The code implements the interpretable elastic prototype algorithms decrisbed in "Elastic Prototype Classication" paper found here. The code will need the IBM cplex optimization package for python to run.
##Data Folder The data is found in the "data" folder. The data is nutrition information for different food groups. The algorithm picks prototypical examples that are the best representatives of each food cluster.
Training data - data\nutrition_folds\fold1\train\fold_1_train.csv
Testing data - data\nutrition_folds\fold1\test\fold_1_test.csv
##Code
For the ellipse alternating method in the paper -> use ellipse_alternating.py
For the rectangular alternating method in the paper-> use rectangular_alternating.py
For the rectangules hierarchial method in the paper-> use rectangular.py
##Sample commands for running
###ellipse alternating
#####template of command
python <file_name.py> -d <data_directory> -o <output_file_name> -b <number_of_balls_to_start_with> -f <fold_number>
#####example command
python ellipse_alternating.py -d nutrition_folds -o dec1_ellipse_only_seed230_3 -b 5 -f 1
*note the data is provided for 1 fold.
###rectangular alternating
#####template of command
python <file_name.py> -d <data_directory> -o <output_file_name> -b <number_of_balls_to_start_with> -f <fold_number>
#####example command
python rectangular_alternating.py -d nutrition_folds -o dec1_rectangular_alt_seed230_6 -b 5 -f 1
*note the data is provided for 1 fold.
###rectangules hierarchial
#####template of command
python <file_name.py> -d <data_directory> -o <output_file_name> -b <number_of_balls_to_start_with> -f <fold_number>
#####example command
python rectangular.py -d nutrition_folds -o dec1_rectangular_only_seed230_6 -b 5 -f 1
*note the data is provided for 1 fold.
##Output generation
When the program runs it will generate relevant .png files and .pdf files that shows the prototypes/subspaces in the "elastic_prototype/" directory. The .png files will be 2d or 3d plots of the chosen subspaces. The .pdf file will contain the information in the .png files and in addition would also contain more details about the chosen prototypes and classifications.
Some of the output images are given below
