• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Copyright 2020 Huawei Technologies Co., Ltd
2#
3# Licensed under the Apache License, Version 2.0 (the "License");
4# you may not use this file except in compliance with the License.
5# You may obtain a copy of the License at
6#
7# http://www.apache.org/licenses/LICENSE-2.0
8#
9# Unless required by applicable law or agreed to in writing, software
10# distributed under the License is distributed on an "AS IS" BASIS,
11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12# See the License for the specific language governing permissions and
13# limitations under the License.
14# ============================================================================
15"""Scalar Affine Bijector"""
16from mindspore.ops import operations as P
17from ..distribution._utils.custom_ops import log_generic
18from .bijector import Bijector
19
20
21class ScalarAffine(Bijector):
22    """
23    Scalar Affine Bijector.
24    This Bijector performs the operation:
25
26    .. math::
27        Y = a * X + b
28
29    where a is the scale factor and b is the shift factor.
30
31    Args:
32        scale (float, list, numpy.ndarray, Tensor): The scale factor. Default: 1.0.
33        shift (float, list, numpy.ndarray, Tensor): The shift factor. Default: 0.0.
34        name (str): The name of the bijector. Default: 'ScalarAffine'.
35
36    Supported Platforms:
37        ``Ascend`` ``GPU``
38
39    Note:
40        The dtype of `shift` and `scale` must be float.
41        If `shift`, `scale` are passed in as numpy.ndarray or tensor, they have to have
42        the same dtype otherwise an error will be raised.
43
44    Raises:
45        TypeError: When the dtype of `shift` or `scale` is not float,
46                   and when the dtype of `shift` and `scale` is not same.
47
48    Examples:
49        >>> import mindspore
50        >>> import mindspore.nn as nn
51        >>> from mindspore import Tensor
52        >>>
53        >>> # To initialize a ScalarAffine bijector of scale 1.0 and shift 2.
54        >>> scalaraffine = nn.probability.bijector.ScalarAffine(1.0, 2.0)
55        >>> value = Tensor([1, 2, 3], dtype=mindspore.float32)
56        >>> ans1 = scalaraffine.forward(value)
57        >>> print(ans1.shape)
58        (3,)
59        >>> ans2 = scalaraffine.inverse(value)
60        >>> print(ans2.shape)
61        (3,)
62        >>> ans3 = scalaraffine.forward_log_jacobian(value)
63        >>> print(ans3.shape)
64        ()
65        >>> ans4 = scalaraffine.inverse_log_jacobian(value)
66        >>> print(ans4.shape)
67        ()
68    """
69
70    def __init__(self,
71                 scale=1.0,
72                 shift=0.0,
73                 name='ScalarAffine'):
74        """
75        Constructor of ScalarAffine Bijector.
76        """
77        param = dict(locals())
78        param['param_dict'] = {'scale': scale, 'shift': shift}
79        super(ScalarAffine, self).__init__(
80            is_constant_jacobian=True,
81            is_injective=True,
82            name=name,
83            dtype=None,
84            param=param)
85
86        self._scale = self._add_parameter(scale, 'scale')
87        self._shift = self._add_parameter(shift, 'shift')
88
89        self.abs = P.Abs()
90        self.oneslike = P.OnesLike()
91        self.dtypeop = P.DType()
92        self.cast = P.Cast()
93        self.log = log_generic
94
95    @property
96    def scale(self):
97        return self._scale
98
99    @property
100    def shift(self):
101        return self._shift
102
103    def extend_repr(self):
104        """Display instance object as string."""
105        if self.is_scalar_batch:
106            str_info = 'scale = {}, shift = {}'.format(self.scale, self.shift)
107        else:
108            str_info = 'batch_shape = {}'.format(self.batch_shape)
109        return str_info
110
111    def _forward(self, x):
112        r"""
113        .. math::
114            f(x) = a * x + b
115        """
116        x = self._check_value_dtype(x)
117        scale_local = self.cast_param_by_value(x, self.scale)
118        shift_local = self.cast_param_by_value(x, self.shift)
119        forward_v = scale_local * x + shift_local * self.oneslike(x)
120        return forward_v
121
122    def _inverse(self, y):
123        r"""
124        .. math::
125            f(y) = \frac{y - b}{a}
126        """
127        y = self._check_value_dtype(y)
128        scale_local = self.cast_param_by_value(y, self.scale)
129        shift_local = self.cast_param_by_value(y, self.shift)
130        inverse_v = (y - shift_local) / scale_local
131        return inverse_v
132
133    def _forward_log_jacobian(self, x):
134        r"""
135        .. math::
136            f(x) = a * x + b
137            f'(x) = a
138            \log(f'(x)) = \log(a)
139        """
140        x = self._check_value_dtype(x)
141        scale_local = self.cast_param_by_value(x, self.scale)
142        forward_log_j = self.log(self.abs(scale_local))
143        return forward_log_j
144
145    def _inverse_log_jacobian(self, y):
146        r"""
147        .. math::
148            f(y) = \frac{(y - b)}{a}
149            f'(x) = \frac{1.0}{a}
150            \log(f'(x)) = - \log(a)
151        """
152        y = self._check_value_dtype(y)
153        scale_local = self.cast_param_by_value(y, self.scale)
154        inverse_log_j = -1. * self.log(self.abs(scale_local))
155        return inverse_log_j
156