def test_ucb_policy(n_samples, n_actions, context_dim, dataset):
# define a solver
ucbp = UCBPolicy(n_actions=n_actions, lr=0.01)
policies = [ucbp]
results = simulate_cb(dataset, n_samples, policies)
# no operational error
assert results[0]["simple_regret"] > -1.0
def test_epsilon_greedy_policy(n_samples, n_actions, context_dim, dataset):
# define a solver
egp = EpsilonGreedyPolicy(n_actions=n_actions, lr=0.1, epsilon=0.1)
policies = [egp]
results = simulate_cb(dataset, n_samples, policies)
# no operational error
assert results[0]["simple_regret"] > -1.0
def test_linucb_policy(n_samples, n_actions, context_dim, dataset):
# define a solver
linucbp = LinUCBPolicy(n_actions=n_actions,
context_dim=context_dim,
delta=0.25,
train_starts_at=500,
train_freq=50)
policies = [linucbp]
results = simulate_cb(dataset, n_samples, policies)
# must avoid getting stuck at no eating
# not sure about synthetic
assert results[0]["simple_regret"] > -1.0
def test_linear_gaussian_thompson_sampling_policy(n_samples, n_actions,
context_dim, dataset):
lgtsp = LinearGaussianThompsonSamplingPolicy(n_actions=n_actions,
context_dim=context_dim,
eta_prior=6.0,
lambda_prior=0.25,
train_starts_at=500,
posterior_update_freq=50)
policies = [lgtsp]
results = simulate_cb(dataset, n_samples, policies)
# must avoid getting stuck at no eating
# not sure about synthetic
assert results[0]["simple_regret"] > -1.0
