Class: MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies::SNES

Inherits:
Base
  • Object
show all
Defined in:
lib/machine_learning_workbench/optimizer/natural_evolution_strategies/snes.rb

Overview

Separable Natural Evolution Strategies

Instance Attribute Summary collapse

Attributes inherited from Base

#best, #eye, #last_fits, #mu, #ndims, #obj_fn, #opt_type, #parallel_fit, #rescale_lrate, #rescale_popsize, #rng, #sigma

Instance Method Summary collapse

Methods inherited from Base

#cmaes_lrate, #cmaes_popsize, #cmaes_utilities, #initialize, #interface_methods, #lrate, #move_inds, #popsize, #sorted_inds, #standard_normal_sample, #standard_normal_samples, #utils

Constructor Details

This class inherits a constructor from MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies::Base

Instance Attribute Details

#variances ⇒ Object (readonly)

Returns the value of attribute variances.



7
8
9
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/snes.rb', line 7

def variances
  @variances
end

Instance Method Details

#convergence ⇒ Object

Estimate algorithm convergence as total variance



41
42
43
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/snes.rb', line 41

def convergence
  variances.sum
end

#initialize_distribution(mu_init: 0, sigma_init: 1) ⇒ Object 



9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/snes.rb', line 9

def initialize_distribution mu_init: 0, sigma_init: 1
  @mu = case mu_init
    when Array
      raise ArgumentError unless mu_init.size == ndims
      NArray[mu_init]
    when Numeric
      NArray.new([1,ndims]).fill mu_init
    else
      raise ArgumentError, "Something is wrong with mu_init: #{mu_init}"
  end
  @variances = case sigma_init
  when Array
    raise ArgumentError unless sigma_init.size == ndims
    NArray[*sigma_init]
  when Numeric
    NArray.new([ndims]).fill(sigma_init)
  else
    raise ArgumentError, "Something is wrong with sigma_init: #{sigma_init}" \
      "(did you remember to copy the other cases from XNES?)"
  end
  @sigma = @variances.diag
end

#load(data) ⇒ Object

Raises:

  • (ArgumentError) 


49
50
51
52
53
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/snes.rb', line 49

def load data
  raise ArgumentError unless data.size == 2
  @mu, @variances = data.map &:to_na
  @sigma = variances.diag
end

#save ⇒ Object 



45
46
47
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/snes.rb', line 45

def save
  [mu.to_a, variances.to_a]
end

#train(picks: sorted_inds) ⇒ Object 



32
33
34
35
36
37
38
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/snes.rb', line 32

def train picks: sorted_inds
  g_mu = utils.dot(picks)
  g_sigma = utils.dot(picks**2 - 1)
  @mu += sigma.dot(g_mu.transpose).transpose * lrate
  @variances *= (g_sigma * lrate / 2).exponential.flatten
  @sigma = @variances.diag
end