beta_prior=None,
explore_rounds=1500,
batch_train=True)
active_explorer = ActiveExplorer(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
batch_train=True)
adaptive_active_greedy = AdaptiveGreedy(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
active_choice='weighted',
decay_type='percentile',
decay=0.9997,
batch_train=True)
softmax_explorer = SoftmaxExplorer(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior,
batch_train=True)
models = [
bootstrapped_ucb, bootstrapped_ts, one_vs_rest, epsilon_greedy,
epsilon_greedy_nodecay, adaptive_greedy_thr, adaptive_greedy_perc,
explore_first, active_explorer, adaptive_active_greedy, softmax_explorer
]
# These lists will keep track of the rewards obtained by each policy
rewards_ucb, rewards_ts, rewards_ovr, rewards_egr, rewards_egr2, \
rewards_agr, rewards_agr2, rewards_efr, rewards_ac, \
rewards_aac, rewards_sft = [list() for i in range(len(models))]
lst_rewards = [
rewards_ucb, rewards_ts, rewards_ovr, rewards_egr, rewards_egr2,
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),
BootstrappedUCB(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior),
SoftmaxExplorer(deepcopy(base_algorithm),
nchoices=nchoices,
beta_prior=beta_prior)
]
model_names_opts = [
"adaptive_greedy_thres", "adaptive_greedy_perc",
"adaptive_active_greedy", "bootstrapped_ts", "bootstrapped_ucb",
"softmax"
]
'''
models = [AdaptiveGreedy(deepcopy(base_algorithm), nchoices=nchoices, decay_type='threshold')]
models = [AdaptiveGreedy(deepcopy(base_algorithm), nchoices = nchoices, beta_prior=beta_prior, decay_type='percentile', decay=0.9997)]
models = [AdaptiveGreedy(deepcopy(base_algorithm), nchoices = nchoices, beta_prior=beta_prior, active_choice='weighted', decay_type='percentile', decay=0.9997)]
#models = [BootstrappedTS(deepcopy(base_algorithm), nchoices = nchoices, beta_prior=beta_prior)]
model_names = ["adaptive_greedy_thres"]
### 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),
beta_prior=None,
decay_type='percentile',
decay=0.9997,
batch_train=True,
random_state=4444)
active_explorer = ActiveExplorer(deepcopy(base_sgd),
smoothing=None,
nchoices=nchoices,
beta_prior=((3. / nchoices, 4.), 2),