Preparing for a Machine Learning / Data Sciece Interview ? Machine Learning Interview Decoded : Machine Learning , Deep Learning, Natural Language Processing and more @ MachineLearningInterview.com
I have a feedback if that's cool. Maybe try to use some animations and go a little bit more in depth into the topic for better understanding. Thank you though this did clear my understanding of gower.
SGD will converge faster no matter the size of the dataset, but will not converge as close to the local minimum as batch gradient descent (it can still be very very close), this trade off is worth it for how much computationally cheaper it is.
I´m not sure why you need this but may this could help you as well . a + t * (b - a)= x if t = 0 than x= a if t = 1 than x= b ,if t= -3.14 than you get ratio a to b of -3.14 (x - a) / (b - a) =t this is other whay round if x is in between a and b you get t =0.563 where the Ratio is. lerrp , inv lerp and remaping
when I calculate SHAP for one sample the result that I get is different than the result when I calculate SHAP for that sample and the other one (two samples), why? imagine fimdimg SHAP for one sample (one row) and two samples (two rows) which one sample is share between these two seup. I get different SHAP in these two setups for the share sample. What is the reason?
If I am not wrong, the example given is not ideal for explaining complementNB. ComplementNB should be used in imbalanced data when the response variable is not binary i.e. it can have more than 2 options. If the response variable is binary in nature, then the complementNB and multinomialNB will return the same result, making the whole exercise meaningless.
It was mentioned that is it useful for unbalanced data. An example should be simple enough to follow easily. Sure you could have done an example with 10 spam and 1 ham but that doesn't help to teach the concepts better at all
The minority class might just be so underrepresented that creating a point cloud around existing data might not be enough.. How do we know we have enough data??