ek number session ... in easy terms ... BIAS is the inability of ML algorithm to capture the 100 percent or exact relationship. To understand bias one must think why do we need a ML in first place. In mathematics or physics we have absolute relationship or formula between dependent and independent variables like s=ut+1/2 at2 (std 7 Physics) or SI = P*R*T so for computing cases like we have absolute formula we don't need any ML algo. ML try to do the same i.e. estimate a formula, let say I want to calculate the purchasing power (P) so I train a model with different variables like income,age, family income and m model fetches a formula P = wo+ b1*income+b2*age + b3* family income..... So this formula is not absolute or universal as its derived by a specific ML algo for specific data but let say by miracle we derive a formula that exactly calculates the purchasing power with 100 percent accuracy so for that model bias is 0 as the model accurately captures the relationship..... Variance ---- Talking about variance, in short way the difference in fits between data set is called variance , imagine we used that same miracle formula in test data and data fits 100 percent as in we get 100 percent accuracy(for different test set) then we can say that the variance is 0 which means the ML formula is perfect or let say when use the same miracle formula in test set we get 50% accuracy which means the bias was low but variance is high as formula didnt work well with unseen (test) data... SO in an imaginary world if bias is 0 and variance is also 0 then my friend you have discovered a formula not an estimation .... In a practical world we aim for a model with low bias and low variance..... Subscribe Krish Channel if this helped
I would EDA, cuz that is more applicable in the job scenarios, i.e. it depends on the role, but generally, most roles, require strong EDA knowledge, so, I would go for EDA 7 days. next,
Good morning krish.. You have really made my foundation very strong before that I was null in statistic and machine learning since from non technical background.. Now I can read very high level books and could really understand.. You are really great value addition to my learning path..
Hello sir I started every morning with a new session of machine learning. And last 6 days teach me a lot about machine learning algorithms. Thank you very much for this playlist.
Hi Sir, At 1:11:00, I think you had mistakenly spoken the wrong terms for High Bias & low Bias. It should be like for High Bias-> Not perform well, Low Bias-> Perform well. We use Low Bias & low variance for the Generalized Model as it performs well. Correct me if I am wrong.
at 1:11:31 , I guess its wrong if the model captures the good relationship(between dependent and independent variable) in data then it has low bias not high bias. Low bias means that model output the formula is flexible (low bias) to capture the relationship , high bias means that the accuracy is low and model is unable to capture the actual data points .. please verify guys
This video is incredible, and very well explained . But if we have more than one feature in our dataset, should we make the feature selection first and then perform the elbow test?
Is the silhouette score applicable to hierarchical clustering? as some clusters are within other clusters. How do we differentiate a(i) from b(i) then?
sir can you make an urgent lecture on cluster labeling problem ?? document cluster labeling thing ? and what if we enhance this issue as hierarchical cluster labeling thing ?
K means clustering is not mathematically clear. The line you're drawing connecting the two centroids is ok, but how does that perpendicular line drawn. means how is that perpendicular line decided? Also for any new point, will that line be used to classify for k nearest neighbours is to be used?
In k means clustering, is there an assumption in numbers of observations and variables? Would having variables greater than observation affect the results of clustering and make it less accurate?
I don't understand after knowing the clusters we draw the histogram in hierarchical clustering and you are showing we need to draw a parallel like and the number of vertical lines it intersects will be number of clusters?? I mean we already drawing the histogram based on the clusters. Doesn't make sense what you told.
ek number session ... in easy terms ... BIAS is the inability of ML algorithm to capture the 100 percent or exact relationship. To understand bias one must think why do we need a ML in first place. In mathematics or physics we have absolute relationship or formula between dependent and independent variables like s=ut+1/2 at2 (std 7 Physics) or SI = P*R*T so for computing cases like we have absolute formula we don't need any ML algo. ML try to do the same i.e. estimate a formula, let say I want to calculate the purchasing power (P) so I train a model with different variables like income,age, family income and m model fetches a formula P = wo+ b1*income+b2*age + b3* family income..... So this formula is not absolute or universal as its derived by a specific ML algo for specific data but let say by miracle we derive a formula that exactly calculates the purchasing power with 100 percent accuracy so for that model bias is 0 as the model accurately captures the relationship..... Variance ---- Talking about variance, in short way the difference in fits between data set is called variance , imagine we used that same miracle formula in test data and data fits 100 percent as in we get 100 percent accuracy(for different test set) then we can say that the variance is 0 which means the ML formula is perfect or let say when use the same miracle formula in test set we get 50% accuracy which means the bias was low but variance is high as formula didnt work well with unseen (test) data... SO in an imaginary world if bias is 0 and variance is also 0 then my friend you have discovered a formula not an estimation .... In a practical world we aim for a model with low bias and low variance..... Subscribe Krish Channel if this helped
A humble request to you @Krish, make next live streams on Deep Learning.
ya
Yes
I would EDA, cuz that is more applicable in the job scenarios, i.e. it depends on the role, but generally, most roles, require strong EDA knowledge, so, I would go for EDA 7 days. next,
@@kkevinluke looks like your opinion won. And I also agree with you.
I got placed at tiger analytics
Credit goes to u krish
Your videos helped me to crack the interview
Hi Rahul ,congrats .please share interview quesions
Good Evening Krish. Your contents is absolutely a gold mine. Please arrange Deep Learning sessions next :)
Thanks
You are the best teacher that I have in my life in this domain,thanks a lot to share this kind of knowledge...
Good morning krish.. You have really made my foundation very strong before that I was null in statistic and machine learning since from non technical background.. Now I can read very high level books and could really understand.. You are really great value addition to my learning path..
Excellent, just excellent. Thanks
Super Explanation as always. hats off
Thank You
1.75 speed is he best way to watch and lot of information covered in less time
Excellent and knowledge gaining session and every second spend was gain. Thanks alot 😊 keeping helping and sharing the knowledge & concepts 💐💐💐
you are one of the best teachers any student can have..❤
Hello sir I started every morning with a new session of machine learning. And last 6 days teach me a lot about machine learning algorithms. Thank you very much for this playlist.
Please cover XGboost'GBM and catboost in live videos so we can understamd learn better
Hi Sir, At 1:11:00, I think you had mistakenly spoken the wrong terms for High Bias & low Bias. It should be like for High Bias-> Not perform well, Low Bias-> Perform well. We use Low Bias & low variance for the Generalized Model as it performs well. Correct me if I am wrong.
Pata hai bsdk galti se boldiye sir iske liye comment krne ki jarurat nai thi gyaan mat chodo
First thing First !
Great session 👏 👌 👍
Thanks for this great Tutorial.
Thank you for the lecture
Yes DEEP LEARNING NEXT!
finished watching
thanks! really want know about exact definition of bias & var
great teaching
Amazing explanation thank you sir
Krish Naik Sir is Awesome
at 1:11:31 , I guess its wrong if the model captures the good relationship(between dependent and independent variable) in data then it has low bias not high bias. Low bias means that model output the formula is flexible (low bias) to capture the relationship , high bias means that the accuracy is low and model is unable to capture the actual data points .. please verify guys
Thank your sir Krish
Keep it up.
Please start mock interview sessions as well
Hello sir, you are doing great job. do you have any video related to OPTIC clustering?
Beautiful sir....
Hello sir take care of your health
superb.....!!
Thank you so much sir❤️
Depends on the data points
Hi Krish, Are you planning to take ML (Deep Learning) session?
sir pls make video on homogeneity, completeness, V-measure and Davies-Bouldin Index
The video was very good. But how to calculate the feature importance after k-means clustering?
Hello @Krish, thank you for the explanations. Please do an extensive depth in EDA sessions next. I appreciate your efforts very much, thanks again.
will you do deep learning series?
Hi krish sir its learning from you.
Can you please detailed video of Principle components analysis
east or west naik sir is suppper duper best
You are just amazing Sir. 😊
10 out of 10
Please make some videos on soft clustering algorithm (ex. Fuzzy C Means)
10/10 rating
A humble request to you @Krish,make next live session streams on Machine learning practice and practicals
Yes Deep learning course
This video is incredible, and very well explained . But if we have more than one feature in our dataset, should we make the feature selection first and then perform the elbow test?
Is the silhouette score applicable to hierarchical clustering? as some clusters are within other clusters. How do we differentiate a(i) from b(i) then?
@KRISHNAIK SIR, KINDLY PROVIDE THE DBSCAN VIDEO LINK
sir can you make an urgent lecture on cluster labeling problem ?? document cluster labeling thing ? and what if we enhance this issue as hierarchical cluster labeling thing ?
Mil gya bhai ml padhna ka channel ekdum maja aagya sir
Silhouette score
Please let me know on which kind of data like linear ,non linear etc which algorithm works better
Quick qq. High bias meaning better accuracy. ??
No Low Bias & Low variance .
What are the type of Biases can there be in a dataset? how to answer this question ?
Andrew NG of INDIA==Krish Naik Sir
Could someone share github link which is being referenced at 51:51?
Can I've the git hub link here please 😵💫
Hello sir. Do you, by any chance, know about the assumptions of k means cluster analysis in the case of large variance?
K means clustering is not mathematically clear. The line you're drawing connecting the two centroids is ok, but how does that perpendicular line drawn. means how is that perpendicular line decided? Also for any new point, will that line be used to classify for k nearest neighbours is to be used?
Sir can you please provide the github link?
In k means clustering, is there an assumption in numbers of observations and variables? Would having variables greater than observation affect the results of clustering and make it less accurate?
Hello Krish! How it is possible to have 3 centroids when k=2 is specified as you told at 32:00 while introducing kmeans plus?
51:27 k means cant do cluster like this , kmeans created convex pattern in data
10/10
I don't understand after knowing the clusters we draw the histogram in hierarchical clustering and you are showing we need to draw a parallel like and the number of vertical lines it intersects will be number of clusters?? I mean we already drawing the histogram based on the clusters. Doesn't make sense what you told.
how to find eps and impis in dbsan
Sir, if low bias - high variance is overfitting and high bias - high variance is underfitting , then what is high bias - low variance ?
That is practically not possible because u will not get a model that performs bad on training data but somehow performs well on test data.
silhouette Code is dam tough to understand Sir 😞
Sir pls make a video ON pea
pca*
silhoit score
can i know the matrial link ?
anyone knows where I can get data science/ml internships? I am in third yr of comp eng
Where is the Github link for this?
I didnt find the githuub link sir