forked from Tensor-Array/Tensor-Array
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathnormalization.cc
More file actions
77 lines (72 loc) · 3.63 KB
/
normalization.cc
File metadata and controls
77 lines (72 loc) · 3.63 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
/*
Copyright 2024 TensorArray-Creators
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
#include "normalization.hh"
namespace tensor_array
{
namespace layers
{
void NormalizationImpl::layer_init(std::vector<std::pair<std::initializer_list<unsigned int>, const std::type_info&>>&& shape_input)
{
std::vector<unsigned int> temp = shape_input[0].first;
const std::type_info& type = shape_input[0].second;
for (auto& it : dims_mean)
temp[it] = 1U;
if (!this->gamma.has_tensor())
this->gamma = value::values(temp, 1).tensor_cast(type);
if (!this->beta.has_tensor())
this->beta = value::values(temp, 0).tensor_cast(type);
if (!this->moving_mean.has_tensor())
this->moving_mean = value::values(temp, 0).tensor_cast(type);
if (!this->moving_variance.has_tensor())
this->moving_variance = value::values(temp, 1).tensor_cast(type);
this->moving_mean = this->moving_mean.clone();
this->moving_variance = this->moving_variance.clone();
}
value::Tensor NormalizationImpl::calculate(const value::Tensor& input)
{
value::Tensor normal;
if (tensor_array::value::use_grad)
{
value::Tensor temp_mean = input.mean(this->dims_mean);
value::Tensor temp_variance = input.variance(this->dims_mean);
normal = (input - temp_mean) / power(add(temp_variance, value::values(input.get_buffer().shape(), this->eps)), value::values(input.get_buffer().shape(), .5f));
this->moving_mean *= value::values(this->moving_mean.get_buffer().shape(), 1.f - momentum);
this->moving_mean += (value::values(temp_mean.get_buffer().shape(), momentum) * temp_mean);
this->moving_mean = this->moving_mean.clone();
this->moving_variance *= value::values(this->moving_variance.get_buffer().shape(), 1.f - momentum);
this->moving_variance += (value::values(temp_variance.get_buffer().shape(), momentum) * temp_variance);
this->moving_variance = this->moving_variance.clone();
}
else
normal = (input - this->moving_mean) / power(add(this->moving_variance, value::values(input.get_buffer().shape(), this->eps)), value::values(input.get_buffer().shape(), .5f)).clone();
return gamma * normal + beta;
}
NormalizationImpl::NormalizationImpl(const std::initializer_list<unsigned char>& dims_mean, float eps, float momentum) :
eps(eps),
momentum(momentum),
dims_mean(dims_mean),
moving_mean(),
moving_variance(),
gamma(),
beta()
{
this->map_tensor.insert(std::make_pair("gamma", &gamma));
this->map_tensor.insert(std::make_pair("beta", &beta));
this->map_tensor.insert(std::make_pair("moving_mean", &moving_mean));
this->map_tensor.insert(std::make_pair("moving_variance", &moving_variance));
}
NormalizationImpl::~NormalizationImpl()
{
}
}
}