|
1
|
|
|
from abc import ABC, abstractmethod |
|
2
|
|
|
import attr |
|
3
|
|
|
from so_magic.utils import SubclassRegistry |
|
4
|
|
|
|
|
5
|
|
|
|
|
6
|
|
|
class EncoderInterface(ABC): |
|
7
|
|
|
@abstractmethod |
|
8
|
|
|
def encode(self, *args, **kwargs): |
|
9
|
|
|
raise NotImplementedError |
|
10
|
|
|
|
|
11
|
|
|
|
|
12
|
|
|
class EncoderFactoryType(type): |
|
13
|
|
|
|
|
14
|
|
|
@classmethod |
|
15
|
|
|
def create(mcs, *args, **kwargs) -> EncoderInterface: |
|
16
|
|
|
raise NotImplementedError |
|
17
|
|
|
|
|
18
|
|
|
|
|
19
|
|
|
# class NominalVariableEncoderFactory: |
|
20
|
|
|
# @classmethod |
|
21
|
|
|
# def create(cls, *args, **kwargs) -> EncoderInterface: |
|
22
|
|
|
|
|
23
|
|
|
|
|
24
|
|
|
|
|
25
|
|
|
@attr.s(slots=True) |
|
26
|
|
|
class NominalAttributeEncoder(EncoderInterface, ABC): |
|
27
|
|
|
"""Encode the observations of a categorical nominal variable. |
|
28
|
|
|
|
|
29
|
|
|
The client code can supply the possible values for the nominal variable, if known a priori. |
|
30
|
|
|
The possible values are stored in the 'values_set' attribute/property. If they are not supplied |
|
31
|
|
|
they should be computed at runtime (when running the encode method). |
|
32
|
|
|
|
|
33
|
|
|
It also defines and stores the string identifiers for each column produced in the 'columns attribute/property. |
|
34
|
|
|
|
|
35
|
|
|
Args: |
|
36
|
|
|
values_set (list): the possible values of the nominal variable observations, if known a priori |
|
37
|
|
|
""" |
|
38
|
|
|
values_set: list = attr.ib(default=attr.Factory(list)) |
|
39
|
|
|
columns: list = attr.ib(init=False, default=attr.Factory(list)) |
|
40
|
|
|
|
|
41
|
|
|
|
|
42
|
|
|
class EncoderFactoryClassRegistry(metaclass=SubclassRegistry): pass |
|
43
|
|
|
|
|
44
|
|
|
from functools import reduce |
|
45
|
|
|
import pandas as pd |
|
46
|
|
|
|
|
47
|
|
|
|
|
48
|
|
|
@EncoderFactoryClassRegistry.register_as_subclass('nominal_list') |
|
49
|
|
|
class OneHotListEncoder(EncoderInterface): |
|
50
|
|
|
binary_transformer = {True: 1.0, False: 0.0} |
|
51
|
|
|
column_name_joiner = '_' |
|
52
|
|
|
def __init__(self, *args, **kwargs) -> None: |
|
53
|
|
|
pass |
|
54
|
|
|
def encode(self, *args, **kwargs): |
|
55
|
|
|
datapoints = args[0] |
|
56
|
|
|
attribute = args[1] |
|
57
|
|
|
print('ATRTRBUTE', attribute) |
|
58
|
|
|
print('STR', str(attribute)) |
|
59
|
|
|
cc = [_ for _ in datapoints.observations[str(attribute)]] |
|
60
|
|
|
print('LEN1', len(cc)) |
|
61
|
|
|
c = [_ for _ in cc if isinstance(_, list)] |
|
62
|
|
|
print('LEN2', len(c)) |
|
63
|
|
|
self.values_set = reduce(lambda i, j: set(i).union(set(j)), |
|
64
|
|
|
c) |
|
65
|
|
|
self.columns = sorted([f'{str(attribute)}{self.column_name_joiner}{x}' for x in self.values_set]) |
|
66
|
|
|
return pd.DataFrame([self._yield_vector(datarow, str(attribute)) for index, datarow in datapoints.iterrows()], |
|
67
|
|
|
columns=self.columns) |
|
68
|
|
|
|
|
69
|
|
|
def _yield_vector(self, datarow, attribute): |
|
70
|
|
|
decision = {True: self._encode, False: self._encode_none} |
|
71
|
|
|
return decision[isinstance(datarow[str(attribute)], list)](datarow, str(attribute)) |
|
72
|
|
|
|
|
73
|
|
|
def _encode(self, datarow, attribute): |
|
74
|
|
|
return [self.binary_transformer[column in datarow[str(attribute)]] for column in sorted(self.values_set)] |
|
75
|
|
|
|
|
76
|
|
|
def _encode_none(self, _datarow, _attribute): |
|
77
|
|
|
return [0.0] * len(self.values_set) |
|
78
|
|
|
|
|
79
|
|
|
def get_feature_names(self): |
|
80
|
|
|
return self.columns |
|
81
|
|
|
|
|
82
|
|
|
|
|
83
|
|
|
|
|
84
|
|
|
@EncoderFactoryClassRegistry.register_as_subclass('nominal_str') |
|
85
|
|
|
class OneHotStringEncoder(EncoderInterface): |
|
86
|
|
|
binary_transformer = {True: 1.0, False: 0.0} |
|
87
|
|
|
column_name_joiner = '_' |
|
88
|
|
|
def __init__(self, *args, **kwargs) -> None: |
|
89
|
|
|
pass |
|
90
|
|
|
def encode(self, *args, **kwargs): |
|
91
|
|
|
datapoints = args[0] |
|
92
|
|
|
attribute = args[1] |
|
93
|
|
|
print('ATRTRBUTE', attribute) |
|
94
|
|
|
print('STR', str(attribute)) |
|
95
|
|
|
c = [x for x in datapoints.observations[str(attribute)] if isinstance(x, str)] |
|
96
|
|
|
self.values_set = {value for value in c} |
|
97
|
|
|
self.columns = sorted([f'{str(attribute)}{self.column_name_joiner}{x}' for x in self.values_set]) |
|
98
|
|
|
return pd.DataFrame([self._yield_vector(datarow, str(attribute)) for index, datarow in datapoints.iterrows()], |
|
99
|
|
|
columns=self.columns) |
|
100
|
|
|
|
|
101
|
|
|
def _yield_vector(self, datarow, attribute): |
|
102
|
|
|
decision = {True: self._encode, False: self._encode_none} |
|
103
|
|
|
return decision[isinstance(datarow[str(attribute)], str)](datarow, str(attribute)) |
|
104
|
|
|
|
|
105
|
|
|
def _encode(self, datarow, attribute): |
|
106
|
|
|
return [self.binary_transformer[variable_value == datarow[str(attribute)]] for variable_value in sorted(self.values_set)] |
|
107
|
|
|
|
|
108
|
|
|
def _encode_none(self, _datarow, _attribute): |
|
109
|
|
|
return [0.0] * len(self.values_set) |
|
110
|
|
|
|
|
111
|
|
|
def get_feature_names(self): |
|
112
|
|
|
return self.columns |
|
113
|
|
|
|
|
114
|
|
|
|
|
115
|
|
|
@attr.s |
|
116
|
|
|
class EncoderFactory: |
|
117
|
|
|
encoder_factory_classes_registry = attr.ib(default=attr.Factory(lambda: EncoderFactoryClassRegistry)) |
|
118
|
|
|
def create(self, datapoints, variable, scheme='auto'): |
|
119
|
|
|
key = self.get_key(variable) |
|
120
|
|
|
return self.encoder_factory_classes_registry.create(key, datapoints, variable, scheme='auto') |
|
121
|
|
|
|
|
122
|
|
|
def get_key(self, variable): |
|
123
|
|
|
return f'{str(variable.type).lower()}_{str(variable.data_type.__name__)}' |
|
124
|
|
|
|
|
125
|
|
|
|
|
126
|
|
|
@attr.s |
|
127
|
|
|
class MagicEncoderFactory: |
|
128
|
|
|
encoder_factory = attr.ib(init=False, default=attr.Factory(lambda: EncoderFactory())) |
|
129
|
|
|
|
|
130
|
|
|
def create(self, datapoints, variable, scheme='auto'): |
|
131
|
|
|
return self.encoder_factory.create(datapoints, variable, scheme='auto') |
|
132
|
|
|
|
boromir674 /
so-magic
