Glad ...to see this sort of contents ...Believe it or not these channel is going to be the best in ML,DL and AI .. Some thing equivalent to an University ...
I pre-booked the Kindle version India. The book is so lucid to understand. I really appreciate the efforts in the write-up and the knowledge that's been provided in the book. Honestly learned alot.
I wonder why two dislikes, you guy's can comment what is missing so that Abhishek Thakur will respond to you and we all can benifit from this. My suggestion in future if you are dislike please comment why and what is missing from Abhishek. Look below he is patiantly answering most of the comments to his best. Personally I learned a lot but need more hands-on to get this skill. A special thanks Abhishek Thakur for serving in this manner. You are my motivation to deliver content for free.
Amazing video Abhishek. You keep emphasizing on the importance of cross validation setup so how do u go decide the correct cross validation setup for a given problem?
This is one of the most comprehensive tutorial on blending and stacking very amazing work from your side I have a query if i have a multi class targets with probabilities how will i compare the accuracy ??
While creating new features using existing features, which features to consider (features with low correlation with target) or can I use features which have high correlation with target to create new features. Or both ?
Awesome work !!! ❤️ Thankew so much sir 🙏 Earlier i had question in my mind how to find the optimal model cofficients but today i got the answer to that. Also good to to know about rank averaging 🙌
How do you do blending with time series split ? total valid set size < total train set size in this case and L1 will be the time series split folds. Should I then use only the valid set OOF preds to do L2 (probably without regards to time anymore and treat it as regular K-fold here onwards) ? Is this the correct approach ?
Hi Abhishek, thanks for the great content you are providing. I wanted to hear your opinion on AutoML, how do you think it will evolve and how will it change things.
Hi Abhishek. I know that I'm late but can I use optuna to find optimate weights for ensembling? After all, it's just a function that needs to be optimized right? Won't optuna work well?
Really cool video as always Abhishek! I actually meant to ask you how did you setup this code server that you are working on? I recently built a workstation PC for data science stuff, but I do not want be "immobile" because it's a PC. How can I replicate this setup which you use (without using Remote Desktop)? Any resources or links would be super helpful. Thanks :)
Beautiful video Abhishek , This video is more insightful and gave clear roadmap in blending and stacking; How about using mean , median value to use instead of averaging method can we use those can we use it in weighted averaging
Hey Abhishek ! Is it a standard practice to use the ID column in the dataset for training the model? I understand that keeping in the ID column might convey some added information to the model in competitions, but is it an advisable practice while training models for prod?
@@abhishekkrthakur You don't drop the id column from the training set for either of the three L1 level models, and ID column is again used in the blending.py file to merge the datasets together.
@@GairolaAbhijit1991 you made me look at the code again. haha :D yes. i do keep ID column. its a way to track the data points for all the models. but do I use the ID column to train the models? NO! Look carefully, I have always defined which columns to use for training the models :)
Thank you for the video. As usual amazing high quality content! Is there a difference between when the words 'blending' and 'stacking' are used? Are they just synonymous used? I tried to read up on it but could not find a clear distinction.
Many people have many different definitions. here is mine: stacking is when you use a model to combine base models and blending is when you combine models just by weighted average. Although weighted average can also be thought of as a model. Some also say stacking is kfold and blending is on a holdout set. :)
If you want more stable coefficients when finding the best weights for blending, please use a different initialization.
very informative video...pls make videos on image annotation from scratch
like annotating images manually?
@@abhishekkrthakur whatever people in kaggle competition use.
I DM'ed you in twitter regarding this.
sorry to bother Abhishek, can u please share the link from where we can get the source codes u are using . specially the optimal_weights.py file .
Glad ...to see this sort of contents ...Believe it or not these channel is going to be the best in ML,DL and AI ..
Some thing equivalent to an University ...
Thank you 🙏
There is a lot of garbage on youTube but this dude you can tell he's a professional.
Really all your contents are awesome,your books ,videos all r much more than an university .Please keep sharing
Also please make a video on feature selection on tabular data....
Please abhishek sir🙏🙏🙏🙏🙏
Vote for this topic also!!!
Consider it done 💙
@@abhishekkrthakur Thanks a lot sir
I have bought and gone through your book. It is just excellent...I would say it is ultimate... Thanks...
Glad you like it. Please do consider writing a review on Amazon too :)
Wow, what I needed, thanks for the helpful tutorial
absolutely best timing on your videos thanks !!!
The next video on reinforcement learning? you can use the Halite competition in kaggle to make us understand
Its on my list :)
You're a true Grandmaster.😅😅 it's not easy to skip your lessons
Man you are on another level..
I pre-booked the Kindle version India. The book is so lucid to understand. I really appreciate the efforts in the write-up and the knowledge that's been provided in the book. Honestly learned alot.
Thank you! Please do consider writing a review on Amazon too :)
@@abhishekkrthakur I will indeed.
I wonder why two dislikes, you guy's can comment what is missing so that Abhishek Thakur will respond to you and we all can benifit from this. My suggestion in future if you are dislike please comment why and what is missing from Abhishek. Look below he is patiantly answering most of the comments to his best. Personally I learned a lot but need more hands-on to get this skill. A special thanks Abhishek Thakur for serving in this manner. You are my motivation to deliver content for free.
Amazing video Abhishek. You keep emphasizing on the importance of cross validation setup so how do u go decide the correct cross validation setup for a given problem?
Please make video on GAN
ua-cam.com/video/gT8-wDPLOBg/v-deo.html
This is one of the most comprehensive tutorial on blending and stacking very amazing work from your side
I have a query if i have a multi class targets with probabilities how will i compare the accuracy ??
Great video
i bought your book from pothi. Looking more books from your end.
That was a fantastic video GrandMaster! 😁👍
much-awaited video ! thank you so much sir
You are doing linear blending; however, your optimized weights are not adding up to 1. Do you think this is a problem?
While creating new features using existing features, which features to consider (features with low correlation with target) or can I use features which have high correlation with target to create new features. Or both ?
Very good tutorial! What about ensembling when we have time series data?
why are you creating a fold column ?
Please creat a video on feature selection and extraction techniques..
Just amazing Sir, you helped me to clear lost of concepts.
Awesome work !!! ❤️
Thankew so much sir 🙏
Earlier i had question in my mind how to find the optimal model cofficients but today i got the answer to that.
Also good to to know about rank averaging 🙌
How do you do blending with time series split ? total valid set size < total train set size in this case and L1 will be the time series split folds. Should I then use only the valid set OOF preds to do L2 (probably without regards to time anymore and treat it as regular K-fold here onwards) ? Is this the correct approach ?
Thank you for this sir!
I can't wait to learn more from you
Hi Abhishek, thanks for the great content you are providing. I wanted to hear your opinion on AutoML, how do you think it will evolve and how will it change things.
Hi Abhishek. I know that I'm late but can I use optuna to find optimate weights for ensembling? After all, it's just a function that needs to be optimized right? Won't optuna work well?
yes you can. im not sure i made a video about it but you can read it with code in my book: bit.ly/approachingml. its free
Really cool video as always Abhishek!
I actually meant to ask you how did you setup this code server that you are working on?
I recently built a workstation PC for data science stuff, but I do not want be "immobile" because it's a PC. How can I replicate this setup which you use (without using Remote Desktop)?
Any resources or links would be super helpful. Thanks :)
Why can't I like this video more than once? It's unfair!!
Ensembling time series models into one.
Beautiful video Abhishek , This video is more insightful and gave clear roadmap in blending and stacking; How about using mean , median value to use instead of averaging method can we use those can we use it in weighted averaging
thanks! and yeap you can try many variations. even gm, hm, etc. I didnt go in details there and left it as something to explore ;)
@@abhishekkrthakur Noted Thanks 🙏 Sir
Can you do a video for the final optimization of multi targets?
Got it thanks
got multiple values for argument 'X' Got this error possible solutions to solve this?
Great!!
Hey Abhishek ! Is it a standard practice to use the ID column in the dataset for training the model? I understand that keeping in the ID column might convey some added information to the model in competitions, but is it an advisable practice while training models for prod?
No No. Never keep the id column to train the model. Did I do that?
@@abhishekkrthakur You don't drop the id column from the training set for either of the three L1 level models, and ID column is again used in the blending.py file to merge the datasets together.
@@GairolaAbhijit1991 you made me look at the code again. haha :D yes. i do keep ID column. its a way to track the data points for all the models. but do I use the ID column to train the models? NO! Look carefully, I have always defined which columns to use for training the models :)
Thank you for the video. As usual amazing high quality content! Is there a difference between when the words 'blending' and 'stacking' are used? Are they just synonymous used? I tried to read up on it but could not find a clear distinction.
Many people have many different definitions. here is mine: stacking is when you use a model to combine base models and blending is when you combine models just by weighted average. Although weighted average can also be thought of as a model. Some also say stacking is kfold and blending is on a holdout set. :)
This masterpiece will surely help everyone ! Thank you Abhishek Bbhaiiay ! There is one request , can u do a small video in Hindi ! Just for fun :)))
Great video! Thanks! Learned a lot. I will assume that the code from this video can be found in your book. Is this correct?
Not all but some :)
How can I get your Machine Learning book in Bangladesh?
of course I will buy!
Amazon?
there are not Amazon in BD. Any other option?
thank you for this vid.
I cannot see where the target is dropped?
everywhere. if it's not dropped, its not used. if it's used, its used in cross validated manner to generate target encoded variables.
very cool
The video is long, but really easy to follow!
xtrain column names
hehe... eventually fixed
Dropping bombshells here!
Long one. But worth it xD..