This is the best explanation i have ever seen for any ml algorithm. I was very confused on xgboost and don't find a detail and complete explanation on xgboost anywhere. You have connected each dots and explained as a story. Keep up this amazing work. Now, i would definitely go and check other concepts here. Thanks! :)
for multiple features, i believe it will evaluate each feature separately and find the best split which has the highest gain and if one categorial variable is seen then we have to do one hot encoding (if categorical) in case of binary(0,1) It can work
if we have two features then one thing we can do is that we find gain for using each feature and the feature that gives the maximum gain is selected for splitting.
43:18 We can scan all features , and do all possible splits for all features, then we will calculate gain and similarity score , and select feature which has max gain . 43:56 idenntify unique values in catagorical columns , and consider both as a potential split point ., then calculate gain and similarity scores ., and select with maximum gain . 43:59 idenntify unique values in catagorical columns , and consider everyone as a potential split point ., then calculate gain and similarity scores ., and select with maximum gain . split can have two ways 1 . 1 node out of all can always have a single and other can be a rest classes in a group . like { small } and { medium , large } data - like that . 2 . divide it into non empty , non overlapping subsets . in all possible ways .like { very small,small } , { large , very large } . now for possible split , divide the data based on the subset of categories. calculate similarity scores and gain , and go with max gain :)
Awesome Content Sir Sir please give one life advice should i give the GATE 2024 exam or focus on my placements> Please tell both pross and cons of gate and placements. Much Appreciated
For categorical label you take the log odds and revert it back to probability and train the model with new probabilities values as equivalent to what we did with residuals in the regression case. And train it with decision trees. Since we are using the regression trees the model assumes that it is a regression problem and it is not good if we are dealing with categorical data. So those regression tree values don't output in proper probabilities which ranges from 0-1 instead it gives values beyond that range so to countermeasures it we use logodds again to convert those probabilities values. And train another decision trees with new probabilities converted( log odds values) as label and repeat the process.
What you mentioned is not right. You are talking about categorical variables in a classification problem where as the question was for a regression problem. The concept of log odds makes sense only for a classification problem. To use categorical variables in a regression problem would be to use some form of encoding and the resultant numeric values will be split by following which similarity scores will be evaluated. Look at the campusx video for GBM for classification where log odds and probability inter conversion is used for the classification problem
hi sir , we hope you doing good. i got some query an you please answer it . i worked with keras tuner after wathing your video , i just watched your video on keras classifier and now i also heard of NAS neural network search . can you please help me with differences among these three ?
I'd recommend u to upload videos in 2x coz u speak remarkably slowly and consequently ur videos become enormously long, which ideally don't make any sense at all.
I beg to differ. Most audience here love the content and rightfully so: it's first of all free, secondly it's better than all paid courses especially because he goes deep into the fundamentals and his pedagogical technique is stellar. If you are referring to language of choice (and not accent), then you can send him an email directly which is understandable since everyones Hindi isn't strong. But commenting on his accent and telling it is discouraging is not representative of the majority
dont kid here. There is nothing wrong with his hindi accent or anything as such. He teaches so well all these topics which are very much highly paid outside. Going through a research paper and teaching stuff with research paper only a good teacher/human could do this.
This is the best explanation i have ever seen for any ml algorithm. I was very confused on xgboost and don't find a detail and complete explanation on xgboost anywhere. You have connected each dots and explained as a story. Keep up this amazing work. Now, i would definitely go and check other concepts here. Thanks! :)
for multiple features, i believe it will evaluate each feature separately and find the best split which has the highest gain and if one categorial variable is seen then we have to do one hot encoding (if categorical) in case of binary(0,1) It can work
if we have two features then one thing we can do is that we find gain for using each feature and the feature that gives the maximum gain is selected for splitting.
Absolutely love this series. Can you please make videos on LightGBM regressors and classifiers as well? 🙏
Happy birthday sir.I love your commitments, discipline and dedication❤
Kindly sir upload all4 400 data science interview questions list which you tell us in data science interview questions video
Happy Birthday Nitish Sir. One of the best teacher I have ever seen.
Janmdin ki bahut bahut shubkaamnaayein aur badhaaiyaan Sir.
Loved this one ❤,simply explained. Thank you very much.
Thanks you Nitish.
Good explanation sir❤❤
Thank you sir..
You did your best..
❤
43:18
We can scan all features , and do all possible splits for all features, then we will calculate gain and similarity score , and select feature which has max gain .
43:56
idenntify unique values in catagorical columns , and consider both as a potential split point ., then calculate gain and similarity scores ., and select with maximum gain .
43:59
idenntify unique values in catagorical columns , and consider everyone as a potential split point ., then calculate gain and similarity scores ., and select with maximum gain .
split can have two ways
1 . 1 node out of all can always have a single and other can be a rest classes in a group . like { small } and { medium , large } data - like that .
2 . divide it into non empty , non overlapping subsets . in all possible ways .like { very small,small } , { large , very large } .
now for possible split ,
divide the data based on the subset of categories.
calculate similarity scores and gain , and go with max gain :)
best explanation
I was waiting for this... Thank you sir.
sir please make a video of what are stats for which algorithms so we learn parallel
Awesome Content Sir
Sir please give one life advice should i give the GATE 2024 exam or focus on my placements>
Please tell both pross and cons of gate and placements.
Much Appreciated
Thank You Sir.
Happy Birthday Nitish Sir
Excellent letcure ❤
Thank you so much sir. Can i please request a video on Fourier transformation series for deep leanring and computer vision( mainly cnn)
Thanks
Happiest Birthday Nitish Sir🎉
For categorical label you take the log odds and revert it back to probability and train the model with new probabilities values as equivalent to what we did with residuals in the regression case. And train it with decision trees. Since we are using the regression trees the model assumes that it is a regression problem and it is not good if we are dealing with categorical data. So those regression tree values don't output in proper probabilities which ranges from 0-1 instead it gives values beyond that range so to countermeasures it we use logodds again to convert those probabilities values. And train another decision trees with new probabilities converted( log odds values) as label and repeat the process.
What you mentioned is not right. You are talking about categorical variables in a classification problem where as the question was for a regression problem. The concept of log odds makes sense only for a classification problem. To use categorical variables in a regression problem would be to use some form of encoding and the resultant numeric values will be split by following which similarity scores will be evaluated. Look at the campusx video for GBM for classification where log odds and probability inter conversion is used for the classification problem
Happy Birthday Nitish Sir 🎉🎂🎊🎈
Plz upload xgboost for classification also
which device uses for writing pad.
Happy birthday sir
hi sir , we hope you doing good.
i got some query an you please answer it . i worked with keras tuner after wathing your video , i just watched your video on keras classifier and now i also heard of NAS neural network search . can you please help me with differences among these three ?
Sir maths for machine learning book ke lectures banado Zindagi savar jayegi sir koi nahin rok payega
Happy birthday sir 🎉🎉🎉
Thanks Sir😊
Awesome
Itnq detail me playlist banane ke baad toh nind me bhi aap Machine learning algorithm apply kar skte 😁😁
Why the sum of residuals is not coming zero since we are mean subtracting
thanks great video
Super❤
Why aren't your videos automatically translated?
Is this ml?
Upload the next please!
THANK YOU..! 3000❤
I'd recommend u to upload videos in 2x coz u speak remarkably slowly and consequently ur videos become enormously long, which ideally don't make any sense at all.
Man, you're contents are great, but you accent really discourage the audience...
I beg to differ. Most audience here love the content and rightfully so: it's first of all free, secondly it's better than all paid courses especially because he goes deep into the fundamentals and his pedagogical technique is stellar. If you are referring to language of choice (and not accent), then you can send him an email directly which is understandable since everyones Hindi isn't strong. But commenting on his accent and telling it is discouraging is not representative of the majority
Don’t watch
dont kid here. There is nothing wrong with his hindi accent or anything as such. He teaches so well all these topics which are very much highly paid outside. Going through a research paper and teaching stuff with research paper only a good teacher/human could do this.
Happy Birthday Nitish Sir