Tutorial 36- Logistic Regression Mutliclass Classification(OneVsRest)- Part 3| Data Science 

Did u get any explanation for it from anywhere? If yes, plz let me know

@@trashantrathore4995 through softmax function

Is there something like that ? please let me know

Absolutely correct but at the end it is also recommend to normalize the predicted probabilities from all three classes using a softmax/linear-normalizer or any other function that satisfies the normalization properties. Also in the video krish may have forgotten to add that we generally normalizes the probs at the end, scikit does so using linear normalizer but softmax can also we used just like we did at the output layer of neural-net in mutliclass problems

@@GauravSharma-ui4yd Hello dear sir, can you please explain what is probability normalization?

@db1 make small projects

Logistic regression is a linear model fitting a curve or sigmoid so that the value stays positive and between 0 and 1

just for explanation purpose I guess

what we take for o1, o2, o3 is the value of y not the value of sigmoid function

Its generally a best practise to get the probabilities and setting threshold manually and deciding the output

Exactly my question... but I tried it in python and got 1 for all samples... which is kinda odd

@@a.mo7a check the documentation....it might have divided it with some number

point did you get the answer?

@Epoxy To The World no sir

@Epoxy To The World I'm 23 years old and how may I know that what path you are following???

@Epoxy To The World yes or just go with the flow and try to understand the maths behind each and every algorithm.

This is what I understood after doing some research on this question : Let assume, we have 3 binary classification models, each one has been trained to output whether the input belongs to (1,2,3 respectively) or not. Assume we have a query or test point as input. We need to return the labels for this input. Since, it is a multiclass problem, the input could have 3 labels. Assume we set our threshold to 0.6 . Now, we pass the input to all these 3 models. Assume we got these probabilities from each of the 3 models as follows : 0.52,0.7,0.8 . Since, we made our threshold to 0.6, the input doesn't belong to the first because 0.52 is less than our threshold 0.6 . We now output the labels for the query as 2 and 3 since their probabilities are greater than the threshold. I'm not sure about my answer but this is what I got after googling it.

You can easily do so in neural net by applying sigmoid activation at the output layer. But the probs across the classes are not normalized hence you can add a softmax layer after it. But first adding sigmoid and then softmax doesn't makes sense hence we directly apply softmax at the end. Also in this case as well we are recommended to normalize the probs across the classes.