CSC591-Group9-Project

How to run

• Install python atleast 3.9
• Install requirements: pip install -r requirements.txt
• Run cd src
• Run python main.py to generate tables for the auto2.csv file
• Run python main.py -T -a <algo_name> -f <file_name> to run a praticular algorithm with an input file.
• Run python main.py --help to view possible configuration values

project: multi-goal semi-supervised algorithms
(c) Group 9
  
USAGE: python3 main.py [OPTIONS] [-g ACTIONS]
  
OPTIONS:
  -b  --bins        initial number of bins           = 16
  -c  --cliff       cliff's delta threshold          = .147
  -d  --D           different is over sd*d           = .35
  -F  --Far         distance to distant              = .95
  -h  --help        show help                        = false
  -H  --Halves      search space for clustering      = 512
  -I  --IMin        size of smallest cluster         = .5
  -M  --Max         numbers                          = 512
  -p  --P           dist coefficient                 = 2
  -R  --Rest        how many of rest to sample       = 10
  -r  --reuse       child splits reuse a parent pole = true
  -x  --Bootstrap   number of samples to bootstrap   = 512    
  -o  --Conf        confidence interval              = 0.05
  -f  --file        file to generate table of        = ../data/auto2.csv
  -n  --Niter       number of iterations to run      = 20
  -w  --wColor      output with color                = true
  -T  --test        test particular algorithm        = false
  -a  --algo        name of the algorithm            = sway

Structure

• Data files are located at data
• Output files are located at out and are named using the csv file name
• Python files are located at src
• A script is used to generate all output. It iterates over all data files, runs them through our program, and concatenates output to the output directory
  • The output and how long the process takes is saved to to out/data_file.out
• PDF report can be seen HERE or through the overleaf link

Description of algorithms

main.py

main.py compares each algorithm over 20 iterations.

To produce the top table:

• For the given number of iterations:
  • run data.read to get the "all" result
  • run data.sway to get the "sway" result
  • run the explain algorithm, from HW6, to get the "xpln" result
  • run the betters algorithm, from previous homeworks, to get the "top" result
  • saves all of these data's into a table
• after all iterations complete, for each algorithm:
  • combine the stats. This is done by summing up each mean for each column, and dividing by the number of iterations. It also highlights, in green, the best algorithm per y value
  • produces a table that performs bootstrap and cliffsdelta to determine if each combination of algorithms are considered equal

Example: auto2.csv, which is outputted to auto2.out

        CityMPG+    HighwayMPG+    Weight-    Class-    Avg evals
----  ----------  -------------  ---------  --------  -----------
all           21             28       3040      17.7            0
sway       29.55          34.65    2175.75     9.265            5
xpln       29.55           34.2     2258.5     9.635            5
top           33             40     2052.5       8.9           93

              CityMPG+    HighwayMPG+    Weight-    Class-
------------  ----------  -------------  ---------  --------
all to all    =           =              =          =
all to sway   ≠           ≠              ≠          ≠
sway to xpln  ≠           ≠              ≠          ≠
sway to top   ≠           ≠              ≠          ≠

real	0m13.373s
user	0m0.000s
sys	0m0.031s

To produce the bottom table:

• inside the same for loop as the main loop above::
  • after getting all of the results
  • for each comparison base,diff (all to all, all to sway, sway to xpln, sway to top)

    • originally "set" the comparison result to True, or equals, for each y column

    • for each y column, check if the base data is statistically different than the diff data

      • if it is not equal, set the comparison result for that y value to false

sway2

See PCA-alpha