you are great teacher....great data scientist....hats off to you....because of you all the basic , small n small concepts getting cleared...you are faaabbb
I m a great fan of u & ur channel. Thanku so much for working hard. Itna mast Random forest shayad hi kisine explain kiya hoga youtube par. Not only Random Forest actually all other algorithms or different concepts in Data science , all videos of urs r the best of the best.
Hi, thanks for the great explanation but just a silly mistake I think after we did column sampling, we will need to give data point for prediction in same order as the features selected at 30:00
Its good that less but genuine people are following this channel. Knowledge is for everyone but only dedicated minds will make use of it, other just follow their adrenaline rush.
I think there is some mistake while using column sampling, in a prediction part you are passing the same input array to all the decision trees while it should be different according to the sampled dataframe.
if we do not specify any sampling ,by default is it taking hybrid(row and column sampling together)like some rows and feature(column) are taken together for training decision tree?
there is a small mistake in the column_row sampling(last one). df1,2,3 have different features which are trained on different models and while performing the prediction, you passed he same features for all three models and took majority(aggregation). apart from this rest is awesome. I have a question. lets say I run the random forest by giving my dataset with columns from F1 to F10 with target variable Y. I ran some sklearn model or something. i did train test split and i did fit and now, I want to perform prediction. since, each DT model in RF has different and only few cols(let's say 50%) and my test data will have all the cols. Are the test data features gonna be passed to respective DT model with respective features that the model was trained on? or is there some other mechanism? please explain.
All the features are gonna be passed to the DT models, the classification is done by each model using the subset of features they were trained on , which is then aggregated and you get the RandomForest prediction
sir hum random features selection karenge aur us par value prediction bhi karenge dt se lekin agar hume testing / input data whole features ka mila toh prediction kaise hoga ? i think row sampling is fine but features sampling not understood .
While teaching preprocessing you mentioned that we check for duplicate data (rows) and drop it as it can create problem while building model then how random forest handles the duplicate data which comes from sampling with replacement techniques.
Good question, Atul. See, the problem with duplicate data is that it increases model bias which may lead to overfitting. But the way RF works, it is able to handle this bias.
I am still confused on sampling without replacement, if we have 10 features and we selecting let say 5 features for each DT, then my 10 features will be exhausted in 2 DTs only. Then how come it's distributing in i.e. 100 DTs? Same with row sampling we are giving 25% of rows to each DTs then it should be exhausted after 4 DTs, then how come we are training the other DTs. Please help me with this, I am totally confused with 'without replacement' option.
In sampling without replacement: For the 1st tree, if 5 columns are to be selected: For the 1st column, all 10 columns are available for selection. After the 1st column is selected, then the next column could be any of the remaining 9 columns.. So, this avoids repeating of the columns. For the next DT, the process starts from scrath. All the 10 columns are available for selection and the same process as above follows. In sampling with replacement, after the 1st column is selected, while selecting the 2nd column, it will have all the 10 columns to choose from, including the column already selected as 1st column.
I think the explanation of bootstrapping is wrong in the video. As per Wikipedia (en.wikipedia.org/wiki/Bootstrap_aggregating) , in bootstrapping process the size of the sample dataset has to be same as the original dataset and it is not a smaller subset of the original dataset as explained in the video. I think what you have explained (creating smaller samples from the original dataset) is concept of bootstrapping but applicable from statistics standpoint and not in context of Random Forests. Still I will be grateful if you can kindly explain your reasoning with some related articles.
Hi, i have a small question about decision trees. Is it ok to have a decision tree with a max depth of 7, bcs i noticed how it produced a big tree but also had the better accuracy training and test score than if I had reduced the max depth.
@@campusx-official oh okay, both my train and test data show to be ok in terms of not over or under fitting ! Btw thank you sir, your video literally saved my life! Bless you
completed 90 videos with notes within 3 weeks 🥳🥳🥳🥳. I still remember all the topics and working. Your teachings are blessings from god.
Bro could you please share that notes
Yes can u pls provide notes
@@sourabh_321 I have also prepared the notes. But those includes only theory and maths portion, not the coding part.
@@hardik3809 I don't need coding part I need theoretical part if u can provide me this will help me a lot ❤️
bhai bhejde paper aa rhe h
you are great teacher....great data scientist....hats off to you....because of you all the basic , small n small concepts getting cleared...you are faaabbb
I m a great fan of u & ur channel. Thanku so much for working hard. Itna mast Random forest shayad hi kisine explain kiya hoga youtube par. Not only Random Forest actually all other algorithms or different concepts in Data science , all videos of urs r the best of the best.
You deserve 1M Subscribers!!
you Truely deserve more than 1M followers. mei apki bahut badi fan hu sir hu dil se, app jb padhate ho to padhayi se pyaar sa ho jata h.
Best ever channel & best ever teacher for data science . Thanku very very much.
Hi, thanks for the great explanation but just a silly mistake I think after we did column sampling, we will need to give data point for prediction in same order as the features selected at 30:00
It is best Channal for data Science It will grow 1M in 2023 i swear
Nhi hue bhai🙃
Its good that less but genuine people are following this channel. Knowledge is for everyone but only dedicated minds will make use of it, other just follow their adrenaline rush.
the way you teach is absolutely amazing , keep up the good work ,thankyou
Your videos on ML are amazing. Following this playlist!!
But little bit long
@@Human12358yes
But worthy@@Human12358
you are thanos for complex topics
Your explanation on each small point is next level! Great job sir.
Great content totally diffrent way of teaching!! Mark my word nobody spent so much time on a Single project but here the story is diffrent!!
Bahut badiya sir. God bless you 🙏🙏
massive respect ❤,
you are a great teacher
I think there is some mistake while using column sampling, in a prediction part you are passing the same input array to all the decision trees while it should be different according to the sampled dataframe.
yes, it would be
yes we cant predict values like that if we have input data scaled with diff. features , yes sir se thoda mistake hogaya hain.
In column sampling prediction, How can we give same data to each decision trees as there can be the case that all decision tree have different columns
if we do not specify any sampling ,by default is it taking hybrid(row and column sampling together)like some rows and feature(column) are taken together for training decision tree?
Thank you so much sir 🙏🙏🙏
I m a great fan of u & ur channel
there is a small mistake in the column_row sampling(last one). df1,2,3 have different features which are trained on different models and while performing the prediction, you passed he same features for all three models and took majority(aggregation). apart from this rest is awesome.
I have a question. lets say I run the random forest by giving my dataset with columns from F1 to F10 with target variable Y. I ran some sklearn model or something. i did train test split and i did fit and now, I want to perform prediction. since, each DT model in RF has different and only few cols(let's say 50%) and my test data will have all the cols. Are the test data features gonna be passed to respective DT model with respective features that the model was trained on? or is there some other mechanism? please explain.
same doubt bro .comment if u get the answer
All the features are gonna be passed to the DT models, the classification is done by each model using the subset of features they were trained on , which is then aggregated and you get the RandomForest prediction
Very Well Explained
Thanks Sir
sir hum random features selection karenge aur us par value prediction bhi karenge dt se lekin agar hume testing / input data whole features ka mila toh prediction kaise hoga ? i think row sampling is fine but features sampling not understood .
While teaching preprocessing you mentioned that we check for duplicate data (rows) and drop it as it can create problem while building model then how random forest handles the duplicate data which comes from sampling with replacement techniques.
Good question, Atul. See, the problem with duplicate data is that it increases model bias which may lead to overfitting. But the way RF works, it is able to handle this bias.
I am still confused on sampling without replacement, if we have 10 features and we selecting let say 5 features for each DT, then my 10 features will be exhausted in 2 DTs only. Then how come it's distributing in i.e. 100 DTs? Same with row sampling we are giving 25% of rows to each DTs then it should be exhausted after 4 DTs, then how come we are training the other DTs. Please help me with this, I am totally confused with 'without replacement' option.
In sampling without replacement:
For the 1st tree, if 5 columns are to be selected: For the 1st column, all 10 columns are available for selection. After the 1st column is selected, then the next column could be any of the remaining 9 columns.. So, this avoids repeating of the columns.
For the next DT, the process starts from scrath. All the 10 columns are available for selection and the same process as above follows.
In sampling with replacement, after the 1st column is selected, while selecting the 2nd column, it will have all the 10 columns to choose from, including the column already selected as 1st column.
I am not able to find the link for the visualisation tool
best of the best!
ca we take two output in RF
column sampling function not working .
can anybody resolve it ?
I think the explanation of bootstrapping is wrong in the video. As per Wikipedia (en.wikipedia.org/wiki/Bootstrap_aggregating) , in bootstrapping process the size of the sample dataset has to be same as the original dataset and it is not a smaller subset of the original dataset as explained in the video. I think what you have explained (creating smaller samples from the original dataset) is concept of bootstrapping but applicable from statistics standpoint and not in context of Random Forests. Still I will be grateful if you can kindly explain your reasoning with some related articles.
You are great
Best one
Hi, i have a small question about decision trees. Is it ok to have a decision tree with a max depth of 7, bcs i noticed how it produced a big tree but also had the better accuracy training and test score than if I had reduced the max depth.
Yes you can have a decision tree of max depth 7 or more. It depends on the data
@@campusx-official oh okay, both my train and test data show to be ok in terms of not over or under fitting !
Btw thank you sir, your video literally saved my life! Bless you
finished watching
sir please provide notes for this ml playlist
Thank u sir
THANK YOU 3000
How come 65th video after 62nd(ensemble technique) ??
Check the playlist
@@campusx-official okay got you!
Wow
This video was a bit confusing ngl
Since you're speaking Indian, at least write the title in Indian!!! So we avoid wasting our time!
It's his choice.bkl