Skip to content

Latest commit

 

History

History
94 lines (63 loc) · 2.65 KB

demo_optimization.md

File metadata and controls

94 lines (63 loc) · 2.65 KB
Raw

Usage of optimization and runOptimization

Import all functions in runOptimization

  • runOptimization provide the all the functions of optimization and interface to run optimization
from runOptimization import *

Initialization

Step 1: Specify a input data stored in csv file
fn = "GermanCredit_Age35.csv"
Step 2: Specify the target attribute to rank on
  • If specify target attribute as col(fn)-1, it will generate a score for each user by summing all attributes with equally weight to rank on. col(fn) is the number of attributes in the input data.
  • Target attribute can be value from [0,col(fn)-1] while col(fn)-1 represents using weighted summation score as target attribute
target_att = 6
Step 3: Specify which value of sentitive attribute represents the protected group
sensi_att = 1
Step 4: Specify size of K
  • K represents size of intermediate layer in optimization neural network. Higher K represents more accurate prediction. Also takes more time to converge.
opt_k = 4
Step 5: Choose the accuracy measure will be used in the optimization process
  • Choose from ["scoreDiff", "positionDiff", "kendallDis", "spearmanDis", "pearsonDis"].
  • scoreDiff represents the score difference between two rankings.
  • positionDiff represents the rank position difference between two rankings.
  • kendallDis represents the kendall distance between two rankings.
  • spearmanDis represents spearman correlation between two rankings. Only used the correlation and ignore the p-value in this case.
  • pearsonDis represents pearson correlation between two rankings. Only used the correlation and ignore the p-value in this case.
acc_measure = "scoreDiff"
Step 6: Set the cut point at where to compute the fairness measure
cut_point = 10
Step 7: Specify the file to output optimization results
out_fn = "testOptimization.csv"

Test optimization process using above setting

main(fn, target_att, sensi_att, opt_k, acc_measure, cut_point, out_fn)
print "Finished optimization"
print "OPT result stores in "+ out_fn
Finished reading csv!
Finished fairness normalizer calculation!
Starting optimization @  4 ACCM  scoreDiff  time:  1485470812.1
(250, 1.0368499278231544)
(500, 0.54193378842745721)
(750, 0.42883044927988201)
(1000, 0.21465836597579119)
(1250, 0.20320949600353819)
(1500, 0.20011416240307803)
(1750, 0.19853254631944087)
Ending optimization @  4 ACCM  scoreDiff  time:  1485470993.81
Finished optimization
OPT result stores in testOptimization.csv