def simulateAuction( **kwargs ):
agentType = kwargs.get('agentType')
nAgents = kwargs.get('nAgents')
margDist = kwargs.get('margDist')
nGames = kwargs.get('nGames')
m = margDist.m
winningBids = numpy.zeros((nGames,m))
for g in xrange(nGames):
agentList = [agentFactory(agentType = agentType, m = m) for i in xrange(nAgents)]
bids = numpy.atleast_2d([agent.bid(margDistPrediction = margDist) for agent in agentList])
winningBids[g,:] = numpy.max(bids,0)
# return numpy.max(bids,0)
return winningBids
def bayesSCPP(**kwargs):
"""
save pkl will save the hist class; to recover bayesian probabilities must call hist.bayesMargDistSCPP()
"""
oDir = kwargs.get('oDir')
agentType = kwargs.get('agentType')
nAgents = kwargs.get('nAgents',8)
m = kwargs.get('m',5)
minPrice = kwargs.get('minPrice',0)
maxPrice = kwargs.get('maxPrice',50)
maxSim = kwargs.get('maxSim', 1000)
nGames = kwargs.get('nGames', 100)
parallel = kwargs.get('parallel', False)
nProc = kwargs.get('nProc', multiprocessing.cpu_count() - 1)
tol = kwargs.get('tol',0.001)
plot = kwargs.get('plot', True)
log = kwargs.get('log', True)
saveBayesPkl = kwargs.get('saveBayesPkl',True)
# saveBayesNpz = kwargs.get('saveBayesNpz',True)
saveBayesTxt = kwargs.get('saveBayesTxt',True)
saveHistPkl = kwargs.get('saveHistPkl',False)
# saveHistNpz = kwargs.get('saveHistNpz',False)
savenpz = kwargs.get('savenpz',False)
verbose = kwargs.get('verbose', True)
if not oDir:
str = "-----ERROR-----\n" +\
"In bayesSCPP(...)\n" +\
"Must Provide output directory\n"
raise ValueError(str)
if plot:
pltDir = os.path.join(oDir,'plots')
if not os.path.exists(pltDir):
os.makedirs(pltDir)
else:
[os.remove(filename) for filename in glob.glob(os.path.join(pltDir,'*.png'))]
# shutil.rmtree(pltDir,ignore_errors=True)
# os.makedirs(pltDir)
if not os.path.exists(oDir):
os.makedirs(oDir)
if saveBayesPkl:
bayesPklDir = os.path.join(oDir,'bayesSCPP_bayesPkl')
if not os.path.exists(bayesPklDir):
os.makedirs(bayesPklDir)
else:
[os.remove(f) for f in glob.glob(os.path.join(bayesPklDir,'*.pkl'))]
if saveBayesTxt:
bayesTxtDir = os.path.join(oDir,'bayesSCPP_bayesTxt')
if not os.path.exists(bayesTxtDir):
os.makedirs(bayesTxtDir)
else:
[os.remove(f) for f in glob.glob(os.path.join(bayesTxtDir,'*txt'))]
if saveHistPkl:
histPklDir = os.path.join(oDir,'bayesSCPP_histPkl')
if not os.path.exists(histPklDir):
os.makedirs(histPklDir)
else:
[os.remove(f) for f in glob.glob(os.path.join(histPklDir,'*.pkl'))]
if log:
logFile = os.path.join(oDir,'bayesSCPP_{0}.txt'.format(agentType))
if os.path.exists(logFile):
os.remove(logFile)
with open(logFile,'a') as f:
f.write("oDir = {0}\n".format(oDir))
f.write("agentType = {0}\n".format(agentType))
f.write("nAgents = {0}\n".format(nAgents))
f.write("tol = {0}\n".format(tol))
f.write("m = {0}\n".format(m))
f.write("minPrice = {0}\n".format(minPrice))
f.write("maxPrice = {0}\n".format(maxPrice))
f.write("maxSim = {0}\n".format(maxSim))
f.write("nGames = {0}\n".format(nGames))
f.write("parallel = {0}\n".format(parallel))
f.write("plot = {0}\n".format(plot))
if verbose:
print "oDir = {0}".format(oDir)
print "agentType = {0}".format(agentType)
print "nAgents = {0}".format(nAgents)
print "tol = {0}".format(tol)
print "m = {0}".format(m)
print "minPrice = {0}".format(minPrice)
print "maxPrice = {0}".format(maxPrice)
print "maxSim = {0}".format(maxSim)
print "nGames = {0}".format(nGames)
print "parallel = {0}".format(parallel)
print "plot = {0}".format(plot)
currHist = hist()
if plot:
oFile = os.path.join(pltDir,'bayesSCPP_{0}_{1}.png'.format(agentType,0))
title='bayesSCPP, {0}, Initial Distribution'.format(agentType)
currHist.bayesMargDistSCPP().graphPdfToFile(fname = oFile,
title=title)
klList = []
ksList = []
agentFactoryParams = {'agentType' : agentType,
'm' : m,
'minPrice' : minPrice,
'maxPrice' : maxPrice}
for sim in xrange(maxSim):
oldHist = copy.deepcopy(currHist)
for i in xrange(nGames):
agentList = [agentFactory(**agentFactoryParams) for i in xrange(nAgents)]
if parallel:
pool = multiprocessing.Pool(nProc)
bayesMarg = currHist.bayesMargDistSCPP()
results = [pool.apply_async(bidHelper, kwds = {'agent': agent, 'bayesMargDist': bayesMarg} ) for agent in agentList]
pool.close()
pool.join()
bids = numpy.zeros((nAgents,m))
for idx,r in enumerate(results):
bids[idx,:] = r.get()
r._value = []
else:
bids = numpy.atleast_2d([agent.bid(pricePrediction = currHist.bayesMargDistSCPP()) for agent in agentList])
winningBids = numpy.max(bids,0)
for idx, wb in enumerate(winningBids):
currHist.upcount(idx,wb,mag=1)
# [currHist.upcount(idx, wb, mag=1) for idx, wb in enumerate(winningBids)]
currBayesMargDist = currHist.bayesMargDistSCPP()
oldBayesMargDist = oldHist.bayesMargDistSCPP()
klList.append(klDiv(currBayesMargDist, oldBayesMargDist))
ksList.append(ksStat(currBayesMargDist, oldBayesMargDist))
if verbose:
print ''
print 'itr = {0}'.format(sim+1)
print '\tNumber of Games = {0}'.format((sim+1)*nGames)
print '\tkld = {0}'.format(klList[-1])
print '\tks = {0}'.format(ksList[-1])
if plot:
oPlot = os.path.join(pltDir,'bayesSCPP_{0}_{1}.png'.format(agentType,(sim+1)*nGames))
title='BayesSCPP {0}, klD = {1:.6}, ks = {2:.6} itr = {3}'.format(agentType,klList[-1],ksList[-1],(sim+1)*nGames)
currBayesMargDist.graphPdfToFile(fname = oPlot, title=title)
if saveBayesPkl:
of = os.path.join(bayesPklDir,'bayesSCPP_bayesMargDist_{0}_{1}.pkl'\
.format(agentType,int((sim+1)*nGames)))
with open(of, 'w') as f:
pickle.dump(currBayesMargDist,f)
if saveBayesTxt:
of = os.path.join(bayesTxtDir,'bayesSCPP_bayesTxt_{0}_{1}.txt'\
.format(agentType,int((sim+1)*nGames)))
currBayesMargDist.savetxt(of)
if saveHistPkl:
of = os.path.join(bayesPklDir,'bayesSCPP_hist_{0}_{1}.pkl'\
.format(agentType,(sim+1)*nGames))
with open(of, 'w') as f:
pickle.dump(currHist,f)
if klList[-1] < tol:
break
if log:
with open(logFile,'a') as f:
f.write('Done after {0} games ({1} iterations)\n'.format((sim+1)*nGames,sim))
f.write('kl = {0}\n'.format(klList))
f.write('ks = {0}\n'.format(ksList))
klFile = os.path.join(oDir,'kl.json')
with open(klFile,'w') as f:
json.dump(klList,f)
ksFile = os.path.join(oDir,'ks.json')
with open(ksFile,'w') as f:
json.dump(ksList,f)
if verbose:
print 'Done'
def comp2Agents(**kwargs):
oDir = kwargs.get('oDir')
pp1 = kwargs.get('pp1')
n1 = kwargs.get('n1',4)
pp2 = kwargs.get('pp2')
n2 = kwargs.get('n2',4)
agentType1 = kwargs.get('agentType1')
agentType2 = kwargs.get('agentType2')
m = kwargs.get('m',5)
minValuation = kwargs.get('minValuation',0)
maxValuation = kwargs.get('maxValuation',50)
nGames = kwargs.get('nGames',1000)
parallel = kwargs.get('parallel',True)
nProc = kwargs.get('nProc', multiprocessing.cpu_count() - 1)
verbose = kwargs.get('verbose', True)
if verbose:
print ''
print 'In comp2Agents(...)'
print 'oDir = {0}'.format(oDir)
print 'agentType1 = {0}'.format(agentType1)
print 'n1 = {0}'.format(n1)
print 'agentType2 = {0}'.format(agentType2)
print 'n2 = {0}'.format(n2)
print 'm = {0}'.format(m)
print 'nGames = {0}'.format(nGames)
print 'minValuation = {0}'.format(minValuation)
print 'maxValuation = {0}'.format(maxValuation)
print ''
if parallel:
pool = multiprocessing.Pool(nProc)
nGameList = [nGames//nProc]*nProc
nGameList[-1] += (nGames % nProc)
if verbose:
print 'Running parallel simulation.'
print 'Number of cores = {0}'.format(nProc)
print 'Number of simulations per core = {0}'.format(nGameList)
print 'Total Number of simulations = {0}'.format((sum(nGameList)))
results = []
for p in xrange(nProc):
subArgs = copy.deepcopy(kwargs)
subArgs.update(kwargs)
subArgs['parallel'] = False
subArgs['nGames'] = nGameList[p]
subArgs['verbose'] = False
subArgs['oDir'] = None
results.append(pool.apply_async(comp2Agents, kwds = subArgs))
pool.close()
pool.join()
for idx, r in enumerate(results):
if idx == 0:
agentSurplus = r.get()
else:
agentSurplus = numpy.concatenate((agentSurplus,r.get()))
r._value = []
else:
auction = simultaneousAuction( m = m,
nPrice = 2,
reserve = 0)
for i in xrange(n1):
agent = agentFactory(agentType = agentType1,
m = m,
vmin = minValuation,
vmax = maxValuation )
agent.pricePrediction = pp1
auction.attachAgents(agent)
for i in xrange(n2):
agent = agentFactory(agentType = agentType2,
m = m,
vmin = minValuation,
vmax = maxValuation )
agent.pricePrediction = pp2
auction.attachAgents(agent)
agentSurplus = numpy.zeros((nGames,n1+n2))
for itr in xrange(nGames):
if verbose:
print 'Simulating {0} out of {1} auctions'.format(itr,nGames)
[agent.randomValuation() for agent in auction.agentList]
if verbose:
for idx, agent in enumerate(auction.agentList):
print 'agent[{0}] = {1} ; l = {2}, v = {3}'.format(idx, agent.type(), agent.l, agent.v)
auction.runAuction()
auction.notifyAgents()
surplus = auction.agentSurplus()
agentSurplus[itr,:] = surplus
if verbose:
print 'Agent Surplus = {0}'.format(surplus)
if oDir:
oDir = os.path.realpath(oDir)
if not os.path.exists(oDir):
os.makedirs(oDir)
sFile = os.path.join(oDir,'{0}_{1}_agentSurplus.txt'.format(agentType1,agentType2))
numpy.savetxt(sFile, agentSurplus)
return agentSurplus
def simAuctionHelper(**kwargs):
agentType = kwargs.get('agentType')
if isinstance(agentType,list):
nAgents = len(agentType)
elif isinstance(agentType,str):
nAgents = kwargs.get('nAgents',8)
agentType = [agentType]*nAgents
nGames = kwargs.get('nGames')
pricePrediction = kwargs.get('pricePrediction')
m = kwargs.get('m',5)
minValuation = kwargs.get('minValuation',0)
maxValuation = kwargs.get('maxValuation',50)
verbose = kwargs.get('verbose', False)
retType = kwargs.get('retType','bids')
l = kwargs.get('l')
if retType == 'hob':
selfIdx = kwargs.get('selfIdx')
if selfIdx == None:
raise ValueError("ERROR - simulateAuction(...):\n" + \
"\t Must specify selfIdx when retType == 'hob'")
if verbose:
print 'In simulateAuction(...)'
print 'agentType = {0}'.format(agentType)
print 'nGames = {0}'.format(nGames)
print 'm = {0}'.format(m)
print 'minValuation = {0}'.format(minValuation)
print 'maxValuation = {0}'.format(maxValuation)
print 'retType = {0}'.format(retType)
if retType == 'bids':
ret = numpy.zeros((nGames,nAgents,m))
elif retType == 'firstPrice' or retType == 'hob':
ret = numpy.zeros((nGames,m))
else:
raise ValueError("simulateAuction - Unknown return type")
agents = [agentFactory(agentType = atype, m = m, vmin = minValuation, vmax = maxValuation) for atype in agentType]
for itr in xrange(nGames):
if verbose:
print 'running serial game {0}'.format(itr)
if retType == 'firstPrice' or retType == 'secondPrice' or retType == 'hob':
gameBids = numpy.zeros((nAgents,m))
if isinstance(pricePrediction,list):
for agentIdx, agent, pp in zip(numpy.arange(nAgents),agents,pricePrediction):
agent.randomValuation(l = l)
if retType == 'bids':
ret[itr, agentIdx, :] = agent.bid(pricePrediction = pp)
elif retType == 'firstPrice' or retType == 'secondPrice' or retType == 'hob':
gameBids[agentIdx,:] = agent.bid(pricePrediction = pp)
else:
for agentIdx, agent in enumerate(agents):
agent.randomValuation(l=l)
if retType == 'bids':
ret[itr, agentIdx, :] = agent.bid(pricePrediction = pricePrediction)
elif retType == 'firstPrice' or retType == 'secondPrice' or retType == 'hob':
gameBids[agentIdx,:] = agent.bid(pricePrediction = pricePrediction)
if retType == 'firstPrice':
ret[itr,:] = numpy.max(gameBids,0)
elif retType == 'hob':
ret[itr,:] = numpy.max( numpy.delete(gameBids,selfIdx,0), 0 )
elif retType == 'secondPrice':
for goodIdx in xrange(gameBids.shape[1]):
goodBids = gameBids[:,m]
goodArgMax = numpy.argmax(goodBids)
ret[itr,goodIdx] = numpy.max(numpy.delete(goodBids,goodArgMax))
return ret
def hobSim(**kwargs):
"""
Function to simulate an auction and record other highest agent bids.
Parameters
----------
agentTypeList : list of strings, required
Types of agents in the auction.
selfIdx : int, zero indexed, 0 <= selfIdx < len(agentTypeList)
Index for agent in list representing self.
nGames : int, optional
Number of simulations to run. Defaults to 10,000
parallel : bool, optional
Run simulations in parallel or serial (mainly for debugg). Defaults True.
simulateAuctionJointGMM
pricePrediction : Either margDistSCPP or jointGMM instance
Price Predictions for agents to use.
m := int, optional
Number of goods up for auction. Defaults to 5.
vmin : float, optional
Minimum allowable valuation. Defaults to 0.
vmax : float, optional
Maximum allowable valuation. Defaults to 50.
outputs
-------
highestOtherBids : ndarray (nGames,m)
Array of highest other agent bids.
"""
agentTypeList = kwargs.get('agentTypeList')
selfIdx = kwargs.get('selfIdx')
nGames = kwargs.get('nGames')
parallel = kwargs.get('parallel',True)
nProc = kwargs.get('nProc', multiprocessing.cpu_count())
pricePrediction = kwargs.get('pricePrediction')
m = kwargs.get('m',5)
vmin = kwargs.get('vmin',0)
vmax = kwargs.get('vmax',50)
highestOtherBids = numpy.zeros((nGames,m))
if parallel:
pool = multiprocessing.Pool(nProc)
# winning
else:
n = len(agentTypeList)
agents = [agentFactory(agentType = agentType, m = m, vmin = vmin, vmax = vmax) for agentType in agentTypeList]
otherAgents = numpy.delete(numpy.arange(n),selfIdx,0)
bids = numpy.zeros((n,m))
for itr in xrange(nGames):
if isinstance(pricePrediction,list):
for idx, agent, pp in zip(numpy.arange(n),agents,pricePrediction):
#draw a new valuation
agent.randomValuation()
bids[idx] = agent.bid(pricePrediction = pp)
else:
for idx, agent in enumerate(agents):
#draw a new valuation
agent.randomValuation()
bids[idx] = agent.bid(pricePrediction = pricePrediction)
highestOtherBids[itr,:] = numpy.max(bids[otherAgents],0)
return highestOtherBids
