data_frame 0.1.8

data/lib/data_frame/core/filter.rb

@@ -0,0 +1,48 @@
+module DF #:nodoc:
+  module Filter #:nodoc:
+    
+    # Takes a block to evaluate on each row.  The row can be converted into
+    # an OpenStruct or a Hash for easier filter methods. Note, don't try this
+    # with a hash or open struct unless you have facets available.
+    def filter!(as=Array, &block)
+      as = infer_class(as)
+      items = []
+      self.items.each do |row|
+        value = block.call(cast_row(row, as))
+        items << row if value
+      end
+      @items = items.dup
+      self
+    end
+
+    def filter(as=Array, &block)
+      new_data_frame = self.clone
+      new_data_frame.filter!(as, &block)
+    end
+
+    def filter_by_category(hash)
+      new_data_frame = self.dup
+      hash.each do |key, value|
+        key = key.to_underscore_sym
+        next unless self.labels.include?(key)
+        value = [value] unless value.is_a?(Array) or value.is_a?(Range)
+        new_data_frame.filter!(:hash) {|row| value.include?(row[key])}
+      end
+      new_data_frame
+    end
+
+    def filter_by_category!(hash)
+      hash.each do |key, value|
+        key = key.to_underscore_sym
+        next unless self.labels.include?(key)
+        value = [value] unless value.is_a?(Array) or value.is_a?(Range)
+        self.filter!(:hash) {|row| value.include?(row[key])}
+      end
+    end
+    
+  end
+end
+
+class DataFrame
+  include DF::Filter
+end

data/lib/data_frame/core/import.rb

@@ -0,0 +1,113 @@
+module DF #:nodoc:
+  module Import #:nodoc:
+    
+    module InferCSV #:nodoc:
+
+      protected
+        def default_csv_opts; {:converters => :all}; end
+
+        def infer_csv_contents(obj, opts={})
+          contents = File.read(obj) if File.exist?(obj)
+          begin
+            open(obj) {|f| contents = f.read} unless contents
+          rescue
+            nil
+          end
+          contents ||= obj if obj.is_a?(String)
+          return nil unless contents
+          table = FCSV.parse(contents, default_csv_opts.merge(opts))
+          labels = opts.fetch(:headers, true) ? table.shift : []
+          while table.last.empty?
+            table.pop
+          end
+          [labels, table]
+        end
+
+    end # InferCSV
+
+    module ClassMethods #:nodoc:
+
+      include InferCSV
+
+      # This is the neatest part of this neat gem.
+      # DataFrame.from_csv can be called in a lot of ways:
+      # DataFrame.from_csv(csv_contents)
+      # DataFrame.from_csv(filename)
+      # DataFrame.from_csv(url)
+      # If you need to define converters for FasterCSV, do it before calling
+      # this method: 
+      # FasterCSV::Converters[:special] = lambda{|f| f == 'foo' ? 'bar' : 'foo'}
+      # DataFrame.from_csv('http://example.com/my_special_url.csv', :converters => :special)
+      # This returns bar where 'foo' was found and 'foo' everywhere else.
+      def from_csv(obj, opts={})
+        labels, table = infer_csv_contents(obj, opts)
+        name = infer_name_from_contents(obj, opts)
+        return nil unless labels and table
+        df = new(*labels)
+        df.import(table)
+        df.name = name
+        df
+      end
+
+      protected
+
+        # Only works for names sources, urls and files
+        def infer_name_from_contents(obj, opts={})
+          begin
+            File.split(obj).last.split('.')[0..-2].join('.').titleize
+          rescue
+            nil
+          end
+        end
+
+    end # Class Methods
+    
+    module InstanceMethods #:nodoc:
+
+      include InferCSV
+      
+      def add_item(item)
+        self.items << item
+      end
+      alias :add :add_item
+
+      # Loads a batch of rows.  Expects an array of arrays, else you don't
+      # know what you have. 
+      def import(rows)
+        case rows
+        when Array
+          import_array(rows)
+        when String
+          labels, table = infer_csv_contents(rows, :headers => false)
+          import(table)
+        else
+          raise ArgumentError, "Don't know how to import data from #{rows.class}"
+        end
+        true
+      end
+      
+      protected
+        # Imports a table as an array of arrays.  
+        # If the array is one-dimensional and there is more than one label, it
+        # imports only one row. 
+        def import_array(rows)
+          raise ArgumentError, "Can only work with arrays" unless rows.is_a?(Array)
+          if self.labels.size > 1 and rows.dimensions == 1
+            self.add_item(rows)
+          else
+            # self.items = self.items + rows
+            rows.each do |row|
+              self.add_item(row)
+            end
+          end
+        end
+      
+    end # Instance Methods
+    
+  end
+end
+
+class DataFrame
+  include DF::Import::InstanceMethods
+  extend DF::Import::ClassMethods
+end

data/lib/data_frame/core/pre_process.rb

@@ -0,0 +1,69 @@
+module DF #:nodoc:
+  module PreProcess #:nodoc:
+    # A weird name.  This creates a column for every category in a column
+    # and marks each row by its value 
+    def j_binary_ize!(*columns)
+      # Allows to mix a hash with the columns.
+      options = columns.find_all {|e| e.is_a?(Hash)}.inject({}) {|h, e| h.merge!(e)}
+      columns.delete_if {|e| e.is_a?(Hash)}
+
+      # Generates new columns
+      columns.each do |col|
+        values = render_column(col.to_underscore_sym)
+        values.categories.each do |category|
+          full_name = (col.to_s + "_" + category.to_s).to_sym
+          if options[:allow_overlap]
+            category_map = values.inject([]) do |list, e|
+              list << values.all_categories(e)
+            end
+            self.append!(full_name, category_map.map{|e| e.include?(category)})
+          else
+            self.append!(full_name, values.category_map.map{|e| e == category})
+          end
+        end
+      end
+    end
+
+    # Adds a column, numerical_column_name that shows the same data as a
+    # nominal value, but as a number. 
+    def numericize!(*columns)
+      columns.each do |col|
+        values = render_column(col.to_underscore_sym)
+        categories = values.categories
+        value_categories = values.map {|v| values.category(v)}
+
+        i = 0
+        category_map = value_categories.uniq.inject({}) do |h, c|
+          h[c] = i
+          i += 1
+          h
+        end
+
+        blank = Array.new(category_map.size, 0)
+        reverse_category_map = category_map.inject({}) {|h, e| h[e.last] = e.first; h}
+
+        new_values = values.inject([]) do |list, val|
+          a = blank.dup
+          a[category_map[values.category(val)]] = 1
+          list << a
+        end
+
+        new_name = "numerical #{col.to_s}".to_underscore_sym
+        self.append!(new_name, new_values)
+      end
+    end
+    
+    def categorize!(*cs)
+      store_range_hashes
+      cs.each do |column|
+        self.replace!(column, category_map_from_stored_range_hash(column))
+      end
+      restore_range_hashes
+    end
+    
+  end
+end
+
+class DataFrame
+  include DF::PreProcess
+end

data/lib/data_frame/core/saving.rb

@@ -0,0 +1,29 @@
+module DF #:nodoc:
+  module Saving #:nodoc:
+
+    # Saves a data frame as CSV.  
+    # Examples:
+    # df.save('/tmp/some_filename.csv')
+    # df.save('/tmp/some_filename.csv', :include_header => false) # No header information is saved
+    # df.save('/tmp/some_filename.csv', :only => [:list, :of, :columns])
+    # df.save('/tmp/some_filename.csv', :subset => [:list, :of, :columns])
+    # df.save('/tmp/some_filename.csv', 
+    #   :filter => {:column_name => :category_value, 
+    #     :another_column_name => (range..values)}) # Filter by category
+    def save(filename, opts={})
+
+      df = self
+      df = df.subset_from_columns(*Array(opts[:only])) if opts[:only]
+      df = df.subset_from_columns(*Array(opts[:subset])) if opts[:subset]
+      df = df.filter_by_category(opts[:filter]) if opts[:filter]
+      df = df.filter_by_category(opts[:filter_by_category]) if opts[:filter_by_category]
+
+      File.open(filename, "w") { |f| f.write df.to_csv(opts.fetch(:include_header, true)) }
+    end
+
+  end
+end
+
+class DataFrame
+  include DF::Saving
+end

data/lib/data_frame/core/training.rb

@@ -0,0 +1,46 @@
+module Training #:nodoc:
+  
+  # Remove the training set if reset
+  # Return cached training_set, if there is one
+  # Get the proportion or 80%
+  # Get the number of items to choose, n, or a proportion of the items
+  # Store and return n random items
+  def training_set(opts={})
+    if opts[:reset]
+      @training_set = nil 
+      @test_set = nil
+    end
+    return @training_set if @training_set
+    
+    items_size = self.items.size
+    proportion = opts.fetch(:proportion, 0.8)
+    n = opts[:n]
+    n ||= (items_size * proportion).to_i
+    n = self.items.size if n > items_size
+    n = 0 if n < 0
+    
+    @training_set = []
+    while n > @training_set.size
+      @training_set << random_next(items_size) while n > @training_set.size
+      @training_set.uniq!
+    end
+    @training_set
+  end
+  
+  
+  def test_set(opts={})
+    @test_set = nil if opts[:reset]
+    return @test_set if @test_set
+    @test_set = self.items.exclusive_not(self.training_set)
+  end
+  
+  protected
+    def random_next(n)
+      self.items[rand(n)]
+    end
+  
+end
+
+class DataFrame
+  include Training
+end

data/lib/data_frame/data_frame.rb

@@ -0,0 +1,115 @@
+# This allows me to have named columns and optionally named rows in a
+# data frame, to work calculations (usually on the columns), to
+# transpose the matrix and store the transposed matrix until the object
+# is tainted. 
+class DataFrame
+  
+  def inspect
+    "DataFrame rows: #{self.rows.size} labels: #{self.labels.inspect}"
+  end
+  
+  # The labels of the data items
+  attr_reader :labels
+  alias :variables :labels
+  
+  # The items stored in the frame
+  attr_accessor :items
+  
+  # An optional name, useful for arff files
+  attr_accessor :name
+  
+  def initialize(*labels)
+    labels = labels.first if labels.size == 1 and labels.first.is_a?(Array)
+    @labels = labels.map {|e| e.to_underscore_sym }
+    @items = TransposableArray.new
+  end
+  
+  def row_labels
+    @row_labels ||= []
+  end
+  
+  def row_labels=(ary)
+    raise ArgumentError, "Row labels must be an array" unless ary.is_a?(Array)
+    @row_labels = ary
+  end
+  
+  # The rows as an array of arrays, an alias for items.
+  alias :rows :items
+  
+  def render_row(sym)
+    i = self.row_labels.index(sym)
+    return nil unless i
+    @items[i]
+  end
+  
+  # Return the column, given its name
+  def render_column(sym)
+    i = @labels.index(sym.to_underscore_sym)
+    return nil unless i
+    @items.transpose[i]
+  end
+  
+  # The columns as a Dictionary or Hash
+  # This is cached, call columns(true) to reset the cache.
+  def columns(reset=false)
+    @columns = nil if reset
+    return @columns if @columns
+    
+    container = defined?(Dictionary) ? Dictionary.new : Hash.new
+    i = 0
+    
+    @columns = @items.transpose.inject(container) do |cont, col|
+      cont[@labels[i]] = col
+      i += 1
+      cont
+    end
+  end
+  alias :to_hash :columns
+  alias :to_dictionary :columns
+  
+  def method_missing(sym, *args, &block)
+    if self.labels.include?(sym)
+      render_column(sym)
+    elsif self.row_labels.include?(sym)
+      render_row(sym)
+    elsif @items.respond_to?(sym)
+      @items.send(sym, *args, &block)
+    else
+      super
+    end
+  end
+  
+  protected
+  
+    def validate_column(column)
+      column = column.to_sym
+      raise ArgumentError, "Must provide the name of an existing column.  Provided #{column.inspect}, needed to provide one of #{self.labels.inspect}" unless self.labels.include?(column)
+      column
+    end
+    
+    def infer_class(obj)
+      obj = obj.to_s.classify.constantize if obj.is_a?(Symbol)
+      obj = obj.classify.constantize if obj.is_a?(String)
+      obj
+    end
+    
+    def cast_row(row, as)
+      if as == Hash
+        obj = {}
+        self.labels.each_with_index do |label, i|
+          obj[label] = row[i]
+        end
+        obj
+      elsif as == OpenStruct
+        obj = OpenStruct.new
+        self.labels.each_with_index do |label, i|
+          obj.table[label] = row[i]
+        end
+        obj
+      elsif as == Array
+        row
+      else
+        as.new(*row)
+      end
+    end
+end

data/lib/data_frame/id3.rb

@@ -0,0 +1,28 @@
+module DF #:nodoc:
+  # Uses Ilya Grigorik's ID3 decision_tree gem.  Installs it if you don't have it.
+  module ID3
+    begin
+      gem 'decisiontree'
+      require 'decisiontree'
+    rescue
+      `sudo gem install decisiontree`
+      gem 'decisiontree'
+      require 'decisiontree'
+    end
+
+    def create_id3(dependent_column, opts={})
+      # Need to put the dependent column in the last column
+      # Probably have other pre processing as well.
+      default = opts.fetch(:default, 1)
+      @id3 = DecisionTree::ID3Tree.new(self.labels, self.training_data, default, :discrete)
+      # ...
+    end
+
+    def id3
+    end
+  end
+end
+
+class DataFrame
+  include DF::ID3
+end