def regret_vs_m(algorithms, m_vals, N, T, epsilon, simulations=10):
models = []
regret = np.zeros((len(algorithms), len(m_vals), simulations))
for m_indx, m in enumerate(m_vals):
model = Parallel.create(N, m, epsilon)
models.append(model)
print "built model {0}".format(m)
for s in xrange(simulations):
for a_indx, algorithm in enumerate(algorithms):
regret[a_indx, m_indx, s] = algorithm.run(T, model)
return regret, models
for T_indx, T in enumerate(T_vals):
for a_indx, algorithm in enumerate(algorithms):
for s in xrange(simulations):
regret[a_indx, T_indx, s] = algorithm.run(T, model)
print T
return regret
experiment = Experiment(3)
experiment.log_code()
simulations = 10000
N = 50
m = 2
epsilon = .3
model = Parallel.create(N, m, epsilon)
T_vals = range(10, 6 * model.K, 25)
algorithms = [
GeneralCausal(truncate='None'),
ParallelCausal(),
SuccessiveRejects(),
AlphaUCB(2),
ThompsonSampling()
]
regret = regret_vs_T(model, algorithms, T_vals, simulations=simulations)
finished = now_string()
experiment.plot_regret(regret, T_vals, "T", algorithms, legend_loc=None)
def estimate_infrequent(self, h):
qij_hat = np.true_divide(self.trials, h)
s_indx = np.argsort(qij_hat) #indexes of elements from s in sorted(s)
m_hat = Parallel.calculate_m(qij_hat[s_indx])
infrequent = s_indx[0:m_hat]
return infrequent
model.set_epsilon(epsilon)
for s in xrange(simulations):
for a_indx, algorithm in enumerate(algorithms):
regret[a_indx, T_indx, s] = algorithm.run(T, model)
return regret
experiment = Experiment(2)
experiment.log_code()
N = 50
simulations = 10000
a = 9.0
m = 2
model = Parallel.create(N, m, .1)
Tmin = int(ceil(4 * model.K / a))
Tmax = 10 * model.K
T_vals = range(Tmin, Tmax, 100)
algorithms = [
GeneralCausal(truncate='None'),
ParallelCausal(),
SuccessiveRejects(),
AlphaUCB(2),
ThompsonSampling()
]
regret = regret_vs_T_vary_epsilon(model,
algorithms,