def select(self, label_index, unlabel_index, oracle=None, cost=None, budget=40):
"""Randomly selects a batch of instance-label pairs under the
constraints of meeting the budget conditions.
Parameters
----------
label_index: ignore
unlabel_index: {list, np.ndarray, MultiLabelIndexCollection}
The indexes of unlabeled samples. It should be a 1d array of indexes (column major, start from 0) or
MultiLabelIndexCollection or a list of tuples with 2 elements, in which,
the 1st element is the index of instance and the 2nd element is the index of labels.
oracle: Oracle,(default=None)
Oracle indicate the cost for each label.
Oracle in active learning whose role is to label the given query.And it can also give the cost of
each corresponding label.The Oracle includes the label and cost information at least.
Oracle(labels=labels, cost=cost)
cost: np.array, (default=None), shape [1, n_classes] or [n_classes]
The costs of querying each class.if not provide,it will all be 1.
budget: int, optional (default=40)
The budget of the select cost.If cost for eatch labels is 1,will degenerate into the batch_size.
Returns
-------
selected_ins_lab_pair: MultiLabelIndexCollection
The selected instance label pair.
"""
unlabel_index = self._check_multi_label_ind(unlabel_index)
n_classes = unlabel_index._label_size
assert(len(cost) == n_classes)
if oracle is None and cost is None:
raise ValueError('There is no information about the cost of each laebl. \
Please input Oracle or cost for the label at least.')
if oracle:
_, costs = oracle.query_by_index(range(n_classes))
else:
costs = cost
instance_pair = MultiLabelIndexCollection(label_size=n_classes)
un_ind = copy.deepcopy(unlabel_index)
current_cost = 0.
while True:
rand = np.random.choice(len(un_ind))
i_j = flattern_multilabel_index(un_ind.index)[rand]
j_class = i_j[1]
current_cost += costs[j_class]
if current_cost > budget:
break
instance_pair.update(i_j)
un_ind.difference_update(i_j)
# return instance_pair
return [tuple(i) for i in list(instance_pair)]
def _check_feature_ind(self, container):
if not isinstance(container, MultiLabelIndexCollection):
try:
if isinstance(container[0], tuple):
container = MultiLabelIndexCollection(
container, self.X.shape[1])
else:
container = MultiLabelIndexCollection.construct_by_1d_array(
container, label_mat_shape=self.X.shape)
except:
raise ValueError(
"Please pass a 1d array of indexes or MultiLabelIndexCollection (column major, start from 0)"
"or a list of tuples with 2 elements, in which, the 1st element is the index of instance "
"and the 2nd element is the index of features.")
return container
def _check_multi_label_ind(self, container):
"""Check if the given array is an array of multi label indexes."""
if not isinstance(container, MultiLabelIndexCollection):
try:
if isinstance(container[0], tuple):
container = MultiLabelIndexCollection(
container, self.y.shape[1])
else:
container = MultiLabelIndexCollection.construct_by_1d_array(
container, label_mat_shape=self.y.shape)
except:
raise ValueError(
"Please pass a 1d array of indexes or MultiLabelIndexCollection (column major, "
"start from 0) or a list "
"of tuples with 2 elements, in which, the 1st element is the index of instance "
"and the 2nd element is the index of label.")
return copy.copy(container)
a_ind = IndexCollection(a)
# add a single index, warn if there is a repeated element.
a_ind.add(4)
# discard a single index, warn if not exist.
a_ind.discard(4)
# add a batch of indexes.
a_ind.update([4, 5])
# discard a batch of indexes.
a_ind.difference_update([1, 2])
print(a_ind)
# ---------MultiLabelIndexCollection-------------
from alipy.index import MultiLabelIndexCollection
multi_lab_ind1 = MultiLabelIndexCollection([(0, 1), (0, 2), (0, (3, 4)),
(1, (0, 1))],
label_size=5)
multi_lab_ind1.update((0, 0))
multi_lab_ind1.update([(1, 2), (1, (3, 4))])
multi_lab_ind1.update([(2, )])
multi_lab_ind1.difference_update([(0, )])
print(multi_lab_ind1)
# matlab style 1d index supporting
b = [1, 4, 11]
mi = MultiLabelIndexCollection.construct_by_1d_array(array=b,
label_mat_shape=(3, 4))
print(mi)
print('col major:', mi.get_onedim_index(order='F', ins_num=3))
print('row major:', mi.get_onedim_index(order='C'))
def hierarchical_multilabel_mark(multilabel_index, label_index, label_tree, y_true):
""""Complete instance-label information according to hierarchy in the label-tree.
Parameters
----------
label_index: {list, np.ndarray, MultiLabelIndexCollection}
The indexes of labeled samples. It should be a 1d array of indexes (column major, start from 0) or
MultiLabelIndexCollection or a list of tuples with 2 elements, in which,
the 1st element is the index of instance and the 2nd element is the index of labels.
multilabel_index: {list, np.ndarray, MultiLabelIndexCollection}
The indexes of labeled samples. It should be a 1d array of indexes (column major, start from 0) or
MultiLabelIndexCollection or a list of tuples with 2 elements, in which,
the 1st element is the index of instance and the 2nd element is the index of labels.
label_tree: np.ndarray
The hierarchical relationships among data features.
if node_i is the parent of node_j , then label_tree(i,j)=1
y_true: 2D array, optional (default=None)
Label matrix of the whole dataset. It is a reference which will not use additional memory.
shape [n_samples, n_classes]
Returns
-------
selected_ins_lab_pair: list
A list of tuples that contains the indexes of selected instance-label pairs.
"""
# try to convert the indexes
if not isinstance(multilabel_index, MultiLabelIndexCollection):
try:
if isinstance(multilabel_index[0], tuple):
container = MultiLabelIndexCollection(multilabel_index, np.shape(y_true)[1])
else:
container = MultiLabelIndexCollection.construct_by_1d_array(multilabel_index, label_mat_shape=np.shape(y_true))
except:
raise ValueError(
"Please pass a 1d array of indexes or MultiLabelIndexCollection (column major, "
"start from 0) or a list "
"of tuples with 2 elements, in which, the 1st element is the index of instance "
"and the 2nd element is the index of label.")
multilabel_index = copy.deepcopy(container)
if not isinstance(label_index, MultiLabelIndexCollection):
try:
if isinstance(label_index[0], tuple):
container = MultiLabelIndexCollection(label_index, np.shape(y_true)[1])
else:
container = MultiLabelIndexCollection.construct_by_1d_array(label_index, label_mat_shape=np.shape(y_true))
except:
raise ValueError(
"Please pass a 1d array of indexes or MultiLabelIndexCollection (column major, "
"start from 0) or a list "
"of tuples with 2 elements, in which, the 1st element is the index of instance "
"and the 2nd element is the index of label.")
label_index = copy.deepcopy(container)
n_classes = multilabel_index._label_size
assert(np.shape(label_tree)[0] == n_classes and np.shape(label_tree)[1] == n_classes)
add_label_index = MultiLabelIndexCollection(label_size=n_classes)
for instance_label_pair in multilabel_index:
i_instance = instance_label_pair[0]
j_label = instance_label_pair[1]
if y_true[instance_label_pair] == 1:
for descent_label in range(n_classes):
if label_tree[j_label][descent_label] == 1:
if (not (i_instance, descent_label) in label_index):
add_label_index.update((i_instance, descent_label))
elif y_true[instance_label_pair] == -1:
for parent_label in range(n_classes):
if label_tree[parent_label][j_label] == 1:
if (not (i_instance, parent_label) in label_index):
add_label_index.update((i_instance, parent_label))
for i in add_label_index:
if not i in multilabel_index:
multilabel_index.update(i)
return multilabel_index
# Use the default Logistic Regression classifier
model = LogisticRegression()
# The cost budget is 50 times querying
stopping_criterion = StoppingCriteria('num_of_queries', 50)
AFASMC_result = []
rand_result = []
Stable_result = []
# AFASMC
for i in range(5):
train_idx = tr[i]
test_idx = te[i]
label_ind = MultiLabelIndexCollection(lab[i], label_size=X.shape[1])
unlab_ind = MultiLabelIndexCollection(unlab[i], label_size=X.shape[1])
saver = StateIO(i, train_idx, test_idx, label_ind, unlab_ind)
strategy = QueryFeatureAFASMC(X=X, y=y, train_idx=train_idx)
while not stopping_criterion.is_stop():
# query
selected_feature = strategy.select(observed_entries=label_ind,
unkonwn_entries=unlab_ind)
# update index
label_ind.update(selected_feature)
unlab_ind.difference_update(selected_feature)
# train/test
lab_in_train = map_whole_index_to_train(train_idx, label_ind)