RSM318H1 Chapter Notes - Chapter 3: Search Algorithm, Gradient Descent, Categorical Variable
Document Summary
Target y is being predicted from single feature x. Plot y as function of a and bj where in m+1 dimensions. Gradient descent algorithm: iterative search routine for finding minimum. This function is a valley and goal is to find bottom of valley. Use calculus to determine path of steepest descent down the valley. Is preferable when there a very large number of features that are computationally time consuming. Direction a and bj need to be changed to descend as quickly as possible. Linear regressions can produce a number of statistics: Can use products and powers: i. ii. iii. iv. v. vi. Take step down the path of steepest descent. R-squared t-statistics had no explanatory power at all. Is coefficient of correlation squared if only one feature. Measures proportion of variance in target that is explained by the features. P-value: probability of obtaining a t-statistic as large as the one observed if the parameter.
