Class: MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies::XNES

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

Overview

Exponential Natural Evolution Strategies

Instance Attribute Summary collapse

Attributes inherited from Base

#best, #id, #last_fits, #mu, #ndims, #obj_fn, #opt_type, #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

#log_sigma ⇒ Object (readonly)

Returns the value of attribute log_sigma.



5
6
7
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/xnes.rb', line 5

def log_sigma
  @log_sigma
end

Instance Method Details

#convergence ⇒ Object

Estimate algorithm convergence as total variance



30
31
32
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/xnes.rb', line 30

def convergence
  sigma.trace
end

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



7
8
9
10
11
12
13
14
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/xnes.rb', line 7

def initialize_distribution mu_init: 0, sigma_init: 1
  @mu = NMatrix.new([1, ndims], mu_init, dtype: :float64)
  sigma_init = [sigma_init]*ndims unless sigma_init.kind_of? Enumerable
  @sigma = NMatrix.diag(sigma_init, dtype: :float64)
  # Works with the log of sigma to avoid continuous decompositions (thanks Sun Yi)
  log_sigma_init = sigma_init.map &Math.method(:log)
  @log_sigma = NMatrix.diag(log_sigma_init, dtype: :float64)
end

#load(data) ⇒ Object

Raises:

  • (ArgumentError) 


38
39
40
41
42
43
44
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/xnes.rb', line 38

def load data
  raise ArgumentError unless data.size == 2
  mu_ary, log_sigma_ary = data
  @mu = NMatrix[*mu_ary, dtype: :float64]
  @log_sigma = NMatrix[*log_sigma_ary, dtype: :float64]
  @sigma = log_sigma.exponential
end

#save ⇒ Object 



34
35
36
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/xnes.rb', line 34

def save
  [mu.to_consistent_a, log_sigma.to_consistent_a]
end

#train(picks: sorted_inds) ⇒ Object 



16
17
18
19
20
21
22
23
24
25
26
27
 
# File 'lib/machine_learning_workbench/optimizer/natural_evolution_strategies/xnes.rb', line 16

def train picks: sorted_inds
  g_mu = utils.dot(picks)
  g_log_sigma = popsize.times.inject(NMatrix.zeros_like sigma) do |sum, i|
    u = utils[i]
    ind = picks.row(i)
    ind_sq = ind.outer_flat(ind, &:*)
    sum + (ind_sq - id) * u
  end
  @mu += sigma.dot(g_mu.transpose).transpose * lrate
  @log_sigma += g_log_sigma * (lrate/2)
  @sigma = log_sigma.exponential
end