ML::03

Gradient Descent
----------------

Goal: Estimate the parameters in the hypothesis function.

The gradient descent algorithm: Iterative algorithm to find theta_0, theta_1 so that J(theta_0,theta_1) has a local minimum.

repeat until convergence {
	theta_j := theta_j - ɑ *   d            J(theta_0, theta_1))
                               ---------
                               d theta_j
}
                                 ------- derivative term ------------

j=0,1; parallel assignment within each iteration; ɑ is called "learning rate"; derivative term (Ableitung nach theta_j)

ɑ to small: algorithm slow
ɑ to large: fail to converge or even diverge