def fit(self, X, s, prob_s_eq_1=None, cv_n_folds=3):
'''Train the classifier using X examples and s labels.
Parameters
----------
X : np.array
Input feature matrix (N, D), 2D numpy array
s : np.array
A binary vector of labels, s, which may contain mislabeling
prob_s_eq_1 : iterable (list or np.array)
The probability, for each example, whether it is s==1 P(s==1|x).
If you are not sure, leave prob_s_eq_q = None (default) and
it will be computed for you using cross-validation.
cv_n_folds : int
The number of cross-validation folds used to compute
out-of-sample probabilities for each example in X.
'''
assert_inputs_are_valid(X, s, prob_s_eq_1)
if prob_s_eq_1 is None:
prob_s_eq_1 = cv_pred_proba(
X=X,
s=s,
clf=self.clf,
cv_n_folds=cv_n_folds,
)
if self.e1 is None:
self.e1 = np.mean(prob_s_eq_1[s == 1])
self.clf.fit(X, s)
def fit(self, X, s, pulearning=None):
'''
Parameters
----------
X : np.array
Input feature matrix (N, D), 2D numpy array
s : np.array
A binary vector of labels, s, which may contain mislabeling
pulearning : bool
Set to True if you wish to perform PU learning. PU learning assumes
that positive examples are perfectly labeled (contain no mislabeling)
and therefore frac_neg2pos = 0 (rh0 = 0). If
you are not sure, leave pulearning = None (default).
'''
# Check if we are in the PU learning setting.
if pulearning is None:
pulearning = (self.rh0 == 0)
assert_inputs_are_valid(X, s)
# Set rh0 = 0 if no negatives exist in P.
rh0 = 0.0 if pulearning else self.rh0
rh1 = self.rh1
alpha = float(1 - rh1 + rh0) / 2
sample_weight = np.ones(np.shape(s)) * (1 - alpha)
sample_weight[s == 0] = alpha
self.clf.fit(X, s, sample_weight=sample_weight)
def fit(
self,
X,
s,
pulearning=None,
prob_s_eq_1=None,
cv_n_folds=3,
):
'''
Parameters
----------
X : np.array
Input feature matrix (N, D), 2D numpy array
s : np.array
A binary vector of labels, s, which may contain mislabeling
pulearning : bool
Set to True if you wish to perform PU learning. PU learning assumes
that positive examples are perfectly labeled (contain no mislabeling)
and therefore frac_neg2pos = 0 (rh0 = 0). If
you are not sure, leave pulearning = None (default).
prob_s_eq_1 : iterable (list or np.array)
The probability, for each example, whether it is s==1 P(s==1|x).
If you are not sure, leave prob_s_eq_q = None (default) and
it will be computed for you using cross-validation.
cv_n_folds : int
The number of cross-validation folds used to compute
out-of-sample probabilities for each example in X.
'''
# Check if we are in the PU learning setting.
if pulearning is None:
pulearning = (self.rh0 == 0)
assert_inputs_are_valid(X, s, prob_s_eq_1)
# Set rh0 = 0 if no negatives exist in P.
rh0 = 0.0 if pulearning else self.rh0
rh1 = self.rh1
if prob_s_eq_1 is None:
prob_s_eq_1 = cv_pred_proba(
X=X,
s=s,
clf=self.clf,
cv_n_folds=cv_n_folds,
)
# Liu2016 using probabilities
assert prob_s_eq_1 is not None, "Error: prob_s_eq_1 is None type."
rho_s_opposite = np.ones(np.shape(prob_s_eq_1)) * rh0
rho_s_opposite[s == 0] = rh1
sample_weight = (prob_s_eq_1 -
rho_s_opposite) / prob_s_eq_1 / float(1 - rh1 - rh0)
self.clf.fit(X, s, sample_weight=sample_weight)
def fit(self, X, s):
'''Train the classifier clf with s labels.
X : np.array
Input feature matrix (N, D), 2D numpy array
s : np.array
A binary vector of labels, s, which may contain mislabeling
'''
assert_inputs_are_valid(X, s)
self.clf.fit(X, s)