def __init__(self, base_algorithm, reward_estimator, nchoices, method='rovr',
handle_invalid=True, c=None, pmin=1e-5, beta_prior=None, smoothing=(1,2), **kwargs_costsens):
assert (method == 'rovr') or (method == 'wap')
self.method = method
if method == 'wap':
_check_constructor_input(base_algorithm, nchoices)
else:
assert isinstance(nchoices, int)
assert nchoices > 2
assert ('fit' in dir(base_algorithm)) and ('predict' in dir(base_algorithm))
if c is not None:
assert isinstance(c, float)
if pmin is not None:
assert isinstance(pmin, float)
assert isinstance(handle_invalid, bool)
if type(reward_estimator) == np.ndarray:
assert reward_estimator.shape[1] == nchoices
assert reward_estimator.shape[0] == X.shape[0]
else:
assert ('predict_proba_separate' in dir(reward_estimator)) or ('predict_proba' in dir(reward_estimator))
if beta_prior is not None:
beta_prior = _check_beta_prior(beta_prior, nchoices, 2)
self.base_algorithm = base_algorithm
self.reward_estimator = reward_estimator
self.nchoices = nchoices
self.c = c
self.pmin = pmin
self.handle_invalid = handle_invalid
self.beta_prior = beta_prior
self.smoothing = _check_smoothing(smoothing)
self.kwargs_costsens = kwargs_costsens
def __init__(self, base_algorithm, nchoices, c=None, pmin=1e-5):
_check_constructor_input(base_algorithm, nchoices)
self.base_algorithm = base_algorithm
self.nchoices = nchoices
self.tree = _BinTree(nchoices)
if c is not None:
assert isinstance(c, float)
if pmin is not None:
assert isinstance(pmin, float)
self.c = c
self.pmin = pmin
def __init__(self, base_algorithm, nchoices, c = None, pmin = 1e-5, njobs = -1):
try:
from costsensitive import _BinTree
except:
raise ValueError("This functionality requires package 'costsensitive'.\nCan be installed with 'pip install costsensitive'.")
_check_constructor_input(base_algorithm, nchoices)
self.base_algorithm = base_algorithm
self.nchoices = nchoices
self.tree = _BinTree(nchoices)
if c is not None:
assert isinstance(c, float)
if pmin is not None:
assert isinstance(pmin, float)
self.c = c
self.pmin = pmin
self.njobs = _check_njobs(njobs)