Class: DNN::Layers::BatchNormalization

Inherits:
HasParamLayer show all
Defined in:
lib/dnn/core/normalizations.rb

Instance Attribute Summary collapse

Attributes inherited from HasParamLayer

#trainable

Attributes inherited from Layer

#input_shape, #name

Instance Method Summary collapse

Methods inherited from Layer

#built?, #call, call, from_hash, #output_shape

Constructor Details

#initialize(axis: 0, momentum: 0.9, eps: 1e-7) ⇒ BatchNormalization 



16
17
18
19
20
21
 
# File 'lib/dnn/core/normalizations.rb', line 16

def initialize(axis: 0, momentum: 0.9, eps: 1e-7)
  super()
  @axis = axis
  @momentum = momentum
  @eps = eps
end

Instance Attribute Details

#axis ⇒ Object (readonly)

Returns the value of attribute axis.



9
10
11
 
# File 'lib/dnn/core/normalizations.rb', line 9

def axis
  @axis
end

#beta ⇒ Object (readonly)

Returns the value of attribute beta.



6
7
8
 
# File 'lib/dnn/core/normalizations.rb', line 6

def beta
  @beta
end

#eps ⇒ Object

Returns the value of attribute eps.



11
12
13
 
# File 'lib/dnn/core/normalizations.rb', line 11

def eps
  @eps
end

#gamma ⇒ Object (readonly)

Returns the value of attribute gamma.



5
6
7
 
# File 'lib/dnn/core/normalizations.rb', line 5

def gamma
  @gamma
end

#momentum ⇒ Object

Returns the value of attribute momentum.



10
11
12
 
# File 'lib/dnn/core/normalizations.rb', line 10

def momentum
  @momentum
end

#running_mean ⇒ Object (readonly)

Returns the value of attribute running_mean.



7
8
9
 
# File 'lib/dnn/core/normalizations.rb', line 7

def running_mean
  @running_mean
end

#running_var ⇒ Object (readonly)

Returns the value of attribute running_var.



8
9
10
 
# File 'lib/dnn/core/normalizations.rb', line 8

def running_var
  @running_var
end

Instance Method Details

#backward(dy) ⇒ Object 



48
49
50
51
52
53
54
55
56
57
58
59
60
61
 
# File 'lib/dnn/core/normalizations.rb', line 48

def backward(dy)
  batch_size = dy.shape[@axis]
  if @trainable
    @beta.grad = dy.sum(axis: @axis, keepdims: true)
    @gamma.grad = (@xn * dy).sum(axis: @axis, keepdims: true)
  end
  dxn = @gamma.data * dy
  dxc = dxn / @std
  dstd = -((dxn * @xc) / (@std**2)).sum(axis: @axis, keepdims: true)
  dvar = 0.5 * dstd / @std
  dxc += (2.0 / batch_size) * @xc * dvar
  dmean = dxc.sum(axis: @axis, keepdims: true)
  dxc - dmean / batch_size
end

#build(input_shape) ⇒ Object 



23
24
25
26
27
28
29
 
# File 'lib/dnn/core/normalizations.rb', line 23

def build(input_shape)
  super
  @gamma = Param.new(Xumo::SFloat.ones(*output_shape), Xumo::SFloat[0])
  @beta = Param.new(Xumo::SFloat.zeros(*output_shape), Xumo::SFloat[0])
  @running_mean = Param.new(Xumo::SFloat.zeros(*output_shape))
  @running_var = Param.new(Xumo::SFloat.zeros(*output_shape))
end

#forward(x) ⇒ Object 



31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
 
# File 'lib/dnn/core/normalizations.rb', line 31

def forward(x)
  if DNN.learning_phase
    mean = x.mean(axis: @axis, keepdims: true)
    @xc = x - mean
    var = (@xc**2).mean(axis: @axis, keepdims: true)
    @std = Xumo::NMath.sqrt(var + @eps)
    xn = @xc / @std
    @xn = xn
    @running_mean.data = @momentum * @running_mean.data + (1 - @momentum) * mean
    @running_var.data = @momentum * @running_var.data + (1 - @momentum) * var
  else
    xc = x - @running_mean.data
    xn = xc / Xumo::NMath.sqrt(@running_var.data + @eps)
  end
  @gamma.data * xn + @beta.data
end

#get_params ⇒ Object 



71
72
73
 
# File 'lib/dnn/core/normalizations.rb', line 71

def get_params
  { gamma: @gamma, beta: @beta, running_mean: @running_mean, running_var: @running_var }
end

#load_hash(hash) ⇒ Object 



67
68
69
 
# File 'lib/dnn/core/normalizations.rb', line 67

def load_hash(hash)
  initialize(axis: hash[:axis], momentum: hash[:momentum])
end

#to_hash ⇒ Object 



63
64
65
 
# File 'lib/dnn/core/normalizations.rb', line 63

def to_hash
  super(axis: @axis, momentum: @momentum, eps: @eps)
end