def find_edge_existence_matchings(matchings_dir, d, a, name,
verbose=0,
cycle_cap=3,
chain_cap_list=[3],
edge_success_prob_list = [0.4],
gamma_list=[],
protection_levels=[0.01],
N_trials=10,
subtour_dir = None,
remove_subtours=False,
cycle_file=None,
run_experiment=False,
outfile=None):
timelimit = None
if len(edge_success_prob_list) > 0:
prob = True
use_gamma = False
par_list = edge_success_prob_list
elif len(gamma_list) > 0:
use_gamma = True
prob = False
par_list = gamma_list
if ( len(edge_success_prob_list) > 0 and len(gamma_list) > 0):
raise Warning("cannot specify both edge success prob list and gamma list")
for chain_cap in chain_cap_list:
for par in par_list:
if prob:
edge_fail_func = lambda e: np.random.rand() > par
elif use_gamma:
gamma = par
type = 'nr'
nr_file = os.path.join(matchings_dir,build_filename(name, type, chain_cap))
cfg = kidney_ip.OptConfig(d, a, cycle_cap, chain_cap, verbose=(verbose >= 2),
timelimit=timelimit, edge_success_prob=1, cycle_file=cycle_file)
if os.path.exists(nr_file):
if run_experiment:
sol = kidney_ip.OptSolution.from_file(cfg, nr_file)
nonrob_score = sum(e.score for e in sol.matching_edges)
else:
sol = kidney_ip.optimise_picef(cfg)
sol.save_to_file(nr_file)
nonrob_score = sum(e.score for e in sol.matching_edges)
if verbose >= 1:
print "protection level epsilon = %f" % p
print "non-robust solution:"
print sol.display()
if prob:
type = 'fa'
fa_file = os.path.join(matchings_dir,build_filename(name, type, chain_cap,edge_success_prob=par))
cfg_failaware = kidney_ip.OptConfig(d, a, cycle_cap, chain_cap, verbose=(verbose>=2),
timelimit=timelimit, edge_success_prob=par,cycle_file=cycle_file)
if os.path.exists(fa_file):
if run_experiment:
sol_failaware = kidney_ip.OptSolution.from_file(cfg_failaware, fa_file)
failaware_score = sum(e.score for e in sol_failaware.matching_edges)
else:
pass
else:
sol_failaware = kidney_ip.optimise_picef(cfg_failaware)
sol_failaware.save_to_file(fa_file)
failaware_score = sol_failaware.get_total_score()
if verbose >= 1:
print "failure-aware solution:"
print sol_failaware.display()
if run_experiment and (sol.total_score == 0):
if verbose:
print "optimal matching has zero weight for exchange: %s" % name
with open(outfile, 'a') as csvfile:
csvfile.write("{},{},{},{},{},{},{},{},{},{},{},{},{}\n".format(
name,
0, # i
cycle_cap,
chain_cap,
par,
p,
0, # sol_rob.gamma,
0, # sol_rob.optimistic_score,
0, # failaware_nominal_score,
0, # sol.total_score,
0, # realized_robust_score,
0, # realized_failaware_score,
0 ))# realized_nonrob_score))
else:
max_num_cycles = 0
for p in protection_levels:
type = 'ex'
ex_file = os.path.join(matchings_dir, build_filename(name, type, chain_cap,
edge_success_prob=par,
protection_level=p))
cfg_rob = kidney_ip.OptConfig(d, a, cycle_cap, chain_cap, verbose=(verbose >= 2),
timelimit=timelimit, edge_success_prob=par,
protection_level=p, name=name, subtour_dir=subtour_dir,
cycle_file=cycle_file,
remove_subtours=remove_subtours)
if os.path.exists(ex_file):
if run_experiment:
sol_rob = kidney_ip.OptSolution.from_file(cfg_rob, ex_file)
rob_optimistic_score = sum(e.score for e in sol_rob.matching_edges)
gamma = sol_rob.gamma
else:
pass
else:
sol_rob = kidney_ip.solve_edge_existence_uncertainty(cfg_rob, max_num_cycles=max_num_cycles)
sol_rob.save_to_file(ex_file)
rob_optimistic_score = sum(e.score for e in sol_rob.matching_edges)
gamma = sol_rob.gamma
max_num_cycles = sol_rob.max_num_cycles # don't recalculate the maximum number of cycles...
if verbose:
print "robust solution:"
print sol_rob.display()
if run_experiment:
# get the robust and non-robust scores for N_trials realizations
with open(outfile, 'a') as csvfile:
for i in range(N_trials):
# add edge failures (these don't need to be replicated, they are Bernouli random)
for e in d.es:
e.fail = edge_fail_func(e)
for n in a:
for e in n.edges:
e.fail = edge_fail_func(e)
if verbose >= 2:
print "MATCHINGS AFTER FAILURE"
print "non-robust solution:"
print sol.display()
print "failure-aware solution:"
print sol_failaware.display()
print "robust solution:"
print sol_rob.display()
realized_failaware_score = sol_failaware.score_after_edge_failure(a)
realized_robust_score = sol_rob.score_after_edge_failure(a)
realized_nonrob_score = sol.score_after_edge_failure(a)
csvfile.write("{},{},{},{},{},{},{},{},{},{},{},{},{}\n".format(
name,
i,
cycle_cap,
chain_cap,
par,
p,
gamma,
rob_optimistic_score,
failaware_score,
nonrob_score,
realized_robust_score,
realized_failaware_score,
realized_nonrob_score))
def find_edge_existence_matchings_const_gamma(matchings_dir, d, a,
name,verbose=0,
cycle_cap=3,
chain_cap_list=[3],
gamma_list=[1],
N_trials=10,
run_experiment=False,
outfile=None):
timelimit = None
for chain_cap in chain_cap_list:
empty_matching = False
for gamma in sorted(gamma_list):
type = 'nr'
nr_file = os.path.join(matchings_dir, build_filename(name, type, chain_cap))
cfg = kidney_ip.OptConfig(d, a, cycle_cap, chain_cap, verbose=(verbose >= 2),
timelimit=timelimit, edge_success_prob=1,gamma=0)
if os.path.exists(nr_file):
if run_experiment:
sol = kidney_ip.OptSolution.from_file(cfg, nr_file)
nonrob_score = sum(e.score for e in sol.matching_edges)
else:
sol = kidney_ip.optimise_pctsp(cfg) # THIS USED TO USE PICEF
sol.save_to_file(nr_file)
nonrob_score = sum(e.score for e in sol.matching_edges)
if verbose >= 1:
print "non-robust solution:"
print sol.display()
type = 'gamma'
gam_file = os.path.join(matchings_dir, build_filename(name, type, chain_cap,gamma=gamma))
cfg_rob = kidney_ip.OptConfig(d, a, cycle_cap, chain_cap, verbose=(verbose >= 2),
timelimit=timelimit,
name=name,
gamma=gamma)
if os.path.exists(gam_file):
if run_experiment:
sol_rob = kidney_ip.OptSolution.from_file(cfg_rob, gam_file)
rob_optimistic_score = sum(e.score for e in sol_rob.matching_edges)
else:
pass
elif empty_matching:
sol_rob.gamma = gamma # use the previous matching, but change gamma...
else:
sol_rob = kidney_ip.optimize_robust_pctsp(cfg_rob)
sol_rob.save_to_file(gam_file)
rob_optimistic_score = sum(e.score for e in sol_rob.matching_edges)
if rob_optimistic_score == 0:
empty_matching = True
if verbose:
print "robust solution (gamma = %d):" % gamma
print sol_rob.display()
if run_experiment:
# get the robust and non-robust scores for N_trials realizations
with open(outfile, 'a') as csvfile:
for i in range(N_trials):
# scores after a set number of failures
realized_robust_score = sol_rob.score_after_num_failures(a,gamma,seed=i)
realized_nonrob_score = sol.score_after_num_failures(a,gamma,seed=i)
csvfile.write("{},{},{},{},{},{},{},{},{}\n".format(
name,
i,
cycle_cap,
chain_cap,
gamma,
rob_optimistic_score,
nonrob_score,
realized_robust_score,
realized_nonrob_score))
def find_edge_weight_matchings(matchings_dir, d, a, name,
verbose=0,
cycle_cap=3,
chain_cap_list=[3],
dist='uniform',
protection_levels=[0.01],
alpha_list=[0.5],
cycle_file=None,
run_experiment=False,
N_trials=None,
outfile=None):
timelimit = None
edge_success_prob = 1
for alpha in alpha_list:
# weight distribution:
wt_string = dist + re.sub('\\.', '', '{:g}'.format(alpha))
if dist == 'uniform':
# edge weights are uniformly distributed on [w(1-a),w(1+a)], where a is between 0 and 1
low_frac = 1.0 - alpha
high_frac = 1.0 + alpha
discount_func = lambda edge: edge.score * alpha
realized_weight_func = lambda edge: np.random.uniform(low=edge.score*low_frac, high=edge.score*high_frac)
elif dist == 'bimodal':
# edge weights are either low or high
low_frac = 1.0 - alpha
high_frac = 1.0 + alpha
discount_func = lambda edge: edge.score * alpha
realized_weight_func = lambda edge: np.random.choice([edge.score * low_frac, edge.score * high_frac])
elif dist == 'flat':
# save state for re-generating edge states
edge_assign_seed = random.randint(0, 2**31)
# assign some edges to be constant (0.5), and others to be either 0 or 1
# sets edge discount and weight_type
assign_flat_edges(edge_assign_seed,alpha,d,a)
# weight_type 0 = constant
# weight_type 1 = bimodal
# realized edge weights will be added later
realized_weight_func = lambda edge: 0.5 if edge.weight_type==0 else np.random.choice([0.0,1.0])
for chain_cap in chain_cap_list:
type = 'nr'
nr_file = os.path.join(matchings_dir,build_filename(name, type, chain_cap))
cfg = kidney_ip.OptConfig(d, a, cycle_cap, chain_cap, verbose=(verbose >= 2),
timelimit=timelimit, edge_success_prob=edge_success_prob, cycle_file=cycle_file)
# # if os.path.exists(nr_file):
# # if run_experiment:
# # sol = kidney_ip.OptSolution.from_file(cfg, nr_file)
# # nonrob_score = sum(e.score for e in sol.matching_edges)
# # else:
# # pass
# else:
sol = kidney_ip.optimise_picef(cfg)
sol.save_to_file(nr_file)
nonrob_score = sum(e.score for e in sol.matching_edges)
max_card = 0
for p in protection_levels:
type = 'wt'
wt_file = os.path.join(matchings_dir, build_filename(name, type, chain_cap, wt_string=wt_string, edge_success_prob=None, protection_level=p))
cfg_rob = kidney_ip.OptConfig(d, a, cycle_cap, chain_cap, verbose=(verbose>=2),
timelimit=timelimit, edge_success_prob=edge_success_prob,
protection_level=p,cycle_file=cycle_file, edge_assign_seed=edge_assign_seed)
if os.path.exists(wt_file):
if run_experiment:
sol_rob = kidney_ip.OptSolution.from_file(cfg_rob, wt_file)
rob_optimistic_score = sum(e.score for e in sol_rob.matching_edges)
else:
pass
else:
sol_rob = kidney_ip.solve_edge_weight_uncertainty(cfg_rob, max_card=max_card)
sol_rob.save_to_file(wt_file)
rob_optimistic_score = sum(e.score for e in sol_rob.matching_edges)
max_card = sol_rob.max_card # don't recalculate the maximum cardinality...
if verbose >= 1:
print "protection level epsilon = %f" % p
print "non-robust solution:"
print sol.display()
print "robust solution:"
print sol_rob.display()
if run_experiment:
# assign edges their original weight_type using the seed
assign_flat_edges(sol_rob.edge_assign_seed, alpha, d, a)
with open(outfile, 'a') as csvfile:
for i in range(N_trials):
realized_robust_score = sol_rob.score_with_edge_weight_func(realized_weight_func)
realized_nonrob_score = sol.score_with_edge_weight_func(realized_weight_func)
csvfile.write("{},{},{},{},{},{},{},{},{},{},{}\n".format(
name,
i,
cycle_cap,
chain_cap,
alpha,
p,
sol_rob.gamma,
rob_optimistic_score,
nonrob_score,
realized_robust_score,
realized_nonrob_score))
def weighted_fairness_experiment(beta_list, digraph, altruists, dirname,
outfile, chain_cap_list, edge_prob_list,
frac_edges):
beta_list.sort()
cycle_cap = 3
# chain_cap = 3
verbose = False
timelimit = None
#edge_success_prob = edge_prob
eef_alt_constraints = None
lp_file = None
relax = None
formulation = 'picef'
use_relabelled = False
multi = 1
gap = 0
alpha = 0
min_sensitized = 0
total_score = numpy.zeros(len(beta_list))
fair_score = numpy.zeros(len(beta_list))
num_matched = numpy.zeros(len(beta_list))
sens_matched = numpy.zeros(len(beta_list))
total_num_sensitized = digraph.get_num_sensitized()
total_num_pairs = digraph.get_num_pairs()
num_ndds = len(altruists)
# find maximum number of sensitized patients possible
# copy KPD graph with nonzero weights only for highly sensitized patientss
d_fair = digraph.fair_copy()
a_fair = [a.fair_copy() for a in altruists]
# print("maximum possible fairness: {} (total matched = {})".format(max_sens, fair_sol.num_matched()))
for chain_cap in chain_cap_list:
for edge_success_prob in edge_prob_list:
fair_cfg = kidney_ip.OptConfig(d_fair, a_fair, cycle_cap,
chain_cap, verbose, timelimit,
edge_success_prob,
eef_alt_constraints, lp_file, relax,
multi, gap, min_sensitized)
fair_sol = solve_kep(fair_cfg, formulation, use_relabelled)
max_fair_score = fair_sol.total_score
max_sens_matched = fair_sol.num_sensitized()
# these values should never be used
current_total_score = -1
current_fair_score = -1
current_num_matched = -1
current_sens_matched = -1
for i, beta in enumerate(beta_list):
# if we've already hit the maximum score...
if current_fair_score == max_fair_score:
total_score[i] = current_total_score
fair_score[i] = current_fair_score
num_matched[i] = current_num_matched
sens_matched[i] = current_sens_matched
else:
digraph.augment_weights(beta)
for a in altruists:
a.augment_weights(beta)
cfg = kidney_ip.OptConfig(digraph, altruists, cycle_cap,
chain_cap, verbose, timelimit,
edge_success_prob,
eef_alt_constraints, lp_file,
relax, multi, gap,
min_sensitized)
sol = solve_kep(cfg, formulation, use_relabelled)
digraph.unaugment_weights(beta)
for a in altruists:
a.unaugment_weights(beta)
# get the total scores (unaugment the weights first, and recalculate the score...)
current_total_score = sol.get_score(
digraph, altruists, edge_success_prob)
current_fair_score = sol.get_score(d_fair, a_fair,
edge_success_prob)
current_num_matched = sol.num_matched()
current_sens_matched = sol.num_sensitized()
total_score[i] = current_total_score
fair_score[i] = current_fair_score
num_matched[i] = current_num_matched
sens_matched[i] = current_sens_matched
with open(outfile, 'a') as csvfile:
for i, beta in enumerate(
beta_list): # enumerate(beta_list[0:max_i+1]):
csvfile.write(
"{},{},{},{},{},{},{},{},{},{},{},{},{},{},{},{}\n".
format(dirname, cycle_cap, chain_cap, total_num_pairs,
num_ndds, total_num_sensitized, max_fair_score,
max_sens_matched, beta, alpha, total_score[i],
fair_score[i], num_matched[i], sens_matched[i],
edge_success_prob, frac_edges))
def alpha_lex_experiment(digraph, altruists, dirname, outfile, alpha_list,
chain_cap_list, edge_prob_list, frac_edges):
alpha_list.sort()
cycle_cap = 3
# chain_cap = 4
verbose = False
timelimit = None
#edge_success_prob = edge_prob
eef_alt_constraints = None
lp_file = None
relax = None
formulation = 'picef'
use_relabelled = False
multi = 1
gap = 0
min_sensitized = 0
beta = 0
# find maximum number of sensitized patients possible
# copy KPD graph with nonzero weights only for highly sensitized patientss
d_fair = digraph.fair_copy()
a_fair = [a.fair_copy() for a in altruists]
# graph parameters
total_num_sensitized = digraph.get_num_sensitized()
total_num_pairs = digraph.get_num_pairs()
num_ndds = len(altruists)
with open(outfile, 'a') as csvfile:
for chain_cap in chain_cap_list:
for edge_success_prob in edge_prob_list:
fair_cfg = kidney_ip.OptConfig(d_fair, a_fair, cycle_cap,
chain_cap, verbose, timelimit,
edge_success_prob,
eef_alt_constraints, lp_file,
relax, multi, gap,
min_sensitized)
fair_sol = solve_kep(fair_cfg, formulation, use_relabelled)
max_fair_score = fair_sol.total_score
max_sens_matched = fair_sol.num_sensitized()
# first solve original problem (alpha = 0)
min_fair_score = 0
cfg = kidney_ip.OptConfig(digraph, altruists, cycle_cap,
chain_cap, verbose, timelimit,
edge_success_prob,
eef_alt_constraints, lp_file, relax,
multi, gap, min_fair_score)
sol = solve_kep(cfg, formulation, use_relabelled)
fair_score = sol.get_fair_score(altruists)
num_matched = sol.num_matched()
num_sensitized = sol.num_sensitized()
for alpha in alpha_list:
min_fair_score = alpha * max_fair_score # numpy.linspace(0,1,11) * max_fair_score
# if the previous solution already satisfies this minimum score, just write it again
if fair_score >= min_fair_score:
csvfile.write(
"{},{},{},{},{},{},{},{},{},{},{},{},{},{},{},{}\n"
.format(dirname, cycle_cap, chain_cap,
total_num_pairs, num_ndds,
total_num_sensitized, max_fair_score,
max_sens_matched, beta, alpha,
sol.total_score, fair_score, num_matched,
num_sensitized, edge_success_prob,
frac_edges))
# if the previous solution doesn't meet the alpha criteria, solve again with the new min fair score
else:
cfg = kidney_ip.OptConfig(
digraph, altruists, cycle_cap, chain_cap, verbose,
timelimit, edge_success_prob, eef_alt_constraints,
lp_file, relax, multi, gap, min_fair_score)
sol = solve_kep(cfg, formulation, use_relabelled)
fair_score = sol.get_fair_score(altruists)
num_matched = sol.num_matched()
num_sensitized = sol.num_sensitized()
csvfile.write(
"{},{},{},{},{},{},{},{},{},{},{},{},{},{},{},{}\n"
.format(dirname, cycle_cap, chain_cap,
total_num_pairs, num_ndds,
total_num_sensitized, max_fair_score,
max_sens_matched, beta, alpha,
sol.total_score, fair_score, num_matched,
num_sensitized, edge_success_prob,
frac_edges))
def start():
parser = argparse.ArgumentParser("Solve a kidney-exchange instance")
parser.add_argument("cycle_cap", type=int,
help="The maximum permitted cycle length")
parser.add_argument("chain_cap", type=int,
help="The maximum permitted number of edges in a chain")
parser.add_argument("formulation",
help="The IP formulation (uef, eef, eef_full_red, hpief_prime, hpief_2prime, hpief_prime_full_red, hpief_2prime_full_red, picef, cf)")
parser.add_argument("--use-relabelled", "-r", required=False,
action="store_true",
help="Relabel vertices in descending order of in-deg + out-deg")
parser.add_argument("--eef-alt-constraints", "-e", required=False,
action="store_true",
help="Use slightly-modified EEF constraints (ignored for other formulations)")
parser.add_argument("--timelimit", "-t", required=False, default=None,
type=float,
help="IP solver time limit in seconds (default: no time limit)")
parser.add_argument("--verbose", "-v", required=False,
action="store_true",
help="Log Gurobi output to screen and log file")
parser.add_argument("--edge-success-prob", "-p", required=False,
type=float, default=1.0,
help="Edge success probability, for failure-aware matching. " +
"This can only be used with PICEF and cycle formulation. (default: 1)")
parser.add_argument("--lp-file", "-l", required=False, default=None,
metavar='FILE',
help="Write the IP model to FILE, then exit.")
parser.add_argument("--relax", "-x", required=False,
action='store_true',
help="Solve the LP relaxation.")
parser.add_argument("--multi", "-m", required=False, # added by Duncan
type=int, default=1,
help="Search for multiple solutions. (Specify the maximum number of solutions to search for, default = 1)")
parser.add_argument("--experiment", "-z", required=False, # added by Duncan
action='store_true', default=None,
help="Run experiment")
parser.add_argument("--highly_sensitized", "-H", type=percent, default=0, required=False, # added by Duncan
help="Fraction of highly sensitized patients (on [0-1], default = 0)")
parser.add_argument("--seed", "-s", type=str, default=1, required=False, # added by Duncan
help="Random seed for choosing sensitized vertices")
parser.add_argument("--fairness", "-f", type=str, default=1, required=False, # added by Duncan
help="Output file of fairness data")
parser.add_argument("--all_solutions", "-a", type=str, default=1, required=False, # added by Duncan
help="Output file of all solutionss")
parser.add_argument("--add_weight_type", "-w", type=int, default=0, required=False, # added by Duncan
help="Add random edge weights (0 = constant, 1 = floating point, >1 = number of weight levels")
parser.add_argument("--pool_gap", "-g", type=int, default=0, required=False, # added by Duncan
help="Pool gap (if >0, find suboptimal solutions)")
parser.add_argument("--pareto", "-P", required=False, # added by Duncan
action='store_true', default=None,
help="Calculate Pareto curve")
args = parser.parse_args()
args.formulation = args.formulation.lower()
input_lines = [line for line in sys.stdin if len(line.strip()) > 0]
n_digraph_edges = int(input_lines[0].split()[1])
digraph_lines = input_lines[:n_digraph_edges + 2]
d = kidney_digraph.read_digraph(digraph_lines)
if len(input_lines) > len(digraph_lines):
ndd_lines = input_lines[n_digraph_edges + 2:]
altruists = kidney_ndds.read_ndds(ndd_lines, d)
else:
altruists = []
# add random weights
# args.add_weight_type = 0 # 0 is default, read weights from file (constant, uniform for each edge)
if args.add_weight_type == 1: # random floating point weight between 1 and 2 for each edge
for edge in d.es:
edge.score = random.random() + 1.0
elif args.add_weight_type > 1: # random integer, either 1 or 2
for edge in d.es:
edge.score = random.randint(1, args.add_weight_type)
if args.highly_sensitized > 0:
num_sensitized = int( round(d.n * args.highly_sensitized) )
random.seed(args.seed)
sensitized_pairs = random.sample(range(d.n),num_sensitized)
for i in sensitized_pairs:
d.vs[i].sensitized = True
else:
num_sensitized = 0
sensitized_pairs = []
if args.pareto:
print "formulation : {}".format(args.formulation)
print "using_relabelled : {}".format(args.use_relabelled)
print "cycle_cap : {}".format(args.cycle_cap)
print "chain_cap : {}".format(args.chain_cap)
print "num_pairs : {}".format(d.n)
print "num_ndds : {}".format(len(altruists))
print "max_core_size : {}".format(args.multi)
print "num_sensitized : {}".format(num_sensitized)
print "sensitized_pairs : {}".format(' '.join([str(i) for i in sensitized_pairs]))
# 1) calculate optimal core, and all fairness values within core
min_sens = 0
gap = 0
min_sens = 0
cfg = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, args.multi, gap, min_sens)
core = solve_kep(cfg, args.formulation, args.use_relabelled)
print "optimal_score : {}".format(core.total_score)
core_fairness = [s.num_sensitized() for s in core.solutions]
print "{:10s} {:10s} {:10s}".format('min_sens','num_sens','score')
#for s in core.solutions:
# print "{:10d} {:10d} {:10.3f}".format(0,s.num_sensitized(),s.total_score)
# 2) start with a 0 fairness bound, and increase, solving KEP for optimal solution
# ONLY SEARCH FOR ONE SOLUTION PER FAIRNESS
for min_sensitized in range(num_sensitized):
cfg2 = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, 1, 0, min_sensitized)
sol = solve_kep(cfg2, args.formulation, args.use_relabelled)
if sol.total_score > 0:
print "{:10d} {:10d} {:10.3f}".format(int(min_sensitized), sol.num_sensitized(),sol.total_score)
elif args.experiment:
experiment_num = 1
# run experiment:
# randomly assign a percentage of pairs to be highly sensitized
# use the filename as a seed for python.random
if experiment_num == 1:
# 1) find optimal solution
min_sens=0
cfg = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, args.multi, args.pool_gap, min_sens)
opt_solution = solve_kep(cfg, args.formulation, args.use_relabelled)
# 2) find all solutions within 90% of optimal solution
opt_score = opt_solution.total_score
print "inital opt finished. OPT score = {} (max={})".format(opt_score,args.multi)
pct = 90
low_score = max( opt_score * pct / 100, 1)
gap = opt_score - low_score
print "Now finding all solutions within {}% of OPT. low score = {}".format(pct,low_score)
cfg_2 = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, args.multi, gap)
opt_solution_2 = solve_kep(cfg_2, args.formulation, args.use_relabelled)
opt_solution_2.write_fairness( args.fairness, args.cycle_cap, args.chain_cap, len(altruists) )
#opt_solution_2.write_all_scores(args.all_solutions, args.cycle_cap, args.chain_cap, len(altruists))
# print "new core size : {} (max={})".format(opt_solution.size,args.multi)
# print "score counts ; min_fairness max_fairness"
# fair_by_score = opt_solution.pct_sensitized_by_score()
# for score, counts in opt_solution.score_counter().most_common():
# max_fair = max(fair_by_score[score])
# min_fair = min(fair_by_score[score])
# print"{:5} {:10} ; {:6.2f} {:6.2f}".format(score,counts,min_fair,max_fair)
#
# fairness = opt_solution.pct_sensitized_counter()
# print "--- fairness ---"
# print "max percent sens: {}".format(max(fairness.keys()))
# print "min percent sens: {}".format(min(fairness.keys()))
elif experiment_num == 2:
# 1) find optimal solution without constraining fairness
min_sensitized = 0
cfg = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, args.multi, args.pool_gap, min_sensitized)
opt_solution = solve_kep(cfg, args.formulation, args.use_relabelled)
print "No constraint on fairness. OPT = {}".format(opt_solution.total_score)
print "# sensitized pairs used (total) = {} ({})".format(opt_solution.num_sensitized(),num_sensitized)
print "----------- sensitized verts -----------"
print ' '.join([str(i) for i in sorted(sensitized_pairs)])
print "----------- solution -----------"
print opt_solution.display()
# now force 3 sensitized pairs to be used
min_sensitized = 1
cfg2 = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, args.multi, args.pool_gap, min_sensitized)
opt_solution2 = solve_kep(cfg2, args.formulation, args.use_relabelled)
print "Fairness constrined to {}. OPT = {}".format(min_sensitized, opt_solution2.total_score)
print "# sensitized pairs used (total) = {} ({})".format(opt_solution2.num_sensitized(),num_sensitized)
print "----------- sensitized verts -----------"
print ' '.join([str(i) for i in sorted(sensitized_pairs)])
print "----------- solution -----------"
print opt_solution2.display()
# now force 10 sensitized pairs to be used
min_sensitized = 13
cfg3 = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, args.multi, args.pool_gap, min_sensitized)
opt_solution3 = solve_kep(cfg3, args.formulation, args.use_relabelled)
print "Fairness constrined to {}. OPT = {}".format(min_sensitized, opt_solution3.total_score)
print "# sensitized pairs used (total) = {} ({})".format(opt_solution3.num_sensitized(),num_sensitized)
print "----------- sensitized verts -----------"
print ' '.join([str(i) for i in sorted(sensitized_pairs)])
print "----------- solution -----------"
print opt_solution3.display()
elif experiment_num == 3:
# 1) calculate core (optimal), and fairness values within core
min_sens = 0
cfg = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, args.multi, 0, 0)
core = solve_kep(cfg, args.formulation, args.use_relabelled)
core_fairness = [s.num_sensitized() for s in core.solutions]
print "optimal score: {}".format(core.total_score)
print "fairness values (num sensitized = {}):".format(num_sensitized)
print core_fairness
# 2) start with a 0% fairness bound, and increase as long as KEP is feasible. for each increase in fairness,
# solve KEP and record optimal solution
# ONLY SEARCH FOR ONE SOLUTION
for min_sensitized in range(num_sensitized):
cfg2 = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, 1, 0, min_sensitized)
sol = solve_kep(cfg2, args.formulation, args.use_relabelled)
print "min_sensitized : {} score : {}".format(min_sensitized, sol.total_score)
else:
start_time = time.time()
cfg = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap, args.verbose,
args.timelimit, args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax, args.multi) # added multi -- Duncan
opt_solution = solve_kep(cfg, args.formulation, args.use_relabelled)
time_taken = time.time() - start_time
if args.multi > 1: # added by Duncan
print "formulation: {}".format(args.formulation)
print "formulation_name: {}".format(opt_solution.formulation_name)
print "using_relabelled: {}".format(args.use_relabelled)
print "cycle_cap: {}".format(args.cycle_cap)
print "chain_cap: {}".format(args.chain_cap)
print "total_time: {}".format(time_taken)
print "ip_solve_time: {}".format(opt_solution.ip_model.runtime)
print "# pairs : {}".format(cfg.digraph.n)
print "# NDDs : {}".format(len(cfg.ndds))
print "total_score: {}".format(opt_solution.total_score)
print "max core size: {}".format(args.multi)
print "core size : {}".format(opt_solution.size)
if args.output_core:
opt_solution.write_vertex_participation(args.output_core, args.cycle_cap, args.chain_cap, len(altruists) )
else:
print "formulation: {}".format(args.formulation)
print "formulation_name: {}".format(opt_solution.formulation_name)
print "using_relabelled: {}".format(args.use_relabelled)
print "cycle_cap: {}".format(args.cycle_cap)
print "chain_cap: {}".format(args.chain_cap)
print "edge_success_prob: {}".format(args.edge_success_prob)
print "ip_time_limit: {}".format(args.timelimit)
print "ip_vars: {}".format(opt_solution.ip_model.numVars)
print "ip_constrs: {}".format(opt_solution.ip_model.numConstrs)
print "total_time: {}".format(time_taken)
print "ip_solve_time: {}".format(opt_solution.ip_model.runtime)
print "solver_status: {}".format(opt_solution.ip_model.status)
print "total_score: {}".format(opt_solution.total_score)
opt_solution.display()
def start():
parser = argparse.ArgumentParser("Solve a kidney-exchange instance")
parser.add_argument("cycle_cap",
type=int,
help="The maximum permitted cycle length")
parser.add_argument(
"chain_cap",
type=int,
help="The maximum permitted number of edges in a chain")
parser.add_argument(
"formulation",
help=
"The IP formulation (uef, eef, eef_full_red, hpief_prime, hpief_2prime, hpief_prime_full_red, hpief_2prime_full_red, picef, cf)"
)
parser.add_argument(
"--use-relabelled",
"-r",
required=False,
action="store_true",
help="Relabel vertices in descending order of in-deg + out-deg")
parser.add_argument(
"--eef-alt-constraints",
"-e",
required=False,
action="store_true",
help=
"Use slightly-modified EEF constraints (ignored for other formulations)"
)
parser.add_argument(
"--timelimit",
"-t",
required=False,
default=None,
type=float,
help="IP solver time limit in seconds (default: no time limit)")
parser.add_argument("--verbose",
"-v",
required=False,
action="store_true",
help="Log Gurobi output to screen and log file")
parser.add_argument(
"--edge-success-prob",
"-p",
required=False,
type=float,
default=1.0,
help="Edge success probability, for failure-aware matching. " +
"This can only be used with PICEF and cycle formulation. (default: 1)")
parser.add_argument("--lp-file",
"-l",
required=False,
default=None,
metavar='FILE',
help="Write the IP model to FILE, then exit.")
parser.add_argument("--relax",
"-x",
required=False,
action='store_true',
help="Solve the LP relaxation.")
args = parser.parse_args()
args.formulation = args.formulation.lower()
input_lines = [line for line in sys.stdin if len(line.strip()) > 0]
n_digraph_edges = int(input_lines[0].split()[1])
digraph_lines = input_lines[:n_digraph_edges + 2]
d = kidney_digraph.read_digraph(digraph_lines)
if len(input_lines) > len(digraph_lines):
ndd_lines = input_lines[n_digraph_edges + 2:]
altruists = kidney_ndds.read_ndds(ndd_lines, d)
else:
altruists = []
start_time = time.time()
cfg = kidney_ip.OptConfig(d, altruists, args.cycle_cap, args.chain_cap,
args.verbose, args.timelimit,
args.edge_success_prob, args.eef_alt_constraints,
args.lp_file, args.relax)
opt_solution = solve_kep(cfg, args.formulation, args.use_relabelled)
time_taken = time.time() - start_time
print "formulation: {}".format(args.formulation)
print "formulation_name: {}".format(opt_solution.formulation_name)
print "using_relabelled: {}".format(args.use_relabelled)
print "cycle_cap: {}".format(args.cycle_cap)
print "chain_cap: {}".format(args.chain_cap)
print "edge_success_prob: {}".format(args.edge_success_prob)
print "ip_time_limit: {}".format(args.timelimit)
print "ip_vars: {}".format(opt_solution.ip_model.numVars)
print "ip_constrs: {}".format(opt_solution.ip_model.numConstrs)
print "total_time: {}".format(time_taken)
print "ip_solve_time: {}".format(opt_solution.ip_model.runtime)
print "solver_status: {}".format(opt_solution.ip_model.status)
print "total_score: {}".format(opt_solution.total_score)
opt_solution.display()
