Logistic Regression-2 - Beyond Whyy

In the last section, we saw that linear regression hypothesis function needed to be modified inorder to be used for logistic regression problems. Now the obvious question is whether the same cost function can be used here also or is some modification necessary required. The reason that cost function worked for linear regression was because the hypothesis was linear and hence the cost function was convex shaped with a single global minimum. In case of logistic regression, the hypothesis function is no longer linear because of the sigmoid function and using the same definition for cost function would yield a function as shown below. It will be an impossible task to optimize the parameters theta when using such a cost function with so many local minimums.