Class: DNN::Optimizers::RMSProp
- Defined in:
- lib/dnn/core/optimizers.rb
Instance Attribute Summary collapse
-
#muse ⇒ Object
Returns the value of attribute muse.
Attributes inherited from Optimizer
Class Method Summary collapse
Instance Method Summary collapse
-
#initialize(learning_rate = 0.001, muse = 0.9) ⇒ RMSProp
constructor
A new instance of RMSProp.
- #to_hash ⇒ Object
- #update(layer) ⇒ Object
Constructor Details
#initialize(learning_rate = 0.001, muse = 0.9) ⇒ RMSProp
Returns a new instance of RMSProp.
81 82 83 84 85 |
# File 'lib/dnn/core/optimizers.rb', line 81 def initialize(learning_rate = 0.001, muse = 0.9) super(learning_rate) @muse = muse @g = {} end |
Instance Attribute Details
#muse ⇒ Object
Returns the value of attribute muse.
75 76 77 |
# File 'lib/dnn/core/optimizers.rb', line 75 def muse @muse end |
Class Method Details
.load_hash(hash) ⇒ Object
77 78 79 |
# File 'lib/dnn/core/optimizers.rb', line 77 def self.load_hash(hash) self.new(hash[:learning_rate], hash[:muse]) end |
Instance Method Details
#to_hash ⇒ Object
96 97 98 |
# File 'lib/dnn/core/optimizers.rb', line 96 def to_hash super({muse: @muse}) end |
#update(layer) ⇒ Object
87 88 89 90 91 92 93 94 |
# File 'lib/dnn/core/optimizers.rb', line 87 def update(layer) @g[layer] ||= {} layer.params.each_key do |key| @g[layer][key] ||= 0 @g[layer][key] = @muse * @g[layer][key] + (1 - @muse) * layer.grads[key]**2 layer.params[key] -= (@learning_rate / NMath.sqrt(@g[layer][key] + 1e-7)) * layer.grads[key] end end |