davidrichards-data_frame 0.0.18 → 0.0.19
Sign up to get free protection for your applications and to get access to all the features.
- data/README.rdoc +16 -0
- data/VERSION.yml +1 -1
- data/bin/plain_frame +22 -0
- data/lib/data_frame.rb +2 -1
- data/lib/data_frame/arff.rb +43 -36
- data/lib/data_frame/core/column_management.rb +102 -0
- data/lib/data_frame/core/filter.rb +48 -0
- data/lib/data_frame/core/import.rb +112 -0
- data/lib/data_frame/core/pre_process.rb +61 -0
- data/lib/data_frame/core/saving.rb +29 -0
- data/lib/data_frame/core/training.rb +36 -0
- data/lib/data_frame/data_frame.rb +37 -241
- 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/sbn.rb +18 -0
- data/lib/data_frame/transposable_array.rb +1 -1
- data/lib/ext/array.rb +11 -0
- data/spec/data_frame/arff_spec.rb +1 -0
- data/spec/data_frame/core/column_management_spec.rb +97 -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 +71 -0
- data/spec/data_frame/core/saving_spec.rb +61 -0
- data/spec/data_frame/core/training_spec.rb +51 -0
- data/spec/data_frame/data_frame_spec.rb +10 -226
- data/spec/data_frame/id3_spec.rb +22 -0
- data/spec/ext/array_spec.rb +13 -0
- data/spec/fixtures/discrete_testing.csv +4 -0
- data/spec/fixtures/discrete_training.csv +21 -0
- metadata +33 -6
@@ -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 +1,36 @@
|
|
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
|
+
@training_set = nil if opts[:reset]
|
10
|
+
return @training_set if @training_set
|
11
|
+
|
12
|
+
items_size = self.items.size
|
13
|
+
proportion = opts.fetch(:proportion, 0.8)
|
14
|
+
n = opts[:n]
|
15
|
+
n ||= (items_size * proportion).to_i
|
16
|
+
n = self.items.size if n > items_size
|
17
|
+
n = 0 if n < 0
|
18
|
+
|
19
|
+
@training_set = []
|
20
|
+
while n > @training_set.size
|
21
|
+
@training_set << random_next(items_size) while n > @training_set.size
|
22
|
+
@training_set.uniq!
|
23
|
+
end
|
24
|
+
@training_set
|
25
|
+
end
|
26
|
+
|
27
|
+
protected
|
28
|
+
def random_next(n)
|
29
|
+
self.items[rand(n)]
|
30
|
+
end
|
31
|
+
|
32
|
+
end
|
33
|
+
|
34
|
+
class DataFrame
|
35
|
+
include Training
|
36
|
+
end
|
@@ -4,70 +4,6 @@
|
|
4
4
|
# is tainted.
|
5
5
|
class DataFrame
|
6
6
|
|
7
|
-
class << self
|
8
|
-
|
9
|
-
# This is the neatest part of this neat gem.
|
10
|
-
# DataFrame.from_csv can be called in a lot of ways:
|
11
|
-
# DataFrame.from_csv(csv_contents)
|
12
|
-
# DataFrame.from_csv(filename)
|
13
|
-
# DataFrame.from_csv(url)
|
14
|
-
# If you need to define converters for FasterCSV, do it before calling
|
15
|
-
# this method:
|
16
|
-
# FasterCSV::Converters[:special] = lambda{|f| f == 'foo' ? 'bar' : 'foo'}
|
17
|
-
# DataFrame.from_csv('http://example.com/my_special_url.csv', :converters => :special)
|
18
|
-
# This returns bar where 'foo' was found and 'foo' everywhere else.
|
19
|
-
def from_csv(obj, opts={})
|
20
|
-
labels, table = infer_csv_contents(obj, opts)
|
21
|
-
name = infer_name_from_contents(obj, opts)
|
22
|
-
return nil unless labels and table
|
23
|
-
df = new(*labels)
|
24
|
-
df.import(table)
|
25
|
-
df.name = name
|
26
|
-
df
|
27
|
-
end
|
28
|
-
|
29
|
-
protected
|
30
|
-
|
31
|
-
# Only works for names sources, urls and files
|
32
|
-
def infer_name_from_contents(obj, opts={})
|
33
|
-
begin
|
34
|
-
File.split(obj).last.split('.')[0..-2].join('.').titleize
|
35
|
-
rescue
|
36
|
-
nil
|
37
|
-
end
|
38
|
-
end
|
39
|
-
|
40
|
-
def infer_csv_contents(obj, opts={})
|
41
|
-
contents = File.read(obj) if File.exist?(obj)
|
42
|
-
begin
|
43
|
-
open(obj) {|f| contents = f.read} unless contents
|
44
|
-
rescue
|
45
|
-
nil
|
46
|
-
end
|
47
|
-
contents ||= obj if obj.is_a?(String)
|
48
|
-
return nil unless contents
|
49
|
-
table = FCSV.parse(contents, default_csv_opts.merge(opts))
|
50
|
-
labels = table.shift
|
51
|
-
while table.last.empty?
|
52
|
-
table.pop
|
53
|
-
end
|
54
|
-
[labels, table]
|
55
|
-
end
|
56
|
-
|
57
|
-
def default_csv_opts; {:converters => :all}; end
|
58
|
-
end
|
59
|
-
|
60
|
-
# Include the methods from arff.rb
|
61
|
-
include ARFF
|
62
|
-
|
63
|
-
# Loads a batch of rows. Expects an array of arrays, else you don't
|
64
|
-
# know what you have.
|
65
|
-
def import(rows)
|
66
|
-
rows.each do |row|
|
67
|
-
self.add_item(row)
|
68
|
-
end
|
69
|
-
end
|
70
|
-
|
71
7
|
def inspect
|
72
8
|
"DataFrame rows: #{self.rows.size} labels: #{self.labels.inspect}"
|
73
9
|
end
|
@@ -83,15 +19,11 @@ class DataFrame
|
|
83
19
|
attr_accessor :name
|
84
20
|
|
85
21
|
def initialize(*labels)
|
22
|
+
labels = labels.first if labels.size == 1 and labels.first.is_a?(Array)
|
86
23
|
@labels = labels.map {|e| e.to_underscore_sym }
|
87
24
|
@items = TransposableArray.new
|
88
25
|
end
|
89
26
|
|
90
|
-
def add_item(item)
|
91
|
-
self.items << item
|
92
|
-
end
|
93
|
-
alias :add :add_item
|
94
|
-
|
95
27
|
def row_labels
|
96
28
|
@row_labels ||= []
|
97
29
|
end
|
@@ -101,15 +33,22 @@ class DataFrame
|
|
101
33
|
@row_labels = ary
|
102
34
|
end
|
103
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
|
104
46
|
def render_column(sym)
|
105
|
-
i = @labels.index(sym)
|
47
|
+
i = @labels.index(sym.to_underscore_sym)
|
106
48
|
return nil unless i
|
107
49
|
@items.transpose[i]
|
108
50
|
end
|
109
51
|
|
110
|
-
# The rows as an array of arrays, an alias for items.
|
111
|
-
alias :rows :items
|
112
|
-
|
113
52
|
# The columns as a Dictionary or Hash
|
114
53
|
# This is cached, call columns(true) to reset the cache.
|
115
54
|
def columns(reset=false)
|
@@ -128,12 +67,6 @@ class DataFrame
|
|
128
67
|
alias :to_hash :columns
|
129
68
|
alias :to_dictionary :columns
|
130
69
|
|
131
|
-
def render_row(sym)
|
132
|
-
i = self.row_labels.index(sym)
|
133
|
-
return nil unless i
|
134
|
-
@items[i]
|
135
|
-
end
|
136
|
-
|
137
70
|
def method_missing(sym, *args, &block)
|
138
71
|
if self.labels.include?(sym)
|
139
72
|
render_column(sym)
|
@@ -146,174 +79,37 @@ class DataFrame
|
|
146
79
|
end
|
147
80
|
end
|
148
81
|
|
149
|
-
|
150
|
-
labels.each do |label|
|
151
|
-
drop_one!(label)
|
152
|
-
end
|
153
|
-
self
|
154
|
-
end
|
155
|
-
|
156
|
-
def drop_one!(label)
|
157
|
-
i = self.labels.index(label)
|
158
|
-
return nil unless i
|
159
|
-
self.items.each do |item|
|
160
|
-
item.delete_at(i)
|
161
|
-
end
|
162
|
-
self.labels.delete_at(i)
|
163
|
-
self
|
164
|
-
end
|
165
|
-
protected :drop_one!
|
166
|
-
|
167
|
-
def replace!(column, values=nil, &block)
|
168
|
-
column = validate_column(column)
|
169
|
-
if not values
|
170
|
-
values = self.send(column)
|
171
|
-
values.map! {|e| block.call(e)}
|
172
|
-
end
|
173
|
-
replace_column(column, values)
|
174
|
-
self
|
175
|
-
end
|
176
|
-
|
177
|
-
def replace_column(column, values)
|
178
|
-
column = validate_column(column)
|
179
|
-
index = self.labels.index(column)
|
180
|
-
list = []
|
181
|
-
self.items.each_with_index do |item, i|
|
182
|
-
consolidated = item
|
183
|
-
consolidated[index] = values[i]
|
184
|
-
list << consolidated
|
185
|
-
end
|
186
|
-
@items = list.dup
|
187
|
-
end
|
188
|
-
protected :replace_column
|
82
|
+
protected
|
189
83
|
|
190
|
-
|
191
|
-
|
192
|
-
|
193
|
-
|
194
|
-
end
|
195
|
-
protected :validate_column
|
196
|
-
|
197
|
-
# Takes a block to evaluate on each row. The row can be converted into
|
198
|
-
# an OpenStruct or a Hash for easier filter methods. Note, don't try this
|
199
|
-
# with a hash or open struct unless you have facets available.
|
200
|
-
def filter!(as=Array, &block)
|
201
|
-
as = infer_class(as)
|
202
|
-
items = []
|
203
|
-
self.items.each do |row|
|
204
|
-
value = block.call(cast_row(row, as))
|
205
|
-
items << row if value
|
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
|
206
88
|
end
|
207
|
-
|
208
|
-
|
209
|
-
|
210
|
-
|
211
|
-
def filter(as=Array, &block)
|
212
|
-
new_data_frame = self.clone
|
213
|
-
new_data_frame.filter!(as, &block)
|
214
|
-
end
|
215
|
-
|
216
|
-
def infer_class(obj)
|
217
|
-
obj = obj.to_s.classify.constantize if obj.is_a?(Symbol)
|
218
|
-
obj = obj.classify.constantize if obj.is_a?(String)
|
219
|
-
obj
|
220
|
-
end
|
221
|
-
protected :infer_class
|
222
|
-
|
223
|
-
def cast_row(row, as)
|
224
|
-
if as == Hash
|
225
|
-
obj = {}
|
226
|
-
self.labels.each_with_index do |label, i|
|
227
|
-
obj[label] = row[i]
|
228
|
-
end
|
229
|
-
obj
|
230
|
-
elsif as == OpenStruct
|
231
|
-
obj = OpenStruct.new
|
232
|
-
self.labels.each_with_index do |label, i|
|
233
|
-
obj.table[label] = row[i]
|
234
|
-
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)
|
235
93
|
obj
|
236
|
-
elsif as == Array
|
237
|
-
row
|
238
|
-
else
|
239
|
-
as.new(*row)
|
240
94
|
end
|
241
|
-
end
|
242
|
-
protected :cast_row
|
243
|
-
|
244
|
-
# Creates a new data frame, only with the specified columns.
|
245
|
-
def subset_from_columns(*cols)
|
246
|
-
new_labels = self.labels.inject([]) do |list, label|
|
247
|
-
list << label if cols.include?(label)
|
248
|
-
list
|
249
|
-
end
|
250
|
-
new_data_frame = DataFrame.new(*self.labels)
|
251
|
-
new_data_frame.import(self.items)
|
252
|
-
self.labels.each do |label|
|
253
|
-
new_data_frame.drop!(label) unless new_labels.include?(label)
|
254
|
-
end
|
255
|
-
new_data_frame
|
256
|
-
end
|
257
|
-
|
258
|
-
# A weird name. This creates a column for every category in a column
|
259
|
-
# and marks each row by its value
|
260
|
-
def j_binary_ize!(*columns)
|
261
|
-
# Allows to mix a hash with the columns.
|
262
|
-
options = columns.find_all {|e| e.is_a?(Hash)}.inject({}) {|h, e| h.merge!(e)}
|
263
|
-
columns.delete_if {|e| e.is_a?(Hash)}
|
264
95
|
|
265
|
-
|
266
|
-
|
267
|
-
|
268
|
-
|
269
|
-
|
270
|
-
if options[:allow_overlap]
|
271
|
-
category_map = values.inject([]) do |list, e|
|
272
|
-
list << values.all_categories(e)
|
273
|
-
end
|
274
|
-
self.append!(full_name, category_map.map{|e| e.include?(category)})
|
275
|
-
else
|
276
|
-
self.append!(full_name, values.category_map.map{|e| e == category})
|
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]
|
277
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)
|
278
113
|
end
|
279
114
|
end
|
280
|
-
end
|
281
|
-
|
282
|
-
# Adds a unique column to the table
|
283
|
-
def append!(column_name, value=nil)
|
284
|
-
raise ArgumentError, "Can't have duplicate column names" if self.labels.include?(column_name)
|
285
|
-
self.labels << column_name.to_underscore_sym
|
286
|
-
if value.is_a?(Array)
|
287
|
-
self.items.each_with_index do |item, i|
|
288
|
-
item << value[i]
|
289
|
-
end
|
290
|
-
else
|
291
|
-
self.items.each do |item|
|
292
|
-
item << value
|
293
|
-
end
|
294
|
-
end
|
295
|
-
# Because we are tainting the sub arrays, the TaintableArray doesn't know it's been changed.
|
296
|
-
self.items.taint
|
297
|
-
end
|
298
|
-
|
299
|
-
def filter_by_category(hash)
|
300
|
-
new_data_frame = self.dup
|
301
|
-
hash.each do |key, value|
|
302
|
-
key = key.to_underscore_sym
|
303
|
-
next unless self.labels.include?(key)
|
304
|
-
value = [value] unless value.is_a?(Array) or value.is_a?(Range)
|
305
|
-
new_data_frame.filter!(:hash) {|row| value.include?(row[key])}
|
306
|
-
end
|
307
|
-
new_data_frame
|
308
|
-
end
|
309
|
-
|
310
|
-
def filter_by_category!(hash)
|
311
|
-
hash.each do |key, value|
|
312
|
-
key = key.to_underscore_sym
|
313
|
-
next unless self.labels.include?(key)
|
314
|
-
value = [value] unless value.is_a?(Array) or value.is_a?(Range)
|
315
|
-
self.filter!(:hash) {|row| value.include?(row[key])}
|
316
|
-
end
|
317
|
-
end
|
318
|
-
|
319
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
|
@@ -0,0 +1,48 @@
|
|
1
|
+
# The University of California - Irvine has a great set of machine
|
2
|
+
# learning sample data sets. Their data description pages have field
|
3
|
+
# label descriptors. This class extracts them and returns a DataFrame
|
4
|
+
# with the labels of a data set.
|
5
|
+
|
6
|
+
# Turns out, this isn't very useful. So...oh well.
|
7
|
+
# By the way, the code I'm talking about is found here: http://archive.ics.uci.edu/ml/
|
8
|
+
# And to use this class:
|
9
|
+
# require 'lib/data_frame/labels_from_uci'
|
10
|
+
# df = LabelsFromUCI.data_frame 'http://archive.ics.uci.edu/ml/machine-learning-databases/communities/communities.names'
|
11
|
+
# df.import('http://archive.ics.uci.edu/ml/machine-learning-databases/communities/communities.data')
|
12
|
+
|
13
|
+
class LabelsFromUCI
|
14
|
+
|
15
|
+
class << self
|
16
|
+
def process(url)
|
17
|
+
lfu = new(url)
|
18
|
+
lfu.labels
|
19
|
+
end
|
20
|
+
|
21
|
+
def data_frame(url)
|
22
|
+
lfu = new(url)
|
23
|
+
DataFrame.new(lfu.labels)
|
24
|
+
end
|
25
|
+
end
|
26
|
+
|
27
|
+
attr_reader :url, :contents, :labels
|
28
|
+
|
29
|
+
def initialize(url)
|
30
|
+
@url = url
|
31
|
+
open(url) { |f| @contents = f.read }
|
32
|
+
process_labels
|
33
|
+
end
|
34
|
+
|
35
|
+
protected
|
36
|
+
def process_labels
|
37
|
+
@labels = []
|
38
|
+
@contents.each_line do |line|
|
39
|
+
if line =~ label_re
|
40
|
+
@labels << $1
|
41
|
+
end
|
42
|
+
end
|
43
|
+
end
|
44
|
+
|
45
|
+
def label_re
|
46
|
+
/@attribute (\w+)/
|
47
|
+
end
|
48
|
+
end
|