take this example of a small dataset this data set is from a homework problem from https://openlearninglibrary.mit.edu/courses/course-v1:MITx+6.036+1T2019/courseware/Week2/week2_homework/?activate_block_id=block-v1%3AMITx%2B6.036%2B1T2019%2Btype%40sequential%2Bblock%40week2_homework

so here there was a question that instead of initializing weight vector as zeroes what if we initialize to [1000 , -1000] (there is no offset i.e classifiers passing through origin). will the mistakes that the perceptron algorithm makes increases or decreases as it gets converged? Answer : It would significantly increase the number of mistakes.

Intuition that I have: the learning rate parameter affects only the scale of the weight vector, not the direction. so even if we take thetha = [1000, -1000] we cannot determine whether it makes less mistakes than [0,0] or not.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.