for a_indx, algorithm in enumerate(algorithms):
regret[a_indx, m_indx, s] = algorithm.run(T, model)
return regret, models
experiment = Experiment(1)
experiment.log_code()
# Experiment 1
N = 50
epsilon = .3
simulations = 10000
T = 400
algorithms = [
GeneralCausal(truncate='None'),
ParallelCausal(),
SuccessiveRejects(),
AlphaUCB(2),
ThompsonSampling()
]
m_vals = range(2, N, 2)
regret, models = regret_vs_m(algorithms,
m_vals,
N,
T,
epsilon,
simulations=simulations)
experiment.plot_regret(regret,
return m_vals,regret,models
experiment = Experiment(4)
experiment.log_code()
N = 50
N1_vals = range(1,N,3)
pz = .4
q = (0.00001,0.00001,.4,.65)
epsilon = .3
simulations = 10000
T = 400
algorithms = [SuccessiveRejects(),GeneralCausal(),AlphaUCB(2),ThompsonSampling()]
epsilon = .3
pY = ParallelConfounded.pY_epsilon_best(q,pz,epsilon)
m_vals,regret,models = regret_vs_m_general(algorithms,N1_vals,N,T,pz,pY,q,epsilon,simulations = simulations)
experiment.plot_regret(regret,m_vals,"m",algorithms,legend_loc = "lower right",legend_extra = [ParallelCausal])
return regret, pulls
experiment = Experiment(6)
experiment.log_code()
N = 50
N1 = 1
pz = .4
q = (0.00001, 0.00001, .4, .65)
epsilon = .3
pY = ParallelConfounded.pY_epsilon_best(q, pz, epsilon)
simulations = 10000
model = ScaleableParallelConfounded(q, pz, pY, N1, N - N1)
T_vals = range(25, 626, 25)
algorithms = [
GeneralCausal(),
SuccessiveRejects(),
AlphaUCB(2),
ThompsonSampling()
]
regret, pulls = regret_vs_T(model, algorithms, T_vals, simulations=simulations)
experiment.plot_regret(regret, T_vals, "T", algorithms, legend_loc=None)
pY = np.asarray([[.4, .4], [.7, .7]])
models = [
ScaleableParallelConfounded(q, pz, pY, N1, N - N1)
for N1 in range(29, 36, 2)
]
#model = ScaleableParallelConfounded(q,pz,pY,33,N-33)
#print "built model 1"
#model2 = ScaleableParallelConfounded(q,pz,pY,35,N-35)
print "built model 2"
alg = GeneralCausal()
regret = np.zeros((len(models), 1000), dtype=float)
pulls = np.zeros((len(models), 1000, models[0].K), dtype=int)
for s in xrange(1000):
for i, model in enumerate(models):
regret[i, s] = alg.run(T, model)
pulls[i, s, alg.best_action] += 1
etas = [model.contract(model.eta) for model in models]
print regret.mean(axis=1)
for e in etas:
print e
en = [[0, 0, 1 / (N1 + 6), 0, 0, 0, 1 - N1 / (N1 + 6)]
for N1 in range(29, 36, 2)]