using OneVsAll classification algorithm to predict handwritten digits
working of one vs all classification
to understand the working of one-vs-all classification we need to understand logistic regression
what is logistic regression ?
it is a learning algorithm used to predict values between 0 and 1 i.e logical output hence the name logistic regression.
example : predicting whether a person has a particular disease or not predicting whether a car sales value will cross a particular value or not
logistic regression can also be used to predict something that has multi-class output example : classification of digits using its images as input finding out what company a person gets placed to given his skills and qualification etc.
working of logisitic regression .
before understanding logistic regression we need to understand something called sigmoid function.
the image shown here is a sigmoid function . basically it converts any number between (0,1) , this function is necesary as the output we predict should be logical so anything >= 0.5 will be predicted as 1 else 0.
for more detailed explanation of logistic regression follow the link -> https://towardsdatascience.com/logistic-regression-detailed-overview-46c4da4303bc
OneVsAll
one-vs-all is the same as logistic regression except that we repeat the algorithm multiple times
the number of times we repeat the algorithm depends on no of multi-classes present
here we can see the way we have trained our data set using repeated logistic regression . in the above image we have three classes , we need to segregate our data to any of these one class .
working of one-vs-all
algorithm :
for i = 1: no.of.classes
class A = class i ; class B = rest of classes ; perform logistic regression on class A and class B and update the parameters ;
endfor
we use the above algorithm to perform multi-class prediction
for more detailed explanation of one-vs-all follow the link -> https://towardsdatascience.com/multi-class-classification-one-vs-all-one-vs-one-94daed32a87b