net? Thank you in advance!
it's not simple, and is a somewhat new area of research. usually you don't worry about it, after all the only reason you see the variables is because the elastic net procedure decided they should be kept, remember?
Very well explained.
Thanks for comments!
Thanks a lot for the tutorial Sir. Really helpful.Just a small question though.
Is there a code that i can use to see the P value of the regression equation also also along with the R squared and RMSE values?
The output includes all of them.
When doing the scatterplot, why eliminate the target variable? Isn't helpful to see what variables are, and aren't, correlated to it?
It was mainly for discussing multicollinearity where target variable is not needed.
also when the dependent variables are discrete, can we convert them to factors and apply the same approach
You need logistic regression for discrete target variable.
sorry sir am asking lot of questions, in my case am having lot of independent variables that are factors and would like to crunch them in to few to make it simple. in my case target variable is "yes" or "no".glmnet function in R says it also supports binomial.
20:40 you say the purple variable (#6) is more important than blue (#5), but this always disagrees with the variable importance plots, which say nox (variable 5) is most important with rm (variable 6) the second most important. Did you mean to say the blue line was more important? You also say this at16:58
You are right, nox is contributing more to the model.
Okay thanks. Great video by the way there's a lot of great stuff in here in addition to what's in the title so thanks for making it, you're also the only person I've seen who speeds up the code writing portions instead of putting us to sleep to the sound of the keyboard slowly typing code so that was a smart touch too :)
Thanks for feedback!
Thank you sir. I expect more of your videos in youTube.
Thanks for the comments! You can find some useful links below:
Introductory R Videos: goo.gl/NZ55SJ
Machine Learning videos: goo.gl/WHHqWP
Deep Learning with TensorFlow: goo.gl/5VtSuC
Image Analysis & Classification: goo.gl/Md3fMi
Text mining: goo.gl/7FJGmd
Data Visualization: goo.gl/Q7Q2A8
many thanks, one of the helpful videos
Thanks for comments!
Thanks Mr Rai
Thanks for comments!
hats off sir
Thanks!
Love U so much. This video help me a lot
Thanks for comments!
Sir I have got this problem and i have checked my each value i do not have any missing value plz help
Error in na.fail.default(list(LTDA = c(0.26683158, 0.37309602, 0.40553323, :
missing values in object
Look at summary of data using 'summary' function. If there is missing data, it will show up as NAs.
@@bkrai Sir, I have seen through summary but it shows no missing data
lm
Sir please help
Sir I have checked my summary statistics but i did not found any missing value
wow Thank you Boss
Thanks for comments!
thanks allot sir
You are welcome!
Getting an error when trying to run my prediction and my dependent variable is a factor
For factor type dependent variable, try this:
ruclips.net/p/PL34t5iLfZddvv-L5iFFpd_P1jy_7ElWMG
@@bkrai thank you Sir, is it possible to get the confusion matrix showing the F1-score
Amazing
thanks!
Accuracy kese chk karenge sir
using RMSE.
i stopped at ridge step, it shows Error: Stopping
In addition: There were 26 warnings (use warnings() to see them)
DAMNNNNNNN
If you are already using the R code that I provided in the description area, then I'll suggest review the code as it may be a small error. Also check if you have missing data.
@@bkrai Sir the code you provided has some missing data. I tried to fill by looking at your video but it doesn't work. Can you share the exact code in this video? Thank you
Hi sir can you please tell me your mail id, i have a one problem statement, i need to know your suggestion on that dataset.
seemabharat@gmail.com
I get less RMSE for test data😂
It can happen sometimes, but mostly it is the other way.