|
1
|
|
|
from collections import OrderedDict |
|
2
|
|
|
from typing import Any, Tuple, Union |
|
3
|
|
|
|
|
4
|
|
|
import numpy as np |
|
5
|
|
|
from typish import SubscriptableType, Literal, ClsFunction, EllipsisType |
|
6
|
|
|
|
|
7
|
|
|
_Size = Union[int, Literal[Any]] # TODO add type vars as well |
|
8
|
|
|
_Type = Union[type, Literal[Any], np.dtype] |
|
9
|
|
|
_NSizes = Tuple[_Size, EllipsisType] |
|
10
|
|
|
_SizeAndType = Tuple[_Size, _Type] |
|
11
|
|
|
_Sizes = Tuple[_Size, ...] |
|
12
|
|
|
_SizesAndType = Tuple[Tuple[_Size, ...], _Type] |
|
13
|
|
|
_NSizesAndType = Tuple[_NSizes, _Type] |
|
14
|
|
|
_Default = Tuple[Tuple[Literal[Any], EllipsisType], Literal[Any]] |
|
15
|
|
|
|
|
16
|
|
|
|
|
17
|
|
View Code Duplication |
class _NDArrayMeta(SubscriptableType): |
|
|
|
|
|
|
18
|
|
|
_shape = tuple() # Overridden by _NDArray._shape. |
|
19
|
|
|
_type = ... # Overridden by _NDArray._type. |
|
20
|
|
|
|
|
21
|
|
|
@property |
|
22
|
|
|
def dtype(cls) -> np.dtype: |
|
23
|
|
|
""" |
|
24
|
|
|
Return the numpy dtype. |
|
25
|
|
|
:return: the numpy dtype. |
|
26
|
|
|
""" |
|
27
|
|
|
return np.dtype(cls._type) # TODO if type is Any, this wont work |
|
28
|
|
|
|
|
29
|
|
|
@property |
|
30
|
|
|
def shape(cls) -> Tuple[int, int]: |
|
31
|
|
|
""" |
|
32
|
|
|
Return the shape as a tuple of ints. |
|
33
|
|
|
:return: the shape as a tuple of ints. |
|
34
|
|
|
""" |
|
35
|
|
|
return cls._shape |
|
36
|
|
|
|
|
37
|
|
|
def __repr__(cls): |
|
38
|
|
|
shape_ = cls._shape |
|
39
|
|
|
if len(cls._shape) == 2 and cls._shape[1] is ...: |
|
40
|
|
|
shape_ = (cls._shape[0], '...') |
|
41
|
|
|
|
|
42
|
|
|
type_ = getattr(cls._type, '__name__', cls._type) |
|
43
|
|
|
return 'NDArray[{}, {}]'.format(shape_, type_).replace('\'', '') |
|
44
|
|
|
|
|
45
|
|
|
def __str__(cls): |
|
46
|
|
|
return repr(cls) |
|
47
|
|
|
|
|
48
|
|
|
def __eq__(cls, other) -> bool: |
|
49
|
|
|
return (isinstance(other, _NDArrayMeta) |
|
50
|
|
|
and cls._shape == other._shape |
|
51
|
|
|
and cls._type == other._type) |
|
52
|
|
|
|
|
53
|
|
|
def __instancecheck__(cls, instance: np.ndarray) -> bool: |
|
54
|
|
|
""" |
|
55
|
|
|
Checks whether the given instance conforms the current NDArray type by |
|
56
|
|
|
checking the shape and the dtype. |
|
57
|
|
|
:param instance: a numpy.ndarray. |
|
58
|
|
|
:return: True if instance is an instance of cls. |
|
59
|
|
|
""" |
|
60
|
|
|
return (isinstance(instance, np.ndarray) |
|
61
|
|
|
and _NDArrayMeta._is_shape_eq(cls, instance) |
|
62
|
|
|
and _NDArrayMeta._is_type_eq(cls, instance)) |
|
63
|
|
|
|
|
64
|
|
|
def _is_shape_eq(cls, instance: np.ndarray) -> bool: |
|
65
|
|
|
|
|
66
|
|
|
def _is_eq_to(this: Any, that: Any) -> bool: |
|
67
|
|
|
return that is Any or this == that |
|
68
|
|
|
|
|
69
|
|
|
if cls._shape == (Any, ...): |
|
70
|
|
|
return True |
|
71
|
|
|
if len(cls._shape) == 2 and cls._shape[1] is ...: |
|
72
|
|
|
size = cls._shape[0] |
|
73
|
|
|
return all([s == size for s in instance.shape]) |
|
74
|
|
|
if len(instance.shape) != len(cls._shape): |
|
75
|
|
|
return False |
|
76
|
|
|
zipped = zip(instance.shape, cls._shape) |
|
77
|
|
|
return all([_is_eq_to(a, b) for a, b in zipped]) |
|
78
|
|
|
|
|
79
|
|
|
def _is_type_eq(cls, instance: np.ndarray) -> bool: |
|
80
|
|
|
if cls._type is Any: |
|
81
|
|
|
return True |
|
82
|
|
|
return cls.dtype == instance.dtype |
|
83
|
|
|
|
|
84
|
|
|
|
|
85
|
|
View Code Duplication |
class _NDArray(metaclass=_NDArrayMeta): |
|
|
|
|
|
|
86
|
|
|
_shape = (Any, ...) |
|
87
|
|
|
_type = Any |
|
88
|
|
|
|
|
89
|
|
|
@classmethod |
|
90
|
|
|
def _after_subscription(cls, item: Any) -> None: |
|
91
|
|
|
method = ClsFunction(OrderedDict([ |
|
92
|
|
|
(_Size, cls._only_size), |
|
93
|
|
|
(_Type, cls._only_type), |
|
94
|
|
|
(_NSizes, lambda _: ...), |
|
95
|
|
|
(_SizeAndType, cls._size_and_type), |
|
96
|
|
|
(_Sizes, cls._only_sizes), |
|
97
|
|
|
(_SizesAndType, cls._sizes_and_type), |
|
98
|
|
|
(_NSizesAndType, cls._sizes_and_type), |
|
99
|
|
|
(_Default, lambda _: ...), |
|
100
|
|
|
])) |
|
101
|
|
|
|
|
102
|
|
|
if not method.understands(item): |
|
103
|
|
|
raise TypeError('Invalid parameter for NDArray: "{}"'.format(item)) |
|
104
|
|
|
return method(item) |
|
105
|
|
|
|
|
106
|
|
|
@classmethod |
|
107
|
|
|
def _only_size(cls, item: int): |
|
108
|
|
|
# E.g. NDArray[3] |
|
109
|
|
|
# The given item is the size of the single dimension. |
|
110
|
|
|
cls._shape = (item,) |
|
111
|
|
|
|
|
112
|
|
|
@classmethod |
|
113
|
|
|
def _only_type(cls, item: type): |
|
114
|
|
|
# E.g. NDArray[int] |
|
115
|
|
|
# The given item is the type of the single dimension. |
|
116
|
|
|
cls._type = item |
|
117
|
|
|
|
|
118
|
|
|
@classmethod |
|
119
|
|
|
def _size_and_type(cls, item: Tuple[_Size, _Type]): |
|
120
|
|
|
# E.g. NDArray[3, int] |
|
121
|
|
|
# The given item is the size of the single dimension and its type. |
|
122
|
|
|
cls._shape = (item[0],) |
|
123
|
|
|
cls._type = item[1] |
|
124
|
|
|
|
|
125
|
|
|
@classmethod |
|
126
|
|
|
def _only_sizes(cls, item: Tuple[_Size, ...]): |
|
127
|
|
|
# E.g. NDArray[(2, Any, 2)] |
|
128
|
|
|
# The given item is a tuple with just sizes of the dimensions. |
|
129
|
|
|
cls._shape = item |
|
130
|
|
|
|
|
131
|
|
|
@classmethod |
|
132
|
|
|
def _sizes_and_type(cls, item: Tuple[Tuple[_Size, ...], _Type]): |
|
133
|
|
|
# E.g. NDArray[(2, Any, 2), int] |
|
134
|
|
|
# The given item is a tuple with sizes of the dimensions and the type. |
|
135
|
|
|
# Or e.g. NDArray[(3, ...), int] |
|
136
|
|
|
# The given item is a tuple with sizes of n dimensions and the type. |
|
137
|
|
|
cls._only_sizes(item[0]) |
|
138
|
|
|
cls._only_type(item[1]) |
|
139
|
|
|
|
ramonhagenaars /
nptyping
Loading history...