In the classification assignment where the problem statement is about admission to university, how do I initialize the parameters w and b ?

Suppose I keep the num_iters constant at 10000. Then with the initial estimate of b=0, I get a pretty bad decision boundary at the end of num_iters. If I change the value of b to a negative integer, the decision boundary starts to improve. However, if I increase num_iters by around 10 times, then even with initial b = 0, the plot starts getting better and better.

My question is, how do I initialize the parameters so that I get fairly close to the minima of cost function, while keeping num_iters fixed ?

For logistic regression, it is commonto init the w and b values to zeros.

1 Like

You may need to adjust the iterations and learning rate to get good results.

Okay. So I would have to try different values to get an idea of the best one. Thanks !