[docs]
class AbstractSampler(object):
"""
API for Samplers, not all methods need to be implemented depending on the
use case (for example, load_sample may not be defined if positive /
negative cases are generated on the fly).
"""
# Classification categories
@property
def class_ids(self):
raise NotImplementedError
[docs]
def lookup_class_name(self, class_id):
raise NotImplementedError
[docs]
def lookup_class_id(self, class_name):
raise NotImplementedError
# Arbitrary subregion sampling
[docs]
def load_sample(self, tr, pad=None, window_dims=None, visible_thresh=0.1):
raise NotImplementedError
# Binary classification properties
@property
def n_positives(self):
raise NotImplementedError
[docs]
def load_item(self, index, pad=None, window_dims=None):
raise NotImplementedError
[docs]
def load_positive(self, index=None, pad=None, window_dims=None, rng=None):
raise NotImplementedError
[docs]
def load_negative(self, index=None, pad=None, window_dims=None, rng=None):
raise NotImplementedError
# Full image interface
[docs]
def load_image(self, image_id):
raise NotImplementedError
[docs]
def image_ids(self):
raise NotImplementedError
# Preselect
[docs]
def preselect(self, **kwargs):
""" Setup a pool of training examples before the epoch begins """
raise NotImplementedError