batch_train=True)
one_vs_rest = SeparateClassifiers(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
batch_train=True)
epsilon_greedy = EpsilonGreedy(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
batch_train=True)
epsilon_greedy_nodecay = EpsilonGreedy(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
decay=None,
batch_train=True)
adaptive_greedy_thr = AdaptiveGreedy(deepcopy(base_algorithm),
nchoices=nchoices,
decay_type='threshold',
batch_train=True)
adaptive_greedy_perc = AdaptiveGreedy(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
decay_type='percentile',
decay=0.9997,
batch_train=True)
explore_first = ExploreFirst(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=None,
explore_rounds=1500,
batch_train=True)
active_explorer = ActiveExplorer(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
num_samples, num_features = X.shape[0], X.shape[1]
# Contextual bandits setup
nchoices = 2 #hit or miss
base_algorithm = LogisticRegression(random_state=0, solver='lbfgs')
block_size = 1
max = 10
alpha = max / 2
beta = max / 2
beta_prior = ((alpha, beta), max / 2)
# Models
model_opts = [
AdaptiveGreedy(deepcopy(base_algorithm),
nchoices=nchoices,
decay_type='threshold'),
AdaptiveGreedy(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
decay_type='percentile',
decay=0.9997),
AdaptiveGreedy(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
active_choice='weighted',
decay_type='percentile',
decay=0.9997),
BootstrappedTS(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior),
X, y = parse_data("all.txt")
nchoices = y.shape[1]
base_sgd = SGDClassifier(random_state=123, loss='log', warm_start=False)
base_ols = LinearRegression(fit_intercept=True)
## Metaheuristic using different base algorithms and configurations
linucb = LinUCB(nchoices=nchoices, alpha=0.1)
### Important!!! the default hyperparameters for LinUCB in the reference paper
### are very different from what's used in this example
adaptive_active_greedy = AdaptiveGreedy(
deepcopy(base_ols),
nchoices=nchoices,
smoothing=None,
beta_prior=((3. / nchoices, 4.), 2),
active_choice='weighted',
decay_type='percentile',
decay=0.9997,
batch_train=True,
)
softmax_explorer = SoftmaxExplorer(deepcopy(base_sgd),
nchoices=nchoices,
smoothing=(1, 2),
beta_prior=None,
batch_train=True,
refit_buffer=50,
deep_copy_buffer=False,
random_state=3333)
adaptive_greedy_perc = AdaptiveGreedy(deepcopy(base_ols),
nchoices=nchoices,
smoothing=(1, 2),