code files provided:
paper_data_analysis.py: runs all analysis
paper_data_visualization.py: produces all plots
input data format (as specified on Panelot.org):
For each instance, should have the following data:
categories.csv - contains all quotas on all features and their values
respondents.csv - contains all features and values of all respondents
Instructions: put each instance's files in a separate folder, each titled <instance shortname>_k (k = number of participants being put on the panel)
To reproduce Figure 1:
Step 1. run paper_data_analysis.py with the following parameter settings:
# # # # # # # # # # # # # # PARAMETERS # # # # # # # # # # # # # # # # #
# number of panels desired in the lottery
M = 1000
# which instances to analyze
instances = ['sf_a_35', 'sf_b_20', 'sf_c_44', 'sf_d_40', 'sf_e_110', 'cca_75', 'hd_30', 'mass_24','nexus_170','obf_30','newd_40']
# which objective you want to optimize
LEXIMIN = 0
MAXIMIN = 1
NASH = 0
# flags for which types of lotteries you want to compute
OPT = 1 # computes unconstrained optimal distribution - need to run before any others
ILP = 1 # computes both optimal unconstrained and near-optimal unconstrained, wrt to fairness notion specified below
BECK_FIALA = 1 # computes uniform rounded from OPT via beck-fiala (must run OPT first)
RANDOMIZED = 1 # computes uniform rounded from OPT via randomized rounding (must run OPT first)
RANDOMIZED_REPLICATES = 1000 # runs randomized a bunch of times -> report avg and stdev of loss
ILP_MINIMIAX_CHANGE = 0 # takes input distribution specified by fairness objectives and computes minimum change in anyone's probability
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
Step 2. run paper_data_visualization.py with the following parameter settings:
# # # # # # # # # # # # # # PARAMETERS # # # # # # # # # # # # # # # # #
# number of panels desired in the lottery
M = 1000
# instances you want to run plots for
instances = ['sf_a_35', 'sf_b_20', 'sf_c_44', 'sf_d_40', 'sf_e_110', 'cca_75', 'hd_30', 'mass_24','nexus_170','obf_30','newd_40']
instance_names_dict = {'sf_a_35':'sf(a)', 'sf_b_20': 'sf(b)', 'sf_c_44':'sf(c)', 'sf_d_40':'sf(d)', 'sf_e_110':'sf(e)', 'cca_75':'cca', 'hd_30':'hd', 'mass_24':'mass','nexus_170':'nexus','obf_30':'obf','newd_40':'ndem'}
# objectives (can only run one at a time)
LEXIMIN = 0
MAXIMIN = 1
NASH = 0
# which rounding algorithms to analyze
ILP = 1
ILP_MINIMAX_CHANGE = 0
BECK_FIALA = 1
RANDOMIZED = 1
RANDOMIZED_REPLICATES = 1000
THEORY = 1
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
To reproduce Nash Welfare analog of Figure 1 (in Appendix E.3):
Step 1. run paper_data_analysis.py with the following parameter settings:
# # # # # # # # # # # # # # PARAMETERS # # # # # # # # # # # # # # # # #
# number of panels desired in the lottery
M = 1000
# which instances to analyze
instances = ['sf_a_35', 'sf_b_20', 'sf_c_44', 'sf_d_40', 'sf_e_110', 'cca_75', 'hd_30', 'mass_24','nexus_170','obf_30','newd_40']
# which objective you want to optimize
LEXIMIN = 0
MAXIMIN = 0
NASH = 1
# flags for which types of lotteries you want to compute
OPT = 1 # computes unconstrained optimal distribution - need to run before any others
ILP = 1 # computes both optimal unconstrained and near-optimal unconstrained, wrt to fairness notion specified below
BECK_FIALA = 1 # computes uniform rounded from OPT via beck-fiala (must run OPT first)
RANDOMIZED = 1 # computes uniform rounded from OPT via randomized rounding (must run OPT first)
RANDOMIZED_REPLICATES = 1000 # runs randomized a bunch of times -> report avg and stdev of loss
ILP_MINIMIAX_CHANGE = 0 # takes input distribution specified by fairness objectives and computes minimum change in anyone's probability
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
Step 2. run paper_data_visualization.py with the following parameter settings:
# # # # # # # # # # # # # # PARAMETERS # # # # # # # # # # # # # # # # #
# number of panels desired in the lottery
M = 1000
# instances you want to run plots for
instances = ['sf_a_35', 'sf_b_20', 'sf_c_44', 'sf_d_40', 'sf_e_110', 'cca_75', 'hd_30', 'mass_24','nexus_170','obf_30','newd_40']
instance_names_dict = {'sf_a_35':'sf(a)', 'sf_b_20': 'sf(b)', 'sf_c_44':'sf(c)', 'sf_d_40':'sf(d)', 'sf_e_110':'sf(e)', 'cca_75':'cca', 'hd_30':'hd', 'mass_24':'mass','nexus_170':'nexus','obf_30':'obf','newd_40':'ndem'}
# objectives (can only run one at a time)
LEXIMIN = 0
MAXIMIN = 0
NASH = 1
# which rounding algorithms to analyze
ILP = 1
ILP_MINIMAX_CHANGE = 0
BECK_FIALA = 1
RANDOMIZED = 1
RANDOMIZED_REPLICATES = 1000
THEORY = 1
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
To reproduce Figure 2 plus all analogous figures in Appendix E.4:
Step 1. run paper_data_analysis.py with the following parameter settings:
# # # # # # # # # # # # # # PARAMETERS # # # # # # # # # # # # # # # # #
# number of panels desired in the lottery
M = 1000
# which instances to analyze
instances = ['sf_a_35', 'sf_b_20', 'sf_c_44', 'sf_d_40', 'sf_e_110', 'cca_75', 'hd_30', 'mass_24','nexus_170','obf_30','newd_40']
# which objective you want to optimize
LEXIMIN = 1
MAXIMIN = 0
NASH = 0
# flags for which types of lotteries you want to compute
OPT = 1 # computes unconstrained optimal distribution - need to run before any others
ILP = 0 # computes both optimal unconstrained and near-optimal unconstrained, wrt to fairness notion specified below
BECK_FIALA = 1 # computes uniform rounded from OPT via beck-fiala (must run OPT first)
RANDOMIZED = 1 # computes uniform rounded from OPT via randomized rounding (must run OPT first)
RANDOMIZED_REPLICATES = 1000 # runs randomized a bunch of times -> report avg and stdev of loss
ILP_MINIMIAX_CHANGE = 1 # takes input distribution specified by fairness objectives and computes minimum change in anyone's probability
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
Step 2. run paper_data_visualization.py with the following parameter settings
# # # # # # # # # # # # # # PARAMETERS # # # # # # # # # # # # # # # # #
# number of panels desired in the lottery
M = 1000
# instances you want to run plots for
instances = ['sf_a_35', 'sf_b_20', 'sf_c_44', 'sf_d_40', 'sf_e_110', 'cca_75', 'hd_30', 'mass_24','nexus_170','obf_30','newd_40']
instance_names_dict = {'sf_a_35':'sf(a)', 'sf_b_20': 'sf(b)', 'sf_c_44':'sf(c)', 'sf_d_40':'sf(d)', 'sf_e_110':'sf(e)', 'cca_75':'cca', 'hd_30':'hd', 'mass_24':'mass','nexus_170':'nexus','obf_30':'obf','newd_40':'ndem'}
# objectives (can only run one at a time)
LEXIMIN = 1
MAXIMIN = 0
NASH = 0
# which rounding algorithms to analyze
ILP = 0
ILP_MINIMAX_CHANGE = 1
BECK_FIALA = 1
RANDOMIZED = 1
RANDOMIZED_REPLICATES = 1000
THEORY = 1
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # #