machine_learning_workbench 0.6.1 → 0.7.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: aa6944de3c6c7e7ef318e456aaf2479ff577773b
-  data.tar.gz: dce74b8e349c4f15e6a65c32805e8fa6f7e95253
+  metadata.gz: 8897ba173dbfa944cf55b3ca7b57eb3af87bbff7
+  data.tar.gz: 44883310f216b187d5d3ccce669e85f946e6ee5f
 SHA512:
-  metadata.gz: 86858d8e37e499ad296476a92afbaf59ecee6edf9c499c8d1e786ec25d39d676f5852f9faa7f14ed5afeac8f9b8d92a3fd478adb88d85d5d03acb3ca9ac804c6
-  data.tar.gz: 5bd9161716409f4470d4b5d11aad78c9cf907c2acde8486928d3622f175c20f7fb608e27d33b7ede9fcb2ab9339fc95d8d0f6f4d6940c029793e7a854cfa770d
+  metadata.gz: 75d8a1f4d2087746dae316ca47c07925858826bdf393eddde7bb2f82e22b47e2d9c2c6bbaa6e0ce10fea1ccb0b9df1882e80b58519a2107f59732d6f99ea1a76
+  data.tar.gz: d4945335adc99edaabd26b56ac7bb49936d0642d58a09516ac63fae27e11d871db879e12db9cc3d0ec78788363c85c29558d382deb145713fa4051985e330f28

data/.gitignore

@@ -7,6 +7,6 @@
 /spec/reports/
 /tmp/
 Gemfile.lock
-
+/stats/
 # rspec failure tracking
 .rspec_status

data/Rakefile

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 require "bundler/gem_tasks"
 require "rspec/core/rake_task"
 

data/examples/image_compression.rb

@@ -1,3 +1,7 @@
+# frozen_string_literal: true
+
+# Run as: `bundle exec ruby examples/image_compression.rb`
+
 require 'rmagick'
 require 'machine_learning_workbench'
 VectorQuantization = MachineLearningWorkbench::Compressor::VectorQuantization

data/examples/neuroevolution.rb

@@ -1,3 +1,7 @@
+# frozen_string_literal: true
+
+# Run as: `bundle exec ruby examples/neuroevolution.rb`
+
 # Make sure the gem is installed first with `gem install machine_learning_workbench`
 # Alternatively, add `gem 'machine_learning_workbench'` to your Gemfile if using Bundle,
 # followed by a `bundle install`

data/lib/machine_learning_workbench.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 gpu = false             # prepare for switching to GPUs
 if gpu

data/lib/machine_learning_workbench/compressor.rb

@@ -1,3 +1,6 @@
+# frozen_string_literal: true
+
 require_relative 'compressor/vector_quantization'
 require_relative 'compressor/decaying_learning_rate_vq'
 require_relative 'compressor/copy_vq'
+require_relative 'compressor/incr_dict_vq'

data/lib/machine_learning_workbench/compressor/copy_vq.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 module MachineLearningWorkbench::Compressor
   # Train-less VQ, copying new images into centroids
   # Optimized for online training.

data/lib/machine_learning_workbench/compressor/decaying_learning_rate_vq.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 module MachineLearningWorkbench::Compressor
   # VQ with per-centroid decaying learning rates.
   # Optimized for online training.

data/lib/machine_learning_workbench/compressor/incr_dict_vq.rb

@@ -0,0 +1,45 @@
+# frozen_string_literal: true
+
+module MachineLearningWorkbench::Compressor
+  # Incremental Dictionary Train-less VQ, creating new centroids rather than training
+  # Optimized for online training.
+  # TODO: as the deadline grows nigh, the hacks grow foul. Refactor all VQs together.
+  class IncrDictVQ < VectorQuantization
+
+    attr_reader :equal_simil
+    undef :ntrains # centroids are not trained
+
+    def initialize **opts
+      puts "Ignoring learning rate: `lrate: #{opts[:lrate]}`" if opts[:lrate]
+      puts "Ignoring similarity: `simil_type: #{opts[:simil_type]}`" if opts[:simil_type]
+      puts "Ignoring ncentrs: `ncentrs: #{opts[:ncentrs]}`" if opts[:ncentrs]
+      # TODO: try different epsilons to reduce the number of states
+      # for example, in qbert we care what is lit and what is not, not the colors
+      @equal_simil = opts.delete(:equal_simil) || 0.0
+      super **opts.merge({ncentrs: 1, lrate: nil, simil_type: nil})
+      @ntrains = nil # will disable the counting
+    end
+
+    # Overloading lrate check from original VQ
+    def check_lrate lrate; nil; end
+
+    # Train on one vector:
+    # - train only if the image is not already in dictionary
+    # - create new centroid from the image
+    # @return [Integer] index of new centroid
+    def train_one vec, eps: equal_simil
+      mses = centrs.map do |centr|
+        ((centr-vec)**2).sum / centr.size # uhm get rid of division maybe? squares?
+      end
+      min_mse = mses.min
+      # skip training if the centr with smallest mse (most similar) has less than eps error (equal)
+      # TODO: maintain an average somewhere, make eps dynamic
+      return if min_mse < eps
+      puts "Creating centr #{ncentrs} (min_mse: #{min_mse})"
+      centrs << vec
+      @utility = @utility.concatenate 0
+      @ncentrs.tap{ @ncentrs += 1}
+    end
+
+  end
+end

data/lib/machine_learning_workbench/compressor/vector_quantization.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 module MachineLearningWorkbench::Compressor
 
   # Standard Vector Quantization
@@ -15,7 +17,7 @@ module MachineLearningWorkbench::Compressor
       check_lrate lrate # hack: so that we can overload it in dlr_vq
       @lrate = lrate
       @simil_type = simil_type || :dot
-      @encoding_type = encoding_type || :ensemble_norm
+      @encoding_type = encoding_type || :norm_ensemble
       @init_centr_vrange ||= vrange
       @vrange = case vrange
         when Array
@@ -80,6 +82,12 @@ module MachineLearningWorkbench::Compressor
         @ncodes += 1
         @utility[code] += 1
         code
+      when :most_similar_ary
+        code = simils.new_zeros
+        code[simils.max_index] = 1
+        @ncodes += 1
+        @utility += code
+        code
       when :ensemble
         code = simils
         tot = simils.sum
@@ -88,10 +96,17 @@ module MachineLearningWorkbench::Compressor
         @ncodes += 1
         @utility += (contrib - utility) / ncodes # cumulative moving average
         code
-      when :ensemble_norm
+      when :norm_ensemble
         tot = simils.sum
         tot = 1 if tot < 1e-5  # HACK: avoid division by zero
         code = simils / tot
+        @ncodes += 1
+        @utility += (code - utility) / ncodes # cumulative moving average
+        code
+      when :sparse_coding
+        raise NotImplementedError, "do this next"
+
+
         @ncodes += 1
         @utility += (code - utility) / ncodes # cumulative moving average
         code
@@ -104,11 +119,15 @@ module MachineLearningWorkbench::Compressor
       case type
       when :most_similar
         centrs[code]
+      when :most_similar_ary
+        centrs[code.eq(1).where[0]]
       when :ensemble
         tot = code.reduce :+
         centrs.zip(code).map { |centr, contr| centr*contr/tot }.reduce :+
-      when :ensemble_norm
+      when :norm_ensemble
         centrs.zip(code).map { |centr, contr| centr*contr }.reduce :+
+      when :sparse_coding
+        raise NotImplementedError, "do this next"
       else raise ArgumentError, "unrecognized reconstruction type: #{type}"
       end
     end
@@ -148,7 +167,7 @@ module MachineLearningWorkbench::Compressor
       vec_lst.each_with_index do |vec, i|
         trained_idx = train_one vec
         print '.' if debug
-        @ntrains[trained_idx] += 1
+        @ntrains[trained_idx] += 1 if @ntrains
       end
     end
   end

data/lib/machine_learning_workbench/monkey.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 # Monkey patches
 

data/lib/machine_learning_workbench/neural_network.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 require_relative 'neural_network/base'
 require_relative 'neural_network/feed_forward'
 require_relative 'neural_network/recurrent'

data/lib/machine_learning_workbench/neural_network/base.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 module MachineLearningWorkbench::NeuralNetwork
   # Neural Network base class
@@ -26,10 +27,10 @@ module MachineLearningWorkbench::NeuralNetwork
 
     # @param struct [Array<Integer>] list of layer sizes
     # @param act_fn [Symbol] choice of activation function for the neurons
-    def initialize struct, act_fn: nil
+    def initialize struct, act_fn: nil, **act_fn_args
       @struct = struct
       @act_fn_name = act_fn || :sigmoid
-      @act_fn = send(act_fn_name)
+      @act_fn = send act_fn_name, **act_fn_args
       # @state holds both inputs, possibly recurrency, and bias
       # it is a complete input for the next layer, hence size from layer sizes
       @state = layer_row_sizes.collect do |size|
@@ -163,10 +164,10 @@ module MachineLearningWorkbench::NeuralNetwork
     ## Activation functions
 
     # Traditional sigmoid (logistic) with variable steepness
-    def sigmoid k=1
-      # k is steepness:  0<k<1 is flatter, 1<k is flatter
+    def sigmoid steepness: 1
+      # steepness:  0<s<1 is flatter, 1<s is flatter
       # flatter makes activation less sensitive, better with large number of inputs
-      -> (vec) { 1.0 / (NMath.exp(-k * vec) + 1.0) }
+      -> (vec) { 1.0 / (NMath.exp(-steepness * vec) + 1.0) }
     end
     alias logistic sigmoid
 

data/lib/machine_learning_workbench/neural_network/feed_forward.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 module MachineLearningWorkbench::NeuralNetwork
   # Feed Forward Neural Network

data/lib/machine_learning_workbench/neural_network/recurrent.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 module MachineLearningWorkbench::NeuralNetwork
   # Recurrent Neural Network

data/lib/machine_learning_workbench/optimizer.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 module MachineLearningWorkbench::Optimizer
 end
 

data/lib/machine_learning_workbench/optimizer/natural_evolution_strategies/base.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
   # Natural Evolution Strategies base class
@@ -19,11 +20,15 @@ module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
     #   a single instance.
     # @param rescale_popsize [Float] scaling for the default population size
     # @param rescale_lrate [Float] scaling for the default learning rate
-    def initialize ndims, obj_fn, opt_type, rseed: nil, mu_init: 0, sigma_init: 1, parallel_fit: false, rescale_popsize: 1, rescale_lrate: 1
-      raise ArgumentError unless [:min, :max].include? opt_type
-      raise ArgumentError unless obj_fn.respond_to? :call
+    def initialize ndims, obj_fn, opt_type, rseed: nil, mu_init: 0, sigma_init: 1, parallel_fit: false, rescale_popsize: 1, rescale_lrate: 1, utilities: nil, popsize: nil, lrate: nil
+      raise ArgumentError, "opt_type: #{opt_type}" unless [:min, :max].include? opt_type
+      raise ArgumentError, "obj_fn not callable: #{obj_fn}" unless obj_fn.respond_to? :call
+      raise ArgumentError, "utilities only if popsize" if utilities && popsize.nil?
+      raise ArgumentError, "wrong sizes" if utilities && utilities.size != popsize
+      raise ArgumentError, "minimum popsize 5 for default utilities" if popsize&.<(5) && utilities.nil?
       @ndims, @opt_type, @obj_fn, @parallel_fit = ndims, opt_type, obj_fn, parallel_fit
-      @rescale_popsize, @rescale_lrate = rescale_popsize, rescale_lrate
+      @rescale_popsize, @rescale_lrate = rescale_popsize, rescale_lrate # rescale defaults
+      @utilities, @popsize, @lrate = utilities, popsize, lrate # if not set, defaults below
       @eye = NArray.eye(ndims)
       rseed ||= Random.new_seed
       # puts "NES rseed: #{s}"  # currently disabled
@@ -44,18 +49,20 @@ module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
     end
 
     # Memoized automatic magic numbers
+    # Initialization options allow to rescale or entirely override these.
     # NOTE: Doubling popsize and halving lrate often helps
     def utils;   @utilities ||= cmaes_utilities end
     # (see #utils)
-    def popsize; @popsize   ||= cmaes_popsize * rescale_popsize end
+    def popsize; @popsize   ||= Integer(cmaes_popsize * rescale_popsize) end
     # (see #utils)
     def lrate;   @lrate     ||= cmaes_lrate * rescale_lrate end
 
-    # Magic numbers from CMA-ES (TODO: add proper citation)
+    # Magic numbers from CMA-ES (see `README` for citation)
     # @return [NArray] scale-invariant utilities
     def cmaes_utilities
       # Algorithm equations are meant for fitness maximization
       # Match utilities with individuals sorted by INCREASING fitness
+      raise ArgumentError, "Minimum `popsize` should be 5 (is #{popsize})" if popsize < 5
       log_range = (1..popsize).collect do |v|
         [0, Math.log(popsize.to_f/2 - 1) - Math.log(v)].max
       end

data/lib/machine_learning_workbench/optimizer/natural_evolution_strategies/bdnes.rb

@@ -1,11 +1,13 @@
+# frozen_string_literal: true
 
 module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
   # Block-Diagonal Natural Evolution Strategies
   class BDNES < Base
 
-    MAX_RSEED = 10**Random.new_seed.size # same range as Random.new_seed
+    MAX_RSEED = 10**Random.new_seed.size # block random seeds to be on the same range as `Random.new_seed`
 
-    attr_reader :ndims_lst, :blocks, :popsize
+    attr_reader :ndims_lst, :blocks, :popsize, :parallel_update
+    undef :ndims # only `ndims_lst` here
 
     # Initialize a list of XNES, one for each block
     # see class `Base` for the description of the rest of the arguments.
@@ -13,7 +15,8 @@ module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
     #    matrix. Note: entire (reconstructed) individuals will be passed to the `obj_fn`
     #    regardless of the division here described.
     # @param init_opts [Hash] the rest of the options will be passed directly to XNES
-    def initialize ndims_lst, obj_fn, opt_type, parallel_fit: false, rseed: nil, **init_opts
+    # @parellel_update [bool] whether to parallelize block updates
+    def initialize ndims_lst, obj_fn, opt_type, parallel_fit: false, rseed: nil, parallel_update: false, **init_opts
       # mu_init: 0, sigma_init: 1
       # init_opts = {rseed: rseed, mu_init: mu_init, sigma_init: sigma_init}
       # TODO: accept list of `mu_init`s and `sigma_init`s
@@ -21,9 +24,8 @@ module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
       block_fit = -> (*args) { raise "Should never be called" }
       # the BD-NES seed should ensure deterministic reproducibility
       # but each block should have a different seed
-      rseed ||= Random.new_seed
       # puts "BD-NES rseed: #{s}"  # currently disabled
-      @rng = Random.new rseed
+      @rng = Random.new rseed || Random.new_seed
       @blocks = ndims_lst.map do |ndims|
         b_rseed = rng.rand MAX_RSEED
         XNES.new ndims, block_fit, opt_type, rseed: b_rseed, **init_opts
@@ -34,6 +36,8 @@ module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
 
       @best = [(opt_type==:max ? -1 : 1) * Float::INFINITY, nil]
       @last_fits = []
+      @parallel_update = parallel_update
+      require 'parallel' if parallel_update
     end
 
     def sorted_inds_lst
@@ -82,9 +86,22 @@ module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
 
     # duck-type the interface: [:train, :mu, :convergence, :save, :load]
 
+    # TODO: refactor DRY
     def train picks: sorted_inds_lst
-      blocks.zip(sorted_inds_lst).each do |xnes, s_inds|
-        xnes.train picks: s_inds
+      if parallel_update
+        # Parallel.each(blocks.zip(picks)) do |xnes, s_inds|
+        #   xnes.train picks: s_inds
+        # end
+        # Actually it's not this simple.
+        # Forks do not act on the parent, so I need to send back updated mu and sigma
+        # Luckily we have `NES#save` and `NES#load` at the ready
+        # Next: need to implement `#marshal_dump` and `#marshal_load` in `Base`
+        # Actually using `Cumo` rather than `Parallel` may avoid marshaling altogether
+        raise NotImplementedError, "Should dump and load each instance"
+      else
+        blocks.zip(picks).each do |xnes, s_inds|
+          xnes.train picks: s_inds
+        end
       end
     end
 
@@ -92,6 +109,10 @@ module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
       blocks.map(&:mu).reduce { |mem, var| mem.concatenate var, axis: 1 }
     end
 
+    def sigma
+      raise NotImplementedError, "need to write a concatenation like for mu here"
+    end
+
     def convergence
       blocks.map(&:convergence).reduce(:+)
     end

data/lib/machine_learning_workbench/optimizer/natural_evolution_strategies/fnes.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
   # Fixed Variance Natural Evolution Strategies

data/lib/machine_learning_workbench/optimizer/natural_evolution_strategies/rnes.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
   # Radial Natural Evolution Strategies

data/lib/machine_learning_workbench/optimizer/natural_evolution_strategies/snes.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
   # Separable Natural Evolution Strategies

data/lib/machine_learning_workbench/optimizer/natural_evolution_strategies/xnes.rb

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
   # Exponential Natural Evolution Strategies
@@ -11,17 +12,23 @@ module MachineLearningWorkbench::Optimizer::NaturalEvolutionStrategies
           NArray[mu_init]
         when Numeric
           NArray.new([1,ndims]).fill mu_init
+        when NArray
+          raise ArgumentError unless mu_init.size == ndims
+          mu_init.ndim < 2 ? mu_init.reshape(1, ndims) : mu_init
         else
           raise ArgumentError, "Something is wrong with mu_init: #{mu_init}"
       end
       @sigma = case sigma_init
-      when Array
-        raise ArgumentError unless sigma_init.size == ndims
-        NArray[*sigma_init].diag
-      when Numeric
-        NArray.new([ndims]).fill(sigma_init).diag
-      else
-        raise ArgumentError, "Something is wrong with sigma_init: #{sigma_init}"
+        when Array
+          raise ArgumentError unless sigma_init.size == ndims
+          NArray[*sigma_init].diag
+        when Numeric
+          NArray.new([ndims]).fill(sigma_init).diag
+        when NArray
+          raise ArgumentError unless sigma_init.size == ndims**2
+          sigma_init.ndim < 2 ? sigma_init.reshape(ndims, ndims) : sigma_init
+        else
+          raise ArgumentError, "Something is wrong with sigma_init: #{sigma_init}"
       end
       # Works with the log of sigma to avoid continuous decompositions (thanks Sun Yi)
       @log_sigma = NMath.log(sigma.diagonal).diag

data/lib/machine_learning_workbench/systems.rb

@@ -1 +1,3 @@
+# frozen_string_literal: true
+
 require_relative 'systems/neuroevolution'

data/lib/machine_learning_workbench/systems/neuroevolution.rb

@@ -1,2 +1,3 @@
+# frozen_string_literal: true
 
 "Work in progress"

data/lib/machine_learning_workbench/tools.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 require_relative 'tools/execution'
 require_relative 'tools/normalization'
 require_relative 'tools/imaging'

data/lib/machine_learning_workbench/tools/execution.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 module MachineLearningWorkbench::Tools
   module Execution
     $fork_pids ||= []

data/lib/machine_learning_workbench/tools/imaging.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 module MachineLearningWorkbench::Tools
   module Imaging
     Forkable = MachineLearningWorkbench::Tools::Execution

data/lib/machine_learning_workbench/tools/normalization.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 module MachineLearningWorkbench::Tools
   module Normalization
     def self.feature_scaling narr, from: nil, to: [0,1]

data/lib/machine_learning_workbench/tools/verification.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 module MachineLearningWorkbench::Tools
   module Verification
     # TODO: switch to NArray

data/machine_learning_workbench.gemspec

@@ -1,3 +1,4 @@
+# frozen_string_literal: true
 
 lib = File.expand_path("../lib", __FILE__)
 $LOAD_PATH.unshift(lib) unless $LOAD_PATH.include?(lib)

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: machine_learning_workbench
 version: !ruby/object:Gem::Version
-  version: 0.6.1
+  version: 0.7.0
 platform: ruby
 authors:
 - Giuseppe Cuccu
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2018-04-12 00:00:00.000000000 Z
+date: 2018-04-20 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: bundler
@@ -204,6 +204,7 @@ files:
 - lib/machine_learning_workbench/compressor.rb
 - lib/machine_learning_workbench/compressor/copy_vq.rb
 - lib/machine_learning_workbench/compressor/decaying_learning_rate_vq.rb
+- lib/machine_learning_workbench/compressor/incr_dict_vq.rb
 - lib/machine_learning_workbench/compressor/vector_quantization.rb
 - lib/machine_learning_workbench/monkey.rb
 - lib/machine_learning_workbench/neural_network.rb