red-chainer 0.3.1 → 0.3.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 18e976d36f56e110baeae2d799471b4be2e4932778316ffcb3aafa51563bd3fd
-  data.tar.gz: f5429173d5b175738f4358774d9f058e31a86eec1418954cbad72d7875a2a7b8
+  metadata.gz: 33a95bf098a08c334e6a9d29ff791350e6ac0bd9de843054f3ed14f2a005b79b
+  data.tar.gz: 6f4e53b84e93d01e26363b5d43dba73ec4fc515ecbebbd1574ac6a864f52fb83
 SHA512:
-  metadata.gz: b1a617684703b92491e7876d919c2f6ef9c144039b9e12086d5b2ec7aed8644eca3ea8b2f4b32f2ef98fb6f31db4068b67467f95aa0a5868b3a903e073909b2e
-  data.tar.gz: '095870edbd1c68d35da023e9dd9a16096585a264f8e9cedf2c49dc465a62687a46634c58ee8a2d5fca4b2c41872f2f335455e076da66996a47f6c658ca6987e6'
+  metadata.gz: 40054365541bb8956c4a8211fbd489e75d1557a9e5c23bcb49f2cc2b393f0d96574d18259bad8e147525a3b720a5329f904be18c7c4f0891f1d886241b72a65d
+  data.tar.gz: fcb8e641d0efc1ffacc2014c2954d1c5a1e5f947fdcc18306f8946fb307c659e37bb3ccf08b9661d17e13184463b0282ae3594a6c252813d23a53fa0c6182a67

data/README.md

@@ -1,11 +1,14 @@
-# README
+# Red Chainer : A deep learning framework
 
-## Name
-
-Red Chainer
+A flexible framework for neural network for Ruby
 
 ## Description
-Welcome to your new gem! In this directory, you'll find the files you need to be able to package up your Ruby library into a gem. Put your Ruby code in the file `lib/chainer`. To experiment with that code, run `bin/console` for an interactive prompt.
+
+It ported python's [Chainer](https://github.com/chainer/chainer) with Ruby.
+
+## Requirements
+
+* Ruby 2.3 or later
 
 ## Installation
 
@@ -28,15 +31,27 @@ $ gem install red-chainer
 ```
 
 ## Usage
-mnist sample program is [here](./examples/mnist.rb)
+mnist sample program is [here](./examples/mnist/mnist.rb)
 
 ```bash
 # when install Gemfile
-$ bundle exec ruby examples/mnist.rb
+$ bundle exec ruby examples/mnist/mnist.rb
 # when install yourself
-$ ruby examples/mnist.rb
+$ ruby examples/mnist/mnist.rb
 ```
 
 ## License
 
 The MIT license. See [LICENSE.txt](./LICENSE.txt) for details.
+
+## Red Chainer implementation status
+
+|    |  Chainer 2.0<br>(Initial ported version)  | Red Chainer (0.3.1) | example |
+| ---- | ---- | ---- | ---- |
+|  [activation](https://github.com/red-data-tools/red-chainer/tree/master/lib/chainer/functions/activation)  |  15  | 5 | LogSoftmax, ReLU, LeakyReLU, Sigmoid, Tanh |
+|  [loss](https://github.com/red-data-tools/red-chainer/tree/master/lib/chainer/functions/loss)  |  17  | 2 | SoftMax, MeanSquaredError |
+|  [optimizer](https://github.com/red-data-tools/red-chainer/tree/master/lib/chainer/optimizers)  |  9  | 2 | Adam, MomentumSGDRule |
+|  [connection](https://github.com/red-data-tools/red-chainer/tree/master/lib/chainer/functions/connection)  |  12  | 2 | Linear, Convolution2D |
+|  [pooling](https://github.com/red-data-tools/red-chainer/tree/master/lib/chainer/functions/pooling)  |  14  | 3 | Pooling2D, MaxPooling2D, AveragePooling2D |
+|  [example](https://github.com/red-data-tools/red-chainer/tree/master/examples)  |  31  | 3 | MNIST, Iris, CIFAR |
+|  GPU  | use cupy  | ToDo | want to support [Cumo](https://github.com/sonots/cumo) |

data/examples/{iris.rb → iris/iris.rb}

@@ -31,10 +31,11 @@ optimizer = Chainer::Optimizers::Adam.new
 optimizer.setup(model)
 
 iris = Datasets::Iris.new
-x = iris.each.map {|r| r.each.to_a[0..3]}
+iris_table = iris.to_table
+x = iris_table.fetch_values(:sepal_length, :sepal_width, :petal_length, :petal_width).transpose
 
 # target
-y_class = iris.each.map {|r| r.class}
+y_class = iris_table[:class]
 
 # class index array
 # ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
@@ -45,17 +46,12 @@ y = y_class.map{|s|
 }
 
 # y_onehot => One-hot [[1.0, 0.0, 0.0], [1.0, 0.0, 0.0],,, [0.0, 1.0, 0.0], ,, [0.0, 0.0, 1.0]]
-y_onehot = y_class.map{|s|
-  i = class_name.index(s)
-  a = Array.new(class_name.size, 0.0)
-  a[i] = 1.0
-  a
-}
+y_onehot = Numo::SFloat.eye(class_name.size)[y,false]
 
 puts "Iris Datasets"
 puts "No. [sepal_length, sepal_width, petal_length, petal_width] one-hot #=> class"
 x.each_with_index{|r, i|
-  puts "#{'%3d' % i} : [#{r.join(', ')}] #{y_onehot[i]} #=> #{y_class[i]}(#{y[i]})"
+  puts "#{'%3d' % i} : [#{r.join(', ')}] #{y_onehot[i, false].to_a} #=> #{y_class[i]}(#{y[i]})"
 }
 # [5.1, 3.5, 1.4, 0.2, "Iris-setosa"]     => 50 data
 # [7.0, 3.2, 4.7, 1.4, "Iris-versicolor"] => 50 data
@@ -70,8 +66,13 @@ y_train = y_onehot[(1..-1).step(2), true] #=> 75 data (Iris-setosa : 25, Iris-ve
 x_test = x[(0..-1).step(2), true]         #=> 75 data (Iris-setosa : 25, Iris-versicolor : 25, Iris-virginica : 25)
 y_test = y[(0..-1).step(2)]               #=> 75 data (Iris-setosa : 25, Iris-versicolor : 25, Iris-virginica : 25)
 
+puts
+
 # Train
+print("Training ")
+
 10000.times{|i|
+  print(".") if i % 1000 == 0
   x = Chainer::Variable.new(x_train)
   y = Chainer::Variable.new(y_train)
   model.cleargrads()
@@ -80,6 +81,8 @@ y_test = y[(0..-1).step(2)]               #=> 75 data (Iris-setosa : 25, Iris-ve
   optimizer.update()
 }
 
+puts
+
 # Test
 xt = Chainer::Variable.new(x_test)
 yt = model.fwd(xt)

data/lib/chainer/datasets/cifar.rb

@@ -12,18 +12,17 @@ module Chainer
       end
 
       def self.get_cifar(n_classes, with_label, ndim, scale)
-        train_data = []
-        train_labels = []
-        ::Datasets::CIFAR.new(n_classes: n_classes, type: :train).each do |record|
-          train_data << record.pixels
-          train_labels << (n_classes == 10 ? record.label : record.fine_label)
-        end
-
-        test_data = []
-        test_labels = []
-        ::Datasets::CIFAR.new(n_classes: n_classes, type: :test).each do |record|
-          test_data << record.pixels
-          test_labels << (n_classes == 10 ? record.label : record.fine_label)
+        train_table = ::Datasets::CIFAR.new(n_classes: n_classes, type: :train).to_table
+        test_table = ::Datasets::CIFAR.new(n_classes: n_classes, type: :test).to_table
+
+        train_data = train_table[:pixels]
+        test_data = test_table[:pixels]
+        if n_classes == 10
+          train_labels = train_table[:label]
+          test_labels = test_table[:label]
+        else
+          train_labels = train_table[:fine_label]
+          test_labels = test_table[:fine_label]
         end
 
         [

data/lib/chainer/iterators/serial_iterator.rb

@@ -1,7 +1,7 @@
 module Chainer
   module Iterators
     class SerialIterator < Chainer::Dataset::Iterator 
-      attr_reader :epoch
+      attr_reader :epoch, :is_new_epoch
 
       def initialize(dataset, batch_size, repeat: true, shuffle: true)
         @dataset = dataset

data/lib/chainer/reporter.rb

@@ -6,6 +6,8 @@ module Chainer
   class Reporter
     include ReportService
 
+    attr_accessor :observer_names, :observation
+
     def initialize
       @observer_names = {}
       @observation = {}
@@ -16,6 +18,14 @@ module Chainer
       reporter.report(values, observer)
     end
 
+    def self.report_scope(observation)
+      current = @@reporters[-1]
+      old = current.observation
+      current.observation = observation
+      yield
+      current.observation = old
+    end
+
     def report(values, observer=nil)
       # TODO: keep_graph_on_report option
       if observer
@@ -37,6 +47,12 @@ module Chainer
       @observer_names[observer.object_id] = name
     end
 
+    def add_observers(prefix, observers, skipself: true)
+      observers.call(skipself: skipself) do |name, observer|
+        @observer_names[observer.object_id] = "#{prefix}#{name}"
+      end
+    end
+
     def scope(observation)
       @@reporters << self
       old = @observation

data/lib/chainer/training/extension.rb

@@ -5,7 +5,8 @@ module Chainer
       PRIORITY_EDITOR = 200
       PRIORITY_READER = 100
 
-      attr_accessor :name, :priority
+      attr_accessor :name
+      attr_writer :trigger, :priority
 
       def initialize
       end
@@ -14,7 +15,11 @@ module Chainer
       end
 
       def default_name
-        self.class.to_s
+        self.class.name.split('::').last
+      end
+
+      def trigger
+        @trigger || [1, 'iteration']
       end
 
       def priority

data/lib/chainer/training/extensions/evaluator.rb

@@ -1,9 +1,48 @@
 module Chainer
   module Training
     module Extensions
+      # Trainer extension to evaluate models on a validation set.
+      # This extension evaluates the current models by a given evaluation function.
+      #
+      # It creates a Chainer::Reporter object to store values observed in
+      # the evaluation function on each iteration. The report for all iterations
+      # are aggregated to Chainer::DictSummary. The collected mean values
+      # are further reported to the reporter object of the trainer, where the name
+      # of each observation is prefixed by the evaluator name. See
+      # Chainer::Reporter for details in naming rules of the reports.
+      #
+      # Evaluator has a structure to customize similar to that of Chainer::Training::StandardUpdater.
+      # The main differences are:
+      #
+      # - There are no optimizers in an evaluator. Instead, it holds links to evaluate.
+      # - An evaluation loop function is used instead of an update function.
+      # - Preparation routine can be customized, which is called before each evaluation.
+      #   It can be used, e.g., to initialize the state of stateful recurrent networks.
+      #
+      # There are two ways to modify the evaluation behavior besides setting a custom evaluation function.
+      # One is by setting a custom evaluation loop via  the `eval_func` argument.
+      # The other is by inheriting this class and overriding the `evaluate` method.
+      # In latter case, users have to create and handle a reporter object manually.
+      # Users also have to copy the iterators before using them, in order to reuse them at the next time of evaluation.
+      # In both cases, the functions are called in testing mode  (i.e., `chainer.config.train` is set to `false`).
+      #
+      # This extension is called at the end of each epoch by default.
       class Evaluator < Extension
+        # @param [Dataset::Iterator] iterator Dataset iterator for the validation dataset. It can also be a dictionary of iterators.
+        #                                     If this is just an iterator, the iterator is registered by the name 'main'.
+        # @param [Chainer::Link] target Link object or a dictionary of links to evaluate.
+        #                               If this is just a link object, the link is registered by the name 'main'.
+        # @param [Dataset::Convert] converter Converter function to build input arrays.
+        #             												`Chainer::Dataset.concat_examples` is used by default.
+        # @param [integer] device Device to which the training data is sent. Negative value indicates the host memory (CPU).
+        # @param [Function] eval_hook Function to prepare for each evaluation process.
+        #															It is called at the beginning of the evaluation.
+        # 														The evaluator extension object is passed at each call.
+        # @param [Function] eval_func Evaluation function called at each iteration.
+        #                             The target link to evaluate as a callable is used by default.
         def initialize(iterator, target, converter: nil, device: nil, eval_hook: nil, eval_func: nil)
           @priority = Extension::PRIORITY_WRITER
+          @trigger = [1, 'epoch']
 
           if iterator.kind_of?(Dataset::Iterator)
             iterator = { main: iterator }
@@ -20,6 +59,86 @@ module Chainer
           @eval_hook = eval_hook
           @eval_func = eval_func
         end
+
+        # Executes the evaluator extension.
+        #
+        # Unlike usual extensions, this extension can be executed without passing a trainer object.
+        # This extension reports the performance on validation dataset using the `Chainer.report` function.
+        # Thus, users can use  this extension independently from any trainer by manually configuring a `Chainer::Reporter` object.
+        #
+        # @param [Chainer::Training::Trainer] trainer Trainer object that invokes this extension.
+        #                                     It can be omitted in case of calling this extension manually.
+        def call(trainer = nil)
+          reporter = Reporter.new
+          prefix = self.respond_to?(:name) ? "#{self.name}/" : ""
+
+          @targets.each do |name, target|
+            reporter.add_observer("#{prefix}#{name}", target)
+            reporter.add_observers("#{prefix}#{name}", target.method(:namedlinks), skipself: true)
+          end
+
+          result = nil
+          reporter.scope(reporter.observation) do
+            old_train = Chainer.configuration.train
+            Chainer.configuration.train = false
+            result = evaluate()
+            Chainer.configuration.train = old_train
+          end
+
+          Reporter.save_report(result)
+          return result
+        end
+
+        # Evaluates the model and returns a result dictionary.
+        # This method runs the evaluation loop over the validation dataset.
+        # It accumulates the reported values to `DictSummary` and returns a dictionary whose values are means computed by the summary.
+        #
+        # Users can override this method to customize the evaluation routine.
+        # @return dict Result dictionary. This dictionary is further reported via `Chainer.save_report` without specifying any observer.
+        def evaluate
+          iterator = @iterators[:main]
+          target = @targets[:main]
+          eval_func = @eval_func || target
+
+          @eval_hook.(self) if @eval_hook
+
+          if iterator.respond_to?(:reset)
+            iterator.reset
+            it = iterator
+          else
+            it = iterator.dup
+          end
+
+          summary = DictSummary.new
+
+          until it.is_new_epoch do
+            batch = it.next
+            observation = {}
+            Reporter.report_scope(observation) do
+              in_arrays = @converter.(batch, device: @device)
+
+              old_enable_backprop = Chainer.configuration.enable_backprop
+              Chainer.configuration.enable_backprop = false
+
+              if in_arrays.kind_of?(Array)
+                eval_func.(*in_arrays)
+              elsif in_arrays.kind_of?(Hash)
+                eval_func.(**in_arrays)
+              else
+                eval_func.(in_arrays)
+              end
+
+              Chainer.configuration.enable_backprop = old_enable_backprop
+            end
+            summary.add(observation)
+          end
+
+          summary.compute_mean()
+        end
+
+        def default_name
+          "validation"
+        end
       end
     end
   end

data/lib/chainer/training/extensions/exponential_shift.rb

@@ -2,12 +2,12 @@ module Chainer
   module Training
     module Extensions
       # Trainer extension to exponentially shift an optimizer attribute.
-      # 
+      #
       # This extension exponentially increases or decreases the specified attribute of the optimizer.
       # The typical use case is an exponential decay of the learning rate.
       # This extension is also called before the training loop starts by default.
       class ExponentialShift < Extension
-        attr_reader :last_value        
+        attr_reader :last_value
 
         # @param [string] attr Name of the attribute to shift
         # @param [float] rate Rate of the exponential shift.
@@ -62,7 +62,7 @@ module Chainer
           end
         end
 
-        private 
+        private
 
         def get_optimizer(trainer)
           @optimizer || trainer.updater.get_optimizer(:main)

data/lib/chainer/training/extensions/log_report.rb

@@ -9,7 +9,7 @@ module Chainer
 
         def initialize(keys: nil, trigger: [1, 'epoch'], postprocess: nil, log_name: 'log')
           @keys = keys
-          @trigger = Chainer::Training::Util.get_trigger(trigger)
+          @_trigger = Chainer::Training::Util.get_trigger(trigger)
           @postprocess = postprocess
           @log_name = log_name
           @log = []
@@ -25,11 +25,11 @@ module Chainer
           else
             symbolized_observation = Hash[observation.map{|(k,v)| [k.to_sym,v]}]
             filterd_keys = @keys.select {|k| observation.keys.include?(k.to_sym) }
-            @summary.add(filterd_keys.each_with_object({}) {|k, hash| hash[k.to_s] = observation[k.to_sym] })            
+            @summary.add(filterd_keys.each_with_object({}) {|k, hash| hash[k.to_s] = observation[k.to_sym] })
           end
 
-          # if trigger is true, output the result
-          return unless @trigger.(trainer)
+          # if @_trigger is true, output the result
+          return unless @_trigger.(trainer)
 
           stats = @summary.compute_mean
           stats_cpu = {}
@@ -41,9 +41,9 @@ module Chainer
           stats_cpu['epoch'] = updater.epoch
           stats_cpu['iteration'] = updater.iteration
           stats_cpu['elapsed_time'] = trainer.elapsed_time
-        
+
           @postprocess.(stats_cpu) unless @postprocess.nil?
-          
+
           @log << stats_cpu
 
           unless @log_name.nil?
@@ -62,8 +62,8 @@ module Chainer
         end
 
         def serialize(serializer)
-          if @trigger.respond_to?(:serialize)
-            @trigger.serialize(serializer['_trigger'])
+          if @_trigger.respond_to?(:serialize)
+            @_trigger.serialize(serializer['_trigger'])
           end
           # Note that this serialization may lose some information of small
           # numerical differences.

data/lib/chainer/training/extensions/print_report.rb

@@ -2,7 +2,7 @@ module Chainer
   module Training
     module Extensions
       class PrintReport < Extension
-        def initialize(entries, log_report: 'Chainer::Training::Extensions::LogReport', out: STDOUT)
+        def initialize(entries, log_report: 'LogReport', out: STDOUT)
           @entries = entries
           @log_report = log_report
           @out = out
@@ -27,7 +27,7 @@ module Chainer
             @out.write(@header)
             @header = nil
           end
-         
+
           if @log_report.is_a?(String)
             log_report = trainer.get_extension(@log_report)
           elsif @log_report.is_a?(LogReport)

data/lib/chainer/training/extensions/progress_bar.rb

@@ -14,7 +14,7 @@ module Chainer
           @recent_timing = []
         end
 
-        def call(trainer) 
+        def call(trainer)
           if @training_length.nil?
             t = trainer.stop_trigger
             raise TypeError, "cannot retrieve the training length #{t.class}" unless t.is_a?(Chainer::Training::Triggers::IntervalTrigger)
@@ -27,7 +27,7 @@ module Chainer
 
           length, unit = @training_length
           iteration = trainer.updater.iteration
-          
+
           # print the progress bar according to interval
           return unless iteration % @update_interval == 0
 
@@ -69,7 +69,7 @@ module Chainer
           else
             estimated_time = (length - epoch) / speed_e
           end
-          
+
           @out.write(sprintf("%10.5g iters/sec. Estimated time to finish: %s.\n", speed_t, (Time.parse("1991/01/01") + (estimated_time)).strftime("%H:%m:%S")))
 
           # move the cursor to the head of the progress bar

data/lib/chainer/training/extensions/snapshot.rb

@@ -13,6 +13,8 @@ module Chainer
         end
 
         def initialize(save_class: nil, filename_proc: nil, target: nil)
+          @priority = -100
+          @trigger = [1, 'epoch']
           @save_class = save_class || Chainer::Serializers::MarshalSerializer
           @filename_proc = filename_proc || Proc.new { |trainer| "snapshot_iter_#{trainer.updater.iteration}" }
           @target = target

data/lib/chainer/version.rb

@@ -1,4 +1,4 @@
 module Chainer
-  VERSION = "0.3.1"
+  VERSION = "0.3.2"
 end
 

data/red-chainer.gemspec

@@ -20,7 +20,7 @@ Gem::Specification.new do |spec|
   spec.require_paths = ["lib"]
 
   spec.add_runtime_dependency "numo-narray", ">= 0.9.1.1"
-  spec.add_runtime_dependency "red-datasets"
+  spec.add_runtime_dependency "red-datasets", ">= 0.0.5"
 
   spec.add_development_dependency "bundler", "~> 1.15"
   spec.add_development_dependency "rake", "~> 10.0"

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: red-chainer
 version: !ruby/object:Gem::Version
-  version: 0.3.1
+  version: 0.3.2
 platform: ruby
 authors:
 - Yusaku Hatanaka
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2018-05-29 00:00:00.000000000 Z
+date: 2018-06-27 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: numo-narray
@@ -30,14 +30,14 @@ dependencies:
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: '0'
+        version: 0.0.5
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: '0'
+        version: 0.0.5
 - !ruby/object:Gem::Dependency
   name: bundler
   requirement: !ruby/object:Gem::Requirement
@@ -99,8 +99,8 @@ files:
 - examples/cifar/models/resnet18.rb
 - examples/cifar/models/vgg.rb
 - examples/cifar/train_cifar.rb
-- examples/iris.rb
-- examples/mnist.rb
+- examples/iris/iris.rb
+- examples/mnist/mnist.rb
 - lib/chainer.rb
 - lib/chainer/configuration.rb
 - lib/chainer/cuda.rb