Supplementary material for the paper "Theoretical Evaluation of Asymmetric Shapley Values for Root-Cause Analysis"

This repository contains the code we used to run the experiments in our paper.

Information regarding features, groups

Communities and Crime features semantic grouping

Group name	Features
Race	racepctblack, racePctWhite, racePctAsian, racePctHisp
Age	agePct12t21, agePct12t29, agePct16t24, agePct65up
Income	medIncome, pctWWage, pctWFarmSelf, pctWInvInc, pctWSocSec, pctWPubAsst, pctWRetire, medFamInc, perCapInc, NumUnderPov, PctPopUnderPov
Race/Income	whitePerCap, blackPerCap, indianPerCap, AsianPerCap, OtherPerCap, HispPerCap
Education	PctLess9thGrade, PctNotHSGrad, PctBSorMore, PctUnemployed, PctEmploy, PctEmplManu, PctEmplProfServ, PctOccupManu, PctOccupMgmtProf
Family	MalePctDivorce, MalePctNevMarr, FemalePctDiv, TotalPctDiv, PersPerFam, PctFam2Par, PctKids2Par, PctYoungKids2Par, PctTeen2Par, PctWorkMomYoungKids, PctWorkMom, NumKidsBornNeverMar, PctKidsBornNeverMar
Immigration	PctForeignBorn, NumImmig, PctImmigRecent, PctImmigRec5, PctImmigRec8, PctImmigRec10, PctRecentImmig, PctRecImmig5, PctRecImmig8, PctRecImmig10, PctSpeakEnglOnly, PctNotSpeakEnglWell
House	householdsize, PctLargHouseFam, PctLargHouseOccup, PersPerOccupHous, PersPerOwnOccHous, PersPerRentOccHous, PctPersOwnOccup, PctPersDenseHous, PctHousLess3BR, MedNumBR, HousVacant, PctHousOccup, PctHousOwnOcc, PctVacantBoarded, PctVacMore6Mos, MedYrHousBuilt, PctHousNoPhone, PctWOFullPlumb, OwnOccLowQuart, OwnOccMedVal, OwnOccHiQuart, OwnOccQrange, RentLowQ, RentMedian, RentHighQ, RentQrange, MedRent, MedRentPctHousInc, MedOwnCostPctInc, MedOwnCostPctIncNoMtg
Homelessness	NumInShelters, NumStreet
Native	PctBornSameState, PctSameHouse85, PctSameCity85, PctSameState85
Police	LemasSwornFT, LemasSwFTPerPop, LemasSwFTFieldOps, LemasSwFTFieldPerPop, LemasTotalReq, LemasTotReqPerPop, PolicReqPerOffic, PolicPerPop
Race/Police	RacialMatchCommPol, PctPolicWhite, PctPolicBlack, PctPolicHisp, PctPolicAsian, PctPolicMinor, OfficAssgnDrugUnits, NumKindsDrugsSeiz, PolicAveOTWorked, PolicCars, PolicOperBudg, LemasPctPolicOnPatr, LemasGangUnitDeploy, LemasPctOfficDrugUn, PolicBudgPerPop
Land/Pop	population, numbUrban, pctUrban, LandArea, PopDens, PctUsePubTrans

Mobile Telecommunications dataset feature groups

Configuration Management data

Id	Name	# of features	Description
#1	Spectrum	7	Spectrum resources available for a given cell. Includes start frequency of the used band and channel bandwidth.
#2	Antennas, MIMO and modulations	5	Other radio resource like capabilities of the cell including: number of antennas, MIMO capabilities, modulation capabilities.
#3	Dimensioning	33	Configuration settings of the cell resulting from cell dimensioning: range, transmit power, signal strength threshold and hysteresis settings, etc.

Performance Management data

Id	Name	# of features	Description
#4	Cell load	4	Cell load metrics with statistics about the number of connected users and active users.
#5	Neighbor cell load	4	Weighted average of load metrics from neighboring cells (weighted by handover counts).
#6	UE cap. distr.	12	Radio capability distribution of connected users.
#7	TA distr.	3	Timing Advance distribution of UEs within the cell (Timing Advance estimates the distance of the UE from the base station)
#8	Interference	3	Uplink interference distribution
#9	Path loss	3	Uplink path loss distribution (expresses how much the transmitted power is attenuated by the radio channel between the UE and the base station)
#10	Channel quality	4	Rank distribution within the cell

Repo Table of Contents

Directories

nam/

Contains experiments regarding neural networks (NAM). Please refer to the README.md of the subdirectory.

data/

For convenience, we also include the necessary data files in the correct folder structure to run the code. These were downloaded from the UCI repository using the command wget --recursive --no-parent http://archive.ics.uci.edu/ml/machine-learning-databases/00211/ and similar, and also uncompressed when necessary.

output/

Rendered .pdf and .png charts and figures.

Notebooks

example-b-using-absval.ipynb

The incorrect attributions of Example B can also be demonstrated using average absolute values. This notebook contains the numerical simulation.

cacu-features.ipynb

Experiments regarding feature pairs on the Communities and Crime Unnormalized dataset.

cacu-semantic-grops.ipynb

Experiments regarding the semantic grouping of the Communities and Crime Unnormalized dataset.

alldata-features.ipynb

Experiments regarding feature pairs in the four public datasets.

alldata-random-groups.ipynb

Experiments regarding random groupings in the four public datasets.

Environment

We used a python 3.8 environment with the following packages to run the experiments:

• lightgbm=3.3.5
• pandas
• numpy
• tqdm
• matplotlib