|
1
|
|
|
from abc import ABC |
|
2
|
|
|
import attr |
|
3
|
|
|
from so_magic.utils import SubclassRegistry |
|
4
|
|
|
|
|
5
|
|
|
|
|
6
|
|
|
class DiscretizerInterface(ABC): |
|
7
|
|
|
def discretize(self, *args, **kwargs): |
|
8
|
|
|
raise NotImplementedError |
|
9
|
|
|
|
|
10
|
|
|
|
|
11
|
|
|
class AbstractDiscretizer(DiscretizerInterface): |
|
12
|
|
|
def discretize(self, *args, **kwargs): |
|
13
|
|
|
raise NotImplementedError |
|
14
|
|
|
|
|
15
|
|
|
|
|
16
|
|
|
def validate_bin_function(_self, _attribute, value): |
|
17
|
|
|
if not callable(value): |
|
18
|
|
|
raise ValueError(f'Expected a callable object, instead a {type(value).__name__} was given.') |
|
19
|
|
|
|
|
20
|
|
|
@attr.s |
|
21
|
|
|
class BaseDiscretizer(AbstractDiscretizer): |
|
22
|
|
|
binner = attr.ib(init=True, validator=validate_bin_function) |
|
23
|
|
|
|
|
24
|
|
|
def discretize(self, *args, **kwargs): |
|
25
|
|
|
"""Expects args: dataset, feature and kwargs; 'nb_bins'.""" |
|
26
|
|
|
dataset, feature, nb_bins = args[0], args[1], args[2] |
|
27
|
|
|
return self.binner(feature.values(dataset), nb_bins) |
|
28
|
|
|
|
|
29
|
|
|
|
|
30
|
|
|
@attr.s |
|
31
|
|
|
class FeatureDiscretizer(BaseDiscretizer): |
|
32
|
|
|
feature = attr.ib(init=True) |
|
33
|
|
|
|
|
34
|
|
|
def discretize(self, *args, **kwargs): |
|
35
|
|
|
"""Expects args: dataset, nb_bins.""" |
|
36
|
|
|
return super().discretize(args[0], self.feature, args[1]) |
|
37
|
|
|
|
|
38
|
|
|
@attr.s |
|
39
|
|
|
class FeatureDiscretizerFactory: |
|
40
|
|
|
binner_factory = attr.ib(init=True) |
|
41
|
|
|
|
|
42
|
|
|
def categorical(self, feature, **kwargs) -> FeatureDiscretizer: |
|
43
|
|
|
binner_type = 'same-length' |
|
44
|
|
|
if kwargs.get('quantisized', False): |
|
45
|
|
|
binner_type = 'quantisized' |
|
46
|
|
|
return FeatureDiscretizer(self.binner_factory.create_binner(binner_type), feature) |
|
47
|
|
|
|
|
48
|
|
|
def numerical(self, feature, **kwargs) -> FeatureDiscretizer: |
|
49
|
|
|
binner_type = 'same-length' |
|
50
|
|
|
if kwargs.get('quantisized', False): |
|
51
|
|
|
binner_type = 'quantisized' |
|
52
|
|
|
return FeatureDiscretizer(self.binner_factory.create_binner(binner_type), feature) |
|
53
|
|
|
|
|
54
|
|
|
|
|
55
|
|
|
######################################### |
|
56
|
|
|
|
|
57
|
|
|
class BinnerInterface(ABC): |
|
58
|
|
|
def bin(self, values, nb_bins): |
|
59
|
|
|
raise NotImplementedError |
|
60
|
|
|
|
|
61
|
|
|
|
|
62
|
|
|
class BaseBinner(BinnerInterface): |
|
63
|
|
|
|
|
64
|
|
|
def bin(self, values, nb_bins): |
|
65
|
|
|
"""It is assumed numerical (ratio or interval) variable or ordinal (not nominal) categorical variable.""" |
|
66
|
|
|
raise NotImplementedError |
|
67
|
|
|
|
|
68
|
|
|
|
|
69
|
|
|
class BinnerClass(metaclass=SubclassRegistry): pass |
|
70
|
|
|
|
|
71
|
|
|
|
|
72
|
|
|
class BinnerFactory: |
|
73
|
|
|
parent_class = BinnerClass |
|
74
|
|
|
|
|
75
|
|
|
def equal_length_binner(self, *args, **kwargs) -> BaseBinner: |
|
76
|
|
|
raise NotImplementedError |
|
77
|
|
|
|
|
78
|
|
|
def quantisized_binner(self, *args, **kwargs) -> BaseBinner: |
|
79
|
|
|
raise NotImplementedError |
|
80
|
|
|
|
|
81
|
|
|
def create_binner(self, *args, **kwargs) -> BaseBinner: |
|
82
|
|
|
raise NotImplementedError |
|
83
|
|
|
|
boromir674 /
so-magic
