This homework tries two search methods (sequential forward search and exhaustive search) for feature selection in pattern recognition. Before you start, you need to download the following files:

• abalone.dat: the abalone data set
• loadaba.m: script for loading the data set (in which we only use 100 out of 4771 data tuples)
• exhpick3.m: template script for exhaustive search of 3 input variables (or features)
• seqpick.m: template script for sequential forward search for up to 4 input variables (or features)
• knnr.m: k-nearest neighbor rule
• countele.m: element counting, used in knnr.m
• normal.m: column-wise normalization
• louerror.m: leave-one-out error
• vecdist.m: distances between two sets of vectors.

1. Modify lines 24 to 81 of exhpick3.m according to exhaustive search of 3 input variables (or features). You should be able to generate a plot of leave-one-out (LOU) error rates with respective to selected featuers, as shown below. (Note that once you have computed the correct "index" and "misclassify", the program will sort them for you.)
   
2. Modify lines 22 to 49 of seqpick.m according to sequential forward search of up to 4 input variables (or features). You should be able to generate a plot of leave-one-out (LOU) error rates with respective to selected featuers, as shown below. (No sorting is necessary in this case.)
   

   Note that both exhpick3.m and seqpick.m are already executable under MATLAB; they can generate correct plots (with wrong data) without modifications. Your task is to compute the correct LOU error rates using exhaustive and sequential forward searches. (Since we don't have too much time for discussions. If you have any questions, please contact me directly.)

   Please turn in the following items:

   • The modified exhpick3.m and seqpick.m in a floppy disk.
   • Hardcopies of the plots generated by the scripts.