def experiment(k):
# Agent initialization
np.random.seed()
agent = AgentFactored(budgetTot=1000,
deadline=deadline,
nCampaigns=nCampaigns,
nBudget=nIntervals,
nBids=nBids,
maxBudget=100.0)
agent.initGPs()
print "Experiment : ", k
print "A"
# Set the GPs hyperparameters
for c in range(0, nCampaigns):
agent.setGPKernel(c, oracle.gpsClicks[c].kernel_,
oracle.gpsCosts[c].kernel_)
# Init the Core and execute the experiment
env = Environment(copy.copy(campaigns))
core = Core(agent, copy.copy(env), deadline)
core.runEpisode()
np.save(path + "policy_" + str(k), [agent.prevBids, agent.prevBudgets])
np.save(path + "experiment_" + str(k), np.sum(agent.prevConversions,
axis=1))
return np.sum(agent.prevConversions, axis=1), agent
def experiment(k):
np.random.seed()
print "Esperimento: ", k
agentGPUCB = AgentFactored(1000, deadline, ncampaigns, nIntervals, nBids,
maxBudget, 1.0, "GPUCB")
agentGPUCB.initGPs()
envGPUCB = Environment(copy.copy(campaigns))
coreGPUCB = Core(agentGPUCB, envGPUCB, deadline)
meanConvGPUCB = np.zeros(deadline)
# in questo ciclo mi salvo le conversioni medie in ogni istante
for t in range(deadline):
print "Day: ", t + 1
coreGPUCB.step()
meanConvGPUCB[t] = lastMeanConv(agentGPUCB)
# ora invece mi salvo le conversioni istantanee
instConvGPUCB = np.sum(agentGPUCB.prevConversions, axis=1)
positionGPUCB1 = path_dati + "inst_conv_3camp_" + str(k)
positionGPUCB2 = path_dati + "mean_conv_3camp_" + str(k)
np.save(positionGPUCB1, instConvGPUCB)
np.save(positionGPUCB2, meanConvGPUCB)
return
Campaign(a5,
nMeanResearch=1450.0,
nStdResearch=50.0,
probClick=probClick[4],
convParams=convparams))
ncampaigns = len(campaigns)
nBids = 10
nIntervals = 10
deadline = 250
maxBudget = 100
agent = AgentFactored(1000, deadline, ncampaigns, nIntervals, nBids, maxBudget,
1.0)
agent.initGPs()
env = Environment(campaigns)
plotter = PlotterFinal(agent=agent, env=env)
# mi creo una lista con tutte le matrici dell'oracolo di ogni campagna
listMeans = list()
listVar = list()
for i in range(0, ncampaigns):
[trueMeans, trueVar] = plotter.oracleMatrix(indexCamp=i, nsimul=20)
listMeans.append(trueMeans)
listVar.append(trueVar)
if i == 0:
optMatrix = np.array([trueMeans.max(axis=1)])
else:
maxrow = np.array([trueMeans.max(axis=1)])
optMatrix = np.concatenate((optMatrix, maxrow))