Thank you for this presentation. I am a retired behavior analyst and my primary data was always the count of behavior per unit time. I often encountered days on which the learner performed well above what I expected or well below expectations. High points made me question my instruction, expectations, and instructional materials (e.g., a 3rd-grade reading passage accidentally got included among 6th-grade passages). Low points, poorer performance always suggested a bad start to the school day or feeling ill. As you said, it is very bad practice to throw away data in the interest of saving your idea of 'correctness.' Each outlier must be examined and explained.
This is exactly what would help people with dealing with outliers👍 If you can show us a coding example where keeping an outlier intelligently for training would make more sense than just dropping it then that will be great! For now, I am assuming that if a data has too many outliers and you want to incorporate those in your model then better is to use a tree based model (you can correct me if I am wrong). Thank you. 😀
I am someone who cont learn things ,if dont not understand intuition behind that.i feel you are too of my type . i love your videos:) .keep it up .......if possible ,please post some statistics concept videos, which are necessary for DS . thanks ..
Interesting for me, I am dealing with that problem right now. I was hoping you would've gone through the methods for outlier detection/classification; One class SVM, random forests, ... Thanks anyways.
Sir today I have seen many of your video. Thank you for sharing beautiful knowledge. Q: currently preparing time series model for Nifty using option chain data. Taking open interest, premium value and volumes data at every 5 minutes. Repeative failure while normalizing the data since the price changes, some times the strike price also changes which creats outliers. How to deal with this issue. Please guide.
If that 2-second intro music doesn't hype you up, I don't know what will 🔥
haha!
Great summary! Thank you.
Amazing work!!
Awesome explanation!
Thank you for this presentation. I am a retired behavior analyst and my primary data was always the count of behavior per unit time. I often encountered days on which the learner performed well above what I expected or well below expectations. High points made me question my instruction, expectations, and instructional materials (e.g., a 3rd-grade reading passage accidentally got included among 6th-grade passages). Low points, poorer performance always suggested a bad start to the school day or feeling ill. As you said, it is very bad practice to throw away data in the interest of saving your idea of 'correctness.' Each outlier must be examined and explained.
This is exactly what would help people with dealing with outliers👍
If you can show us a coding example where keeping an outlier intelligently for training would make more sense than just dropping it then that will be great!
For now, I am assuming that if a data has too many outliers and you want to incorporate those in your model then better is to use a tree based model (you can correct me if I am wrong).
Thank you. 😀
really great lecture！！！
Nice video 👍 it. Can you do videos on recommendation system.
I am someone who cont learn things ,if dont not understand intuition behind that.i feel you are too of my type . i love your videos:) .keep it up .......if possible ,please post some statistics concept videos, which are necessary for DS . thanks ..
Wonderful
Interesting for me, I am dealing with that problem right now. I was hoping you would've gone through the methods for outlier detection/classification; One class SVM, random forests, ... Thanks anyways.
“Btw, I got this really cool lobster hat for Christmas, hope you like it” 😂
Awesome！
Having domain knowledge is a must or at least someone you can consult somone on it if you dont know.
Thanks!
No problem!
Sir today I have seen many of your video. Thank you for sharing beautiful knowledge.
Q: currently preparing time series model for Nifty using option chain data. Taking open interest, premium value and volumes data at every 5 minutes. Repeative failure while normalizing the data since the price changes, some times the strike price also changes which creats outliers. How to deal with this issue. Please guide.
Amazing videos Ritvik. Just a nitpick, The second strategy should be winsorizing..not seen the term "windsoring" used.
Hi @ritvikmath, are you also planning to make series on Applied machine learning algorithms with the intuition and mathematics behind it? >