data_frame 0.1.8
Sign up to get free protection for your applications and to get access to all the features.
- data/README.rdoc +122 -0
- data/VERSION.yml +4 -0
- data/bin/plain_frame +22 -0
- data/lib/data_frame.rb +26 -0
- data/lib/data_frame/arff.rb +52 -0
- data/lib/data_frame/callback_array.rb +152 -0
- data/lib/data_frame/core/column_management.rb +147 -0
- data/lib/data_frame/core/filter.rb +48 -0
- data/lib/data_frame/core/import.rb +113 -0
- data/lib/data_frame/core/pre_process.rb +69 -0
- data/lib/data_frame/core/saving.rb +29 -0
- data/lib/data_frame/core/training.rb +46 -0
- data/lib/data_frame/data_frame.rb +115 -0
- data/lib/data_frame/id3.rb +28 -0
- data/lib/data_frame/kmeans.rb +10 -0
- data/lib/data_frame/labels_from_uci.rb +48 -0
- data/lib/data_frame/mlp.rb +18 -0
- data/lib/data_frame/model.rb +22 -0
- data/lib/data_frame/parameter_capture.rb +50 -0
- data/lib/data_frame/sbn.rb +18 -0
- data/lib/data_frame/transposable_array.rb +23 -0
- data/lib/ext/array.rb +11 -0
- data/lib/ext/open_struct.rb +5 -0
- data/lib/ext/string.rb +5 -0
- data/lib/ext/symbol.rb +5 -0
- data/spec/data_frame/arff_spec.rb +48 -0
- data/spec/data_frame/callback_array_spec.rb +148 -0
- data/spec/data_frame/core/column_management_spec.rb +128 -0
- data/spec/data_frame/core/filter_spec.rb +88 -0
- data/spec/data_frame/core/import_spec.rb +41 -0
- data/spec/data_frame/core/pre_process_spec.rb +103 -0
- data/spec/data_frame/core/saving_spec.rb +61 -0
- data/spec/data_frame/core/training_spec.rb +72 -0
- data/spec/data_frame/data_frame_spec.rb +141 -0
- data/spec/data_frame/id3_spec.rb +22 -0
- data/spec/data_frame/model_spec.rb +36 -0
- data/spec/data_frame/parameter_capture_spec.rb +32 -0
- data/spec/data_frame/transposable_array_spec.rb +138 -0
- data/spec/data_frame_spec.rb +29 -0
- data/spec/ext/array_spec.rb +13 -0
- data/spec/fixtures/basic.csv +3 -0
- data/spec/fixtures/discrete_testing.csv +4 -0
- data/spec/fixtures/discrete_training.csv +21 -0
- data/spec/spec_helper.rb +8 -0
- metadata +128 -0
@@ -0,0 +1,48 @@
|
|
1
|
+
module DF #:nodoc:
|
2
|
+
module Filter #:nodoc:
|
3
|
+
|
4
|
+
# Takes a block to evaluate on each row. The row can be converted into
|
5
|
+
# an OpenStruct or a Hash for easier filter methods. Note, don't try this
|
6
|
+
# with a hash or open struct unless you have facets available.
|
7
|
+
def filter!(as=Array, &block)
|
8
|
+
as = infer_class(as)
|
9
|
+
items = []
|
10
|
+
self.items.each do |row|
|
11
|
+
value = block.call(cast_row(row, as))
|
12
|
+
items << row if value
|
13
|
+
end
|
14
|
+
@items = items.dup
|
15
|
+
self
|
16
|
+
end
|
17
|
+
|
18
|
+
def filter(as=Array, &block)
|
19
|
+
new_data_frame = self.clone
|
20
|
+
new_data_frame.filter!(as, &block)
|
21
|
+
end
|
22
|
+
|
23
|
+
def filter_by_category(hash)
|
24
|
+
new_data_frame = self.dup
|
25
|
+
hash.each do |key, value|
|
26
|
+
key = key.to_underscore_sym
|
27
|
+
next unless self.labels.include?(key)
|
28
|
+
value = [value] unless value.is_a?(Array) or value.is_a?(Range)
|
29
|
+
new_data_frame.filter!(:hash) {|row| value.include?(row[key])}
|
30
|
+
end
|
31
|
+
new_data_frame
|
32
|
+
end
|
33
|
+
|
34
|
+
def filter_by_category!(hash)
|
35
|
+
hash.each do |key, value|
|
36
|
+
key = key.to_underscore_sym
|
37
|
+
next unless self.labels.include?(key)
|
38
|
+
value = [value] unless value.is_a?(Array) or value.is_a?(Range)
|
39
|
+
self.filter!(:hash) {|row| value.include?(row[key])}
|
40
|
+
end
|
41
|
+
end
|
42
|
+
|
43
|
+
end
|
44
|
+
end
|
45
|
+
|
46
|
+
class DataFrame
|
47
|
+
include DF::Filter
|
48
|
+
end
|
@@ -0,0 +1,113 @@
|
|
1
|
+
module DF #:nodoc:
|
2
|
+
module Import #:nodoc:
|
3
|
+
|
4
|
+
module InferCSV #:nodoc:
|
5
|
+
|
6
|
+
protected
|
7
|
+
def default_csv_opts; {:converters => :all}; end
|
8
|
+
|
9
|
+
def infer_csv_contents(obj, opts={})
|
10
|
+
contents = File.read(obj) if File.exist?(obj)
|
11
|
+
begin
|
12
|
+
open(obj) {|f| contents = f.read} unless contents
|
13
|
+
rescue
|
14
|
+
nil
|
15
|
+
end
|
16
|
+
contents ||= obj if obj.is_a?(String)
|
17
|
+
return nil unless contents
|
18
|
+
table = FCSV.parse(contents, default_csv_opts.merge(opts))
|
19
|
+
labels = opts.fetch(:headers, true) ? table.shift : []
|
20
|
+
while table.last.empty?
|
21
|
+
table.pop
|
22
|
+
end
|
23
|
+
[labels, table]
|
24
|
+
end
|
25
|
+
|
26
|
+
end # InferCSV
|
27
|
+
|
28
|
+
module ClassMethods #:nodoc:
|
29
|
+
|
30
|
+
include InferCSV
|
31
|
+
|
32
|
+
# This is the neatest part of this neat gem.
|
33
|
+
# DataFrame.from_csv can be called in a lot of ways:
|
34
|
+
# DataFrame.from_csv(csv_contents)
|
35
|
+
# DataFrame.from_csv(filename)
|
36
|
+
# DataFrame.from_csv(url)
|
37
|
+
# If you need to define converters for FasterCSV, do it before calling
|
38
|
+
# this method:
|
39
|
+
# FasterCSV::Converters[:special] = lambda{|f| f == 'foo' ? 'bar' : 'foo'}
|
40
|
+
# DataFrame.from_csv('http://example.com/my_special_url.csv', :converters => :special)
|
41
|
+
# This returns bar where 'foo' was found and 'foo' everywhere else.
|
42
|
+
def from_csv(obj, opts={})
|
43
|
+
labels, table = infer_csv_contents(obj, opts)
|
44
|
+
name = infer_name_from_contents(obj, opts)
|
45
|
+
return nil unless labels and table
|
46
|
+
df = new(*labels)
|
47
|
+
df.import(table)
|
48
|
+
df.name = name
|
49
|
+
df
|
50
|
+
end
|
51
|
+
|
52
|
+
protected
|
53
|
+
|
54
|
+
# Only works for names sources, urls and files
|
55
|
+
def infer_name_from_contents(obj, opts={})
|
56
|
+
begin
|
57
|
+
File.split(obj).last.split('.')[0..-2].join('.').titleize
|
58
|
+
rescue
|
59
|
+
nil
|
60
|
+
end
|
61
|
+
end
|
62
|
+
|
63
|
+
end # Class Methods
|
64
|
+
|
65
|
+
module InstanceMethods #:nodoc:
|
66
|
+
|
67
|
+
include InferCSV
|
68
|
+
|
69
|
+
def add_item(item)
|
70
|
+
self.items << item
|
71
|
+
end
|
72
|
+
alias :add :add_item
|
73
|
+
|
74
|
+
# Loads a batch of rows. Expects an array of arrays, else you don't
|
75
|
+
# know what you have.
|
76
|
+
def import(rows)
|
77
|
+
case rows
|
78
|
+
when Array
|
79
|
+
import_array(rows)
|
80
|
+
when String
|
81
|
+
labels, table = infer_csv_contents(rows, :headers => false)
|
82
|
+
import(table)
|
83
|
+
else
|
84
|
+
raise ArgumentError, "Don't know how to import data from #{rows.class}"
|
85
|
+
end
|
86
|
+
true
|
87
|
+
end
|
88
|
+
|
89
|
+
protected
|
90
|
+
# Imports a table as an array of arrays.
|
91
|
+
# If the array is one-dimensional and there is more than one label, it
|
92
|
+
# imports only one row.
|
93
|
+
def import_array(rows)
|
94
|
+
raise ArgumentError, "Can only work with arrays" unless rows.is_a?(Array)
|
95
|
+
if self.labels.size > 1 and rows.dimensions == 1
|
96
|
+
self.add_item(rows)
|
97
|
+
else
|
98
|
+
# self.items = self.items + rows
|
99
|
+
rows.each do |row|
|
100
|
+
self.add_item(row)
|
101
|
+
end
|
102
|
+
end
|
103
|
+
end
|
104
|
+
|
105
|
+
end # Instance Methods
|
106
|
+
|
107
|
+
end
|
108
|
+
end
|
109
|
+
|
110
|
+
class DataFrame
|
111
|
+
include DF::Import::InstanceMethods
|
112
|
+
extend DF::Import::ClassMethods
|
113
|
+
end
|
@@ -0,0 +1,69 @@
|
|
1
|
+
module DF #:nodoc:
|
2
|
+
module PreProcess #:nodoc:
|
3
|
+
# A weird name. This creates a column for every category in a column
|
4
|
+
# and marks each row by its value
|
5
|
+
def j_binary_ize!(*columns)
|
6
|
+
# Allows to mix a hash with the columns.
|
7
|
+
options = columns.find_all {|e| e.is_a?(Hash)}.inject({}) {|h, e| h.merge!(e)}
|
8
|
+
columns.delete_if {|e| e.is_a?(Hash)}
|
9
|
+
|
10
|
+
# Generates new columns
|
11
|
+
columns.each do |col|
|
12
|
+
values = render_column(col.to_underscore_sym)
|
13
|
+
values.categories.each do |category|
|
14
|
+
full_name = (col.to_s + "_" + category.to_s).to_sym
|
15
|
+
if options[:allow_overlap]
|
16
|
+
category_map = values.inject([]) do |list, e|
|
17
|
+
list << values.all_categories(e)
|
18
|
+
end
|
19
|
+
self.append!(full_name, category_map.map{|e| e.include?(category)})
|
20
|
+
else
|
21
|
+
self.append!(full_name, values.category_map.map{|e| e == category})
|
22
|
+
end
|
23
|
+
end
|
24
|
+
end
|
25
|
+
end
|
26
|
+
|
27
|
+
# Adds a column, numerical_column_name that shows the same data as a
|
28
|
+
# nominal value, but as a number.
|
29
|
+
def numericize!(*columns)
|
30
|
+
columns.each do |col|
|
31
|
+
values = render_column(col.to_underscore_sym)
|
32
|
+
categories = values.categories
|
33
|
+
value_categories = values.map {|v| values.category(v)}
|
34
|
+
|
35
|
+
i = 0
|
36
|
+
category_map = value_categories.uniq.inject({}) do |h, c|
|
37
|
+
h[c] = i
|
38
|
+
i += 1
|
39
|
+
h
|
40
|
+
end
|
41
|
+
|
42
|
+
blank = Array.new(category_map.size, 0)
|
43
|
+
reverse_category_map = category_map.inject({}) {|h, e| h[e.last] = e.first; h}
|
44
|
+
|
45
|
+
new_values = values.inject([]) do |list, val|
|
46
|
+
a = blank.dup
|
47
|
+
a[category_map[values.category(val)]] = 1
|
48
|
+
list << a
|
49
|
+
end
|
50
|
+
|
51
|
+
new_name = "numerical #{col.to_s}".to_underscore_sym
|
52
|
+
self.append!(new_name, new_values)
|
53
|
+
end
|
54
|
+
end
|
55
|
+
|
56
|
+
def categorize!(*cs)
|
57
|
+
store_range_hashes
|
58
|
+
cs.each do |column|
|
59
|
+
self.replace!(column, category_map_from_stored_range_hash(column))
|
60
|
+
end
|
61
|
+
restore_range_hashes
|
62
|
+
end
|
63
|
+
|
64
|
+
end
|
65
|
+
end
|
66
|
+
|
67
|
+
class DataFrame
|
68
|
+
include DF::PreProcess
|
69
|
+
end
|
@@ -0,0 +1,29 @@
|
|
1
|
+
module DF #:nodoc:
|
2
|
+
module Saving #:nodoc:
|
3
|
+
|
4
|
+
# Saves a data frame as CSV.
|
5
|
+
# Examples:
|
6
|
+
# df.save('/tmp/some_filename.csv')
|
7
|
+
# df.save('/tmp/some_filename.csv', :include_header => false) # No header information is saved
|
8
|
+
# df.save('/tmp/some_filename.csv', :only => [:list, :of, :columns])
|
9
|
+
# df.save('/tmp/some_filename.csv', :subset => [:list, :of, :columns])
|
10
|
+
# df.save('/tmp/some_filename.csv',
|
11
|
+
# :filter => {:column_name => :category_value,
|
12
|
+
# :another_column_name => (range..values)}) # Filter by category
|
13
|
+
def save(filename, opts={})
|
14
|
+
|
15
|
+
df = self
|
16
|
+
df = df.subset_from_columns(*Array(opts[:only])) if opts[:only]
|
17
|
+
df = df.subset_from_columns(*Array(opts[:subset])) if opts[:subset]
|
18
|
+
df = df.filter_by_category(opts[:filter]) if opts[:filter]
|
19
|
+
df = df.filter_by_category(opts[:filter_by_category]) if opts[:filter_by_category]
|
20
|
+
|
21
|
+
File.open(filename, "w") { |f| f.write df.to_csv(opts.fetch(:include_header, true)) }
|
22
|
+
end
|
23
|
+
|
24
|
+
end
|
25
|
+
end
|
26
|
+
|
27
|
+
class DataFrame
|
28
|
+
include DF::Saving
|
29
|
+
end
|
@@ -0,0 +1,46 @@
|
|
1
|
+
module Training #:nodoc:
|
2
|
+
|
3
|
+
# Remove the training set if reset
|
4
|
+
# Return cached training_set, if there is one
|
5
|
+
# Get the proportion or 80%
|
6
|
+
# Get the number of items to choose, n, or a proportion of the items
|
7
|
+
# Store and return n random items
|
8
|
+
def training_set(opts={})
|
9
|
+
if opts[:reset]
|
10
|
+
@training_set = nil
|
11
|
+
@test_set = nil
|
12
|
+
end
|
13
|
+
return @training_set if @training_set
|
14
|
+
|
15
|
+
items_size = self.items.size
|
16
|
+
proportion = opts.fetch(:proportion, 0.8)
|
17
|
+
n = opts[:n]
|
18
|
+
n ||= (items_size * proportion).to_i
|
19
|
+
n = self.items.size if n > items_size
|
20
|
+
n = 0 if n < 0
|
21
|
+
|
22
|
+
@training_set = []
|
23
|
+
while n > @training_set.size
|
24
|
+
@training_set << random_next(items_size) while n > @training_set.size
|
25
|
+
@training_set.uniq!
|
26
|
+
end
|
27
|
+
@training_set
|
28
|
+
end
|
29
|
+
|
30
|
+
|
31
|
+
def test_set(opts={})
|
32
|
+
@test_set = nil if opts[:reset]
|
33
|
+
return @test_set if @test_set
|
34
|
+
@test_set = self.items.exclusive_not(self.training_set)
|
35
|
+
end
|
36
|
+
|
37
|
+
protected
|
38
|
+
def random_next(n)
|
39
|
+
self.items[rand(n)]
|
40
|
+
end
|
41
|
+
|
42
|
+
end
|
43
|
+
|
44
|
+
class DataFrame
|
45
|
+
include Training
|
46
|
+
end
|
@@ -0,0 +1,115 @@
|
|
1
|
+
# This allows me to have named columns and optionally named rows in a
|
2
|
+
# data frame, to work calculations (usually on the columns), to
|
3
|
+
# transpose the matrix and store the transposed matrix until the object
|
4
|
+
# is tainted.
|
5
|
+
class DataFrame
|
6
|
+
|
7
|
+
def inspect
|
8
|
+
"DataFrame rows: #{self.rows.size} labels: #{self.labels.inspect}"
|
9
|
+
end
|
10
|
+
|
11
|
+
# The labels of the data items
|
12
|
+
attr_reader :labels
|
13
|
+
alias :variables :labels
|
14
|
+
|
15
|
+
# The items stored in the frame
|
16
|
+
attr_accessor :items
|
17
|
+
|
18
|
+
# An optional name, useful for arff files
|
19
|
+
attr_accessor :name
|
20
|
+
|
21
|
+
def initialize(*labels)
|
22
|
+
labels = labels.first if labels.size == 1 and labels.first.is_a?(Array)
|
23
|
+
@labels = labels.map {|e| e.to_underscore_sym }
|
24
|
+
@items = TransposableArray.new
|
25
|
+
end
|
26
|
+
|
27
|
+
def row_labels
|
28
|
+
@row_labels ||= []
|
29
|
+
end
|
30
|
+
|
31
|
+
def row_labels=(ary)
|
32
|
+
raise ArgumentError, "Row labels must be an array" unless ary.is_a?(Array)
|
33
|
+
@row_labels = ary
|
34
|
+
end
|
35
|
+
|
36
|
+
# The rows as an array of arrays, an alias for items.
|
37
|
+
alias :rows :items
|
38
|
+
|
39
|
+
def render_row(sym)
|
40
|
+
i = self.row_labels.index(sym)
|
41
|
+
return nil unless i
|
42
|
+
@items[i]
|
43
|
+
end
|
44
|
+
|
45
|
+
# Return the column, given its name
|
46
|
+
def render_column(sym)
|
47
|
+
i = @labels.index(sym.to_underscore_sym)
|
48
|
+
return nil unless i
|
49
|
+
@items.transpose[i]
|
50
|
+
end
|
51
|
+
|
52
|
+
# The columns as a Dictionary or Hash
|
53
|
+
# This is cached, call columns(true) to reset the cache.
|
54
|
+
def columns(reset=false)
|
55
|
+
@columns = nil if reset
|
56
|
+
return @columns if @columns
|
57
|
+
|
58
|
+
container = defined?(Dictionary) ? Dictionary.new : Hash.new
|
59
|
+
i = 0
|
60
|
+
|
61
|
+
@columns = @items.transpose.inject(container) do |cont, col|
|
62
|
+
cont[@labels[i]] = col
|
63
|
+
i += 1
|
64
|
+
cont
|
65
|
+
end
|
66
|
+
end
|
67
|
+
alias :to_hash :columns
|
68
|
+
alias :to_dictionary :columns
|
69
|
+
|
70
|
+
def method_missing(sym, *args, &block)
|
71
|
+
if self.labels.include?(sym)
|
72
|
+
render_column(sym)
|
73
|
+
elsif self.row_labels.include?(sym)
|
74
|
+
render_row(sym)
|
75
|
+
elsif @items.respond_to?(sym)
|
76
|
+
@items.send(sym, *args, &block)
|
77
|
+
else
|
78
|
+
super
|
79
|
+
end
|
80
|
+
end
|
81
|
+
|
82
|
+
protected
|
83
|
+
|
84
|
+
def validate_column(column)
|
85
|
+
column = column.to_sym
|
86
|
+
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)
|
87
|
+
column
|
88
|
+
end
|
89
|
+
|
90
|
+
def infer_class(obj)
|
91
|
+
obj = obj.to_s.classify.constantize if obj.is_a?(Symbol)
|
92
|
+
obj = obj.classify.constantize if obj.is_a?(String)
|
93
|
+
obj
|
94
|
+
end
|
95
|
+
|
96
|
+
def cast_row(row, as)
|
97
|
+
if as == Hash
|
98
|
+
obj = {}
|
99
|
+
self.labels.each_with_index do |label, i|
|
100
|
+
obj[label] = row[i]
|
101
|
+
end
|
102
|
+
obj
|
103
|
+
elsif as == OpenStruct
|
104
|
+
obj = OpenStruct.new
|
105
|
+
self.labels.each_with_index do |label, i|
|
106
|
+
obj.table[label] = row[i]
|
107
|
+
end
|
108
|
+
obj
|
109
|
+
elsif as == Array
|
110
|
+
row
|
111
|
+
else
|
112
|
+
as.new(*row)
|
113
|
+
end
|
114
|
+
end
|
115
|
+
end
|
@@ -0,0 +1,28 @@
|
|
1
|
+
module DF #:nodoc:
|
2
|
+
# Uses Ilya Grigorik's ID3 decision_tree gem. Installs it if you don't have it.
|
3
|
+
module ID3
|
4
|
+
begin
|
5
|
+
gem 'decisiontree'
|
6
|
+
require 'decisiontree'
|
7
|
+
rescue
|
8
|
+
`sudo gem install decisiontree`
|
9
|
+
gem 'decisiontree'
|
10
|
+
require 'decisiontree'
|
11
|
+
end
|
12
|
+
|
13
|
+
def create_id3(dependent_column, opts={})
|
14
|
+
# Need to put the dependent column in the last column
|
15
|
+
# Probably have other pre processing as well.
|
16
|
+
default = opts.fetch(:default, 1)
|
17
|
+
@id3 = DecisionTree::ID3Tree.new(self.labels, self.training_data, default, :discrete)
|
18
|
+
# ...
|
19
|
+
end
|
20
|
+
|
21
|
+
def id3
|
22
|
+
end
|
23
|
+
end
|
24
|
+
end
|
25
|
+
|
26
|
+
class DataFrame
|
27
|
+
include DF::ID3
|
28
|
+
end
|