def KS(params, eval_func, bins, bmin, h):
idx = bins.index(bmin)
h_hat = [eval_func(params, b) for b in bins[idx:len(h)]]
try:
cdf_h_hat = cdf(h_hat)
except:
traceback.print_exc()
print("func: %s"%eval_func.__name__)
print("params: "+ str(params ))
print(h_hat)
cdf_h = cdf(h[idx:])
return max( [abs(cdf_h[i] - cdf_h_hat[i]) for i in range(len(cdf_h_hat))] )
def __init__(self):
self.agents = {}
self.locations = [(x,y) for x in range(Grid_Size) for y in range(Grid_Size)]
agents_per_race = int((Grid_Size**2)*(1-Empty_Space)/Num_of_Races)
shuffle(self.locations)
loc_idx = 0
minimum_friends = 3
agents = []
for race in range(Num_of_Races):
for i in range(agents_per_race):
new_agent = Agent(race)
potential_friends = agents.copy()
for j in range(minimum_friends):
if potential_friends:
degrees = [degree(agent) for agent in potential_friends]
friend = potential_friends[randomFromCDF(cdf(degrees))]
new_agent.connectTo(friend)
potential_friends.remove(friend)
loc = self.locations[loc_idx]
loc_idx += 1
self.agents[loc] = new_agent
agents.append(new_agent)
# start keeping track of empty locations for unhappy agents to move to
self.empty_locations = []
for i in range(loc_idx, len(self.locations)):
self.empty_locations.append(self.locations[i])
self.steps = 0
self.unhappy_count = 0
def goodness_of_fit(h, bins, bmin, eval_funcs, params, eps, obj_func=KSmod):
'''
eval_funcs : a list of evaluation functions
params : a dictionary of estimated parameter vectors for the functions in eval_funcs
'''
counts = collections.OrderedDict() #count of synthetic data sets with KS_synthetic > KS_est
KS_est = {}
num_data_sets = int((eps**-2)/4)+1
verbose("Generating %d data sets..."%num_data_sets)
idx = bins.index(bmin)
# generate distributions from parameters
distributions = {}
for func in eval_funcs:
distribution = []
for i in range(idx):
distribution.append(0)
for i in range(idx, len(h)):
distribution.append(func(params[func], bins[i]))
print("%s distribution: "%func.__name__)
print(distribution)
distributions[func] = cdf(distribution)
KS_est[func] = KS(params[func], func, bins, bmin, h)
counts[func] = 0
for i in range(num_data_sets):
h_syn = generateData(h, bins, bmin, distributions)
for func in eval_funcs:
params_syn = estimate(h_syn[func], bins, bmin, func, obj_func=obj_func)
KS_syn = KS(params_syn, func, bins, bmin, h_syn[func])
counts[func] += 1 if KS_syn>KS_est[func] else 0
for func in eval_funcs:
counts[func] /= num_data_sets
return counts # the p-values for each of the estimated functions
def preferentialHomophilyNetwork(legislators):
'''
Returns adjacency matrix based on preferential attachment with homophily
'''
for i in range(Num_of_Representatives):
legislators[i].linkTo(legislators[i],1) # create self-link
for rep1 in sample(legislators, Num_of_Representatives): # (a shuffle without changing the original list)
i = legislators.index(rep1)
rep1_issues = sorted(State.issues, key = lambda i: rep1.priorities[i], reverse=True)
potential_friends = []
for rep2 in legislators: # first find all potential friends with homophily
j = legislators.index(rep2)
if rep1!=rep2:
link_strength = 0
sum_pri_diffs = 0
for issue in rep1_issues[0:10]:
similarity = binaryTreeSimilarity(rep1.positions[issue], rep2.positions[issue])
pri_diff = 1 - abs(rep1.priorities[issue] - rep2.priorities[issue])
link_strength += similarity*pri_diff
sum_pri_diffs += pri_diff
link_strength = (link_strength/sum_pri_diffs)*2 - 1
if link_strength > Friend_Threshold:
potential_friends.append(rep2)
degrees = [1 for rep in potential_friends]
for k in range(Minimum_Friends):
degrees = [len(rep.links.values()) for rep in potential_friends]
if degrees:
friend_idx = randomFromCDF(cdf(degrees))
rep2 = potential_friends[friend_idx]
rep1.linkTo(rep2, 1)
potential_friends.remove(rep2) # can't friend someone twice
matrix = [[legislators[i].getLinkValueTo(legislators[j]) for i in range(Num_of_Representatives)]
for j in range(Num_of_Representatives)]
return matrix
def __init__(self, affiliation, priority_issues=[], default_positions={}):
self.priorities = {}
self.positions = {}
self.links = {} # network link strengths to other legislators
self.affiliation = affiliation
# generae priorities by preferential attachment
priorities = [1 for i in range(Num_of_Issues)]
for issue in priority_issues:
priorities[issue] = Priority_Multiplier * (priority_issues.index(issue) + 1)
for i in range(Num_of_Issues**2):
priorities[randomFromCDF(cdf(priorities))] += 1
# normalize the priorities to sum = 1
priorities = pdf(priorities)
shuffle(priorities)
#assign priorities and positions
for issue in State.issues:
self.priorities[issue] = priorities[issue]
self.positions[issue] = getrandbits(Solution_Bit_Length)
if issue in default_positions.keys(): # set the last bits according to default position
b = Ideology_Bit_Depth
#self.positions[issue] = (self.positions[issue]>>b)<<b + default_positions[issue]
self.positions[issue] = default_positions[issue]