class DeepFeatureRep(interface):
r'''
mean_F: \mathbb{R}^{N x H x W x 3} \to \mathbb{R}^{D}
F_inverse: \mathbb{R}^{D} \to \mathbb{R}^{H x W x 3}
mean_F achieves O(1) space wrt N if X is an iterator that yields
\mathbb{R}^{H x W x 3}.
'''
mean_F = method(['self', 'X'])
F_inverse = method(['self', 'F', 'initial_image'], keywords='options')
class AttributeClassifier(interface):
r'''
lookup_scores: returns precomputed scores given a list of strings
score: \mathbb{R}^{N x H x W x 3} \to \mathbb{R}^{N x K}
fields: list of K strings
select: constraints is a list of 2-tuples (field index, value),
attributes is \mathbb{R}^{K}. Returns a list of image file paths which
satisfy the given constraints and are sorted by minkowski distance
(over some set of fields) to the given attribute vector.
score also accepts a list of N strings.
'''
lookup_scores = method(['self', 'X'], keywords='options')
score = method(['self', 'X'], keywords='options')
fields = method(['self'])
select = method(['self', 'constraints', 'attributes'], keywords='options')
class comestible(interface):
eat = method(['self'])
buy_from = method(['self', 'supermarket'])
mix_with = method(['self'], varargs='ingredients')
cook_with = method(['self'], keywords='ingredients')
cook = method(['self', 'temperature'], defaults=1)