# Regression Iterations

If the goal of gradient descent is to find global minimum for cost, why is that we are picking up the value at the end of iterations? What if one of the earlier iteration has a minimum value?
Below is what is found as the last value of b and w at cost 6.7560
alpha = 0.01 b=100.011567727362 w=199.99285075131766
cost 6.745014662580395e-06

Where is one of the earlier iterations, I found the below values
alpha = 0.01 b=100.08531003189917 w=199.947275500705
cost 0.00036684872861835616

I know the b and w values are almost the same, but theoritcally, they are not exactly minimum. What am I missing here?

For each batch gradient data batch:

Because the cost function is convex, the gradient descent process will always move toward lower cost values on every iteration.

If the cost ever increases on the next iteration, it’s a signal that the learning rate is too large and the system isn’t stable.

Thank you. I made a silly mistake. I missed to see the negative exponential at the end. ( 9.028234203503224e-05.)