ppt课件-designreviewofmpsjapanfid109182-utarlington.pptVIP

Example SystemText Reader Given data for a classification problem with N’ = 90 candidate features, there are 9.344 x 1017 subsets of size N=17 and a total of 290 = 1.2379 x 1027 subsets Feature Selection Scanning Methods: Feature Goodness vs. Brute Force Suppose that available features are: z1 = x + n, z2 = x + n, z3 = y + 2n, and z4 = n where x and y are useful and n is noise. FG: Nested subsets {z1}, {z1, z2 }, {z1, z2 , z3 } BF: Optimal subsets {z2}, {z1, z4 }, {z1, z3 , z4 } since x = z1 - z4 and y = z3 - 2?z4 An optimal subset may include features that the FG a