statsample-ekatena 2.0.2
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +7 -0
- data/.gitignore +15 -0
- data/.travis.yml +23 -0
- data/CONTRIBUTING.md +17 -0
- data/Gemfile +2 -0
- data/History.txt +457 -0
- data/LICENSE.txt +12 -0
- data/README.md +175 -0
- data/Rakefile +44 -0
- data/benchmarks/correlation_matrix_15_variables.rb +32 -0
- data/benchmarks/correlation_matrix_5_variables.rb +33 -0
- data/benchmarks/correlation_matrix_methods/correlation_matrix.ds +0 -0
- data/benchmarks/correlation_matrix_methods/correlation_matrix.html +93 -0
- data/benchmarks/correlation_matrix_methods/correlation_matrix.rb +71 -0
- data/benchmarks/correlation_matrix_methods/correlation_matrix.xls +0 -0
- data/benchmarks/correlation_matrix_methods/correlation_matrix_gsl_ruby.ods +0 -0
- data/benchmarks/correlation_matrix_methods/correlation_matrix_with_graphics.ods +0 -0
- data/benchmarks/correlation_matrix_methods/results.ds +0 -0
- data/benchmarks/factor_map.rb +37 -0
- data/benchmarks/helpers_benchmark.rb +5 -0
- data/data/locale/es/LC_MESSAGES/statsample.mo +0 -0
- data/doc_latex/manual/equations.tex +78 -0
- data/examples/boxplot.rb +28 -0
- data/examples/chisquare_test.rb +23 -0
- data/examples/correlation_matrix.rb +32 -0
- data/examples/dataset.rb +30 -0
- data/examples/dominance_analysis.rb +33 -0
- data/examples/dominance_analysis_bootstrap.rb +32 -0
- data/examples/histogram.rb +26 -0
- data/examples/icc.rb +24 -0
- data/examples/levene.rb +29 -0
- data/examples/multiple_regression.rb +20 -0
- data/examples/multivariate_correlation.rb +33 -0
- data/examples/parallel_analysis.rb +40 -0
- data/examples/polychoric.rb +40 -0
- data/examples/principal_axis.rb +26 -0
- data/examples/reliability.rb +31 -0
- data/examples/scatterplot.rb +25 -0
- data/examples/t_test.rb +27 -0
- data/examples/tetrachoric.rb +17 -0
- data/examples/u_test.rb +24 -0
- data/examples/vector.rb +20 -0
- data/examples/velicer_map_test.rb +46 -0
- data/grab_references.rb +29 -0
- data/lib/spss.rb +134 -0
- data/lib/statsample-ekatena/analysis.rb +100 -0
- data/lib/statsample-ekatena/analysis/suite.rb +89 -0
- data/lib/statsample-ekatena/analysis/suitereportbuilder.rb +44 -0
- data/lib/statsample-ekatena/anova.rb +24 -0
- data/lib/statsample-ekatena/anova/contrast.rb +79 -0
- data/lib/statsample-ekatena/anova/oneway.rb +187 -0
- data/lib/statsample-ekatena/anova/twoway.rb +207 -0
- data/lib/statsample-ekatena/bivariate.rb +406 -0
- data/lib/statsample-ekatena/bivariate/pearson.rb +54 -0
- data/lib/statsample-ekatena/codification.rb +182 -0
- data/lib/statsample-ekatena/converter/csv.rb +28 -0
- data/lib/statsample-ekatena/converter/spss.rb +48 -0
- data/lib/statsample-ekatena/converters.rb +211 -0
- data/lib/statsample-ekatena/crosstab.rb +188 -0
- data/lib/statsample-ekatena/daru.rb +115 -0
- data/lib/statsample-ekatena/dataset.rb +10 -0
- data/lib/statsample-ekatena/dominanceanalysis.rb +425 -0
- data/lib/statsample-ekatena/dominanceanalysis/bootstrap.rb +232 -0
- data/lib/statsample-ekatena/factor.rb +104 -0
- data/lib/statsample-ekatena/factor/map.rb +124 -0
- data/lib/statsample-ekatena/factor/parallelanalysis.rb +166 -0
- data/lib/statsample-ekatena/factor/pca.rb +242 -0
- data/lib/statsample-ekatena/factor/principalaxis.rb +243 -0
- data/lib/statsample-ekatena/factor/rotation.rb +198 -0
- data/lib/statsample-ekatena/formula/fit_model.rb +46 -0
- data/lib/statsample-ekatena/formula/formula.rb +306 -0
- data/lib/statsample-ekatena/graph.rb +11 -0
- data/lib/statsample-ekatena/graph/boxplot.rb +236 -0
- data/lib/statsample-ekatena/graph/histogram.rb +198 -0
- data/lib/statsample-ekatena/graph/scatterplot.rb +213 -0
- data/lib/statsample-ekatena/histogram.rb +180 -0
- data/lib/statsample-ekatena/matrix.rb +329 -0
- data/lib/statsample-ekatena/multiset.rb +310 -0
- data/lib/statsample-ekatena/regression.rb +65 -0
- data/lib/statsample-ekatena/regression/multiple.rb +89 -0
- data/lib/statsample-ekatena/regression/multiple/alglibengine.rb +128 -0
- data/lib/statsample-ekatena/regression/multiple/baseengine.rb +251 -0
- data/lib/statsample-ekatena/regression/multiple/gslengine.rb +129 -0
- data/lib/statsample-ekatena/regression/multiple/matrixengine.rb +205 -0
- data/lib/statsample-ekatena/regression/multiple/rubyengine.rb +86 -0
- data/lib/statsample-ekatena/regression/simple.rb +121 -0
- data/lib/statsample-ekatena/reliability.rb +150 -0
- data/lib/statsample-ekatena/reliability/icc.rb +415 -0
- data/lib/statsample-ekatena/reliability/multiscaleanalysis.rb +181 -0
- data/lib/statsample-ekatena/reliability/scaleanalysis.rb +233 -0
- data/lib/statsample-ekatena/reliability/skillscaleanalysis.rb +114 -0
- data/lib/statsample-ekatena/resample.rb +15 -0
- data/lib/statsample-ekatena/shorthand.rb +125 -0
- data/lib/statsample-ekatena/srs.rb +169 -0
- data/lib/statsample-ekatena/test.rb +82 -0
- data/lib/statsample-ekatena/test/bartlettsphericity.rb +45 -0
- data/lib/statsample-ekatena/test/chisquare.rb +73 -0
- data/lib/statsample-ekatena/test/f.rb +52 -0
- data/lib/statsample-ekatena/test/kolmogorovsmirnov.rb +63 -0
- data/lib/statsample-ekatena/test/levene.rb +88 -0
- data/lib/statsample-ekatena/test/t.rb +309 -0
- data/lib/statsample-ekatena/test/umannwhitney.rb +208 -0
- data/lib/statsample-ekatena/test/wilcoxonsignedrank.rb +90 -0
- data/lib/statsample-ekatena/vector.rb +19 -0
- data/lib/statsample-ekatena/version.rb +3 -0
- data/lib/statsample.rb +282 -0
- data/po/es/statsample.mo +0 -0
- data/po/es/statsample.po +959 -0
- data/po/statsample.pot +947 -0
- data/references.txt +24 -0
- data/statsample-ekatena.gemspec +49 -0
- data/test/fixtures/bank2.dat +200 -0
- data/test/fixtures/correlation_matrix.rb +17 -0
- data/test/fixtures/df.csv +15 -0
- data/test/fixtures/hartman_23.matrix +9 -0
- data/test/fixtures/stock_data.csv +500 -0
- data/test/fixtures/tetmat_matrix.txt +5 -0
- data/test/fixtures/tetmat_test.txt +1001 -0
- data/test/helpers_tests.rb +83 -0
- data/test/test_analysis.rb +176 -0
- data/test/test_anova_contrast.rb +36 -0
- data/test/test_anovaoneway.rb +26 -0
- data/test/test_anovatwoway.rb +37 -0
- data/test/test_anovatwowaywithdataset.rb +47 -0
- data/test/test_anovawithvectors.rb +102 -0
- data/test/test_awesome_print_bug.rb +16 -0
- data/test/test_bartlettsphericity.rb +25 -0
- data/test/test_bivariate.rb +164 -0
- data/test/test_codification.rb +78 -0
- data/test/test_crosstab.rb +67 -0
- data/test/test_dominance_analysis.rb +39 -0
- data/test/test_factor.rb +228 -0
- data/test/test_factor_map.rb +38 -0
- data/test/test_factor_pa.rb +56 -0
- data/test/test_fit_model.rb +88 -0
- data/test/test_ggobi.rb +35 -0
- data/test/test_gsl.rb +15 -0
- data/test/test_histogram.rb +109 -0
- data/test/test_matrix.rb +48 -0
- data/test/test_multiset.rb +176 -0
- data/test/test_regression.rb +231 -0
- data/test/test_reliability.rb +223 -0
- data/test/test_reliability_icc.rb +198 -0
- data/test/test_reliability_skillscale.rb +57 -0
- data/test/test_resample.rb +24 -0
- data/test/test_srs.rb +9 -0
- data/test/test_statistics.rb +69 -0
- data/test/test_stest.rb +69 -0
- data/test/test_stratified.rb +17 -0
- data/test/test_test_f.rb +33 -0
- data/test/test_test_kolmogorovsmirnov.rb +34 -0
- data/test/test_test_t.rb +62 -0
- data/test/test_umannwhitney.rb +27 -0
- data/test/test_vector.rb +12 -0
- data/test/test_wilcoxonsignedrank.rb +64 -0
- metadata +570 -0
@@ -0,0 +1,207 @@
|
|
1
|
+
module Statsample
|
2
|
+
module Anova
|
3
|
+
# = Generic Anova two-way.
|
4
|
+
# You could enter the sum of squares or the mean squares for a, b, axb and within.
|
5
|
+
# You should enter the degrees of freedom for a,b and within, because df_axb=df_a*df_b
|
6
|
+
# == Usage
|
7
|
+
# anova=Statsample::Anova::TwoWay(:ss_a=>10,:ss_b=>20,:ss_axb=>10, :ss_within=>20, :df_a=>2, :df_b=>3,df_within=100 @name=>"ANOVA for....")
|
8
|
+
class TwoWay
|
9
|
+
include Summarizable
|
10
|
+
attr_reader :df_a, :df_b, :df_axb, :df_within, :df_total
|
11
|
+
attr_reader :ss_a, :ss_b, :ss_axb, :ss_within, :ss_total
|
12
|
+
attr_reader :ms_a, :ms_b, :ms_axb, :ms_within, :ms_total
|
13
|
+
# Name of ANOVA Analisys
|
14
|
+
attr_accessor :name
|
15
|
+
# Name of a factor
|
16
|
+
attr_accessor :name_a
|
17
|
+
# Name of b factor
|
18
|
+
attr_accessor :name_b
|
19
|
+
# Name of within factor
|
20
|
+
attr_accessor :name_within
|
21
|
+
|
22
|
+
attr_reader :f_a_object, :f_b_object, :f_axb_object
|
23
|
+
def initialize(opts=Hash.new)
|
24
|
+
# First see if sum of squares or mean squares are entered
|
25
|
+
raise ArgumentError, "You should set all d.f." unless [:df_a, :df_b, :df_within].all? {|v| opts.has_key? v}
|
26
|
+
|
27
|
+
@df_a=opts.delete :df_a
|
28
|
+
@df_b=opts.delete :df_b
|
29
|
+
@df_axb=@df_a*@df_b
|
30
|
+
@df_within=opts.delete :df_within
|
31
|
+
@df_total=@df_a+@df_b+@df_axb+@df_within
|
32
|
+
|
33
|
+
if [:ss_a, :ss_b, :ss_axb, :ss_within].all? {|v| opts.has_key? v}
|
34
|
+
@ss_a = opts.delete :ss_a
|
35
|
+
@ss_b = opts.delete :ss_b
|
36
|
+
@ss_axb = opts.delete :ss_axb
|
37
|
+
@ss_within = opts.delete :ss_within
|
38
|
+
|
39
|
+
@ms_a =@ss_a.quo(@df_a)
|
40
|
+
@ms_b =@ss_b.quo(@df_b)
|
41
|
+
@ms_axb =@ss_axb.quo(@df_axb)
|
42
|
+
@ms_within =@ss_within.quo(@df_within)
|
43
|
+
|
44
|
+
elsif [:ms_a, :ms_b, :ms_axb, :ms_within].all? {|v| opts.has_key? v}
|
45
|
+
@ms_a = opts.delete :ms_a
|
46
|
+
@ms_b = opts.delete :ms_b
|
47
|
+
@ms_axb = opts.delete :ms_axb
|
48
|
+
@ms_within = opts.delete :ms_within
|
49
|
+
|
50
|
+
@ss_a =@ms_a*@df_a
|
51
|
+
@ss_b =@ms_b*@df_b
|
52
|
+
@ss_axb =@ms_axb*@df_axb
|
53
|
+
@ss_within =@ms_within*@df_within
|
54
|
+
else
|
55
|
+
raise "You should set all ss or ss"
|
56
|
+
end
|
57
|
+
@ss_total=@ss_a+@ss_b+@ss_axb+@ss_within
|
58
|
+
@ms_total=@ms_a+@ms_b+@ms_axb+@ms_within
|
59
|
+
opts_default={:name=>_("ANOVA Two-Way"),
|
60
|
+
:name_a=>_("A"),
|
61
|
+
:name_b=>_("B"),
|
62
|
+
:name_within=>_("Within")
|
63
|
+
}
|
64
|
+
@opts=opts_default.merge(opts)
|
65
|
+
opts_default.keys.each {|k|
|
66
|
+
send("#{k}=", @opts[k])
|
67
|
+
}
|
68
|
+
@f_a_object=Statsample::Test::F.new(@ms_a,@ms_within,@df_a,@df_within)
|
69
|
+
@f_b_object=Statsample::Test::F.new(@ms_b,@ms_within,@df_b,@df_within)
|
70
|
+
@f_axb_object=Statsample::Test::F.new(@ms_axb,@ms_within,@df_axb,@df_within)
|
71
|
+
end
|
72
|
+
def f_a
|
73
|
+
@f_a_object.f
|
74
|
+
end
|
75
|
+
def f_b
|
76
|
+
@f_b_object.f
|
77
|
+
end
|
78
|
+
def f_axb
|
79
|
+
@f_axb_object.f
|
80
|
+
end
|
81
|
+
def f_a_probability
|
82
|
+
@f_a_object.probability
|
83
|
+
end
|
84
|
+
def f_b_probability
|
85
|
+
@f_b_object.probability
|
86
|
+
end
|
87
|
+
def f_axb_probability
|
88
|
+
@f_axb_object.probability
|
89
|
+
end
|
90
|
+
|
91
|
+
|
92
|
+
def report_building(builder) #:nodoc:
|
93
|
+
builder.section(:name=>@name) do |b|
|
94
|
+
report_building_table(b)
|
95
|
+
end
|
96
|
+
end
|
97
|
+
def report_building_table(builder) #:nodoc:
|
98
|
+
builder.table(:name=>_("%s Table") % @name, :header=>%w{source ss df ms f p}.map {|v| _(v)}) do |t|
|
99
|
+
t.row([@name_a, "%0.3f" % @ss_a, @df_a, "%0.3f" % @ms_a , "%0.3f" % f_a, "%0.4f" % f_a_probability] )
|
100
|
+
t.row([@name_b, "%0.3f" % @ss_b, @df_b, "%0.3f" % @ms_b , "%0.3f" % f_b, "%0.4f" % f_b_probability] )
|
101
|
+
t.row(["%s X %s" % [@name_a, @name_b], "%0.3f" % @ss_axb, @df_axb, "%0.3f" % @ms_axb , "%0.3f" % f_axb, "%0.4f" % f_axb_probability] )
|
102
|
+
t.row([@name_within, "%0.3f" % @ss_within, @df_within, nil,nil,nil] )
|
103
|
+
t.row([_("Total"), "%0.3f" % @ss_total, @df_total, nil,nil,nil] )
|
104
|
+
end
|
105
|
+
end
|
106
|
+
end
|
107
|
+
|
108
|
+
# Two Way Anova with vectors
|
109
|
+
# Example:
|
110
|
+
# v1 = Daru::Vector.new([1,1,2,2])
|
111
|
+
# v2 = Daru::Vector.new([1,2,1,2])
|
112
|
+
# v3 = Daru::Vector.new([5,3,1,5])
|
113
|
+
# anova=Statsample::Anova::TwoWayWithVectors.new(:a=>v1,:b=>v2, :dependent=>v3)
|
114
|
+
#
|
115
|
+
class TwoWayWithVectors < TwoWay
|
116
|
+
# Show summary Levene test
|
117
|
+
attr_accessor :summary_levene
|
118
|
+
# Show summary descriptives for variables (means)
|
119
|
+
attr_accessor :summary_descriptives
|
120
|
+
attr_reader :a_var, :b_var, :dep_var
|
121
|
+
# For now, only equal sample cells allowed
|
122
|
+
def initialize(opts=Hash.new)
|
123
|
+
raise "You should insert at least :a, :b and :dependent" unless [:a, :b, :dependent].all? {|v| opts.has_key? v}
|
124
|
+
@a_var = :a
|
125
|
+
@b_var = :b
|
126
|
+
@dep_var = :dependent
|
127
|
+
@a_vector, @b_vector, @dep_vector =
|
128
|
+
Statsample.only_valid_clone opts[:a], opts[:b], opts[:dependent]
|
129
|
+
|
130
|
+
ds = Daru::DataFrame.new({@a_var=>@a_vector, @b_var=>@b_vector, @dep_var=>@dep_vector})
|
131
|
+
@ds = ds.clone_only_valid
|
132
|
+
_p = @a_vector.factors.size
|
133
|
+
_q = @b_vector.factors.size
|
134
|
+
@x_general = @dep_vector.mean
|
135
|
+
@axb_means = {}
|
136
|
+
@axb_sd = {}
|
137
|
+
@vectors = []
|
138
|
+
n=nil
|
139
|
+
@ds.to_multiset_by_split(a_var,b_var).each_vector(dep_var) {|k,v|
|
140
|
+
@axb_means[k] = v.mean
|
141
|
+
@axb_sd[k] = v.sd
|
142
|
+
@vectors << v
|
143
|
+
n ||= v.size
|
144
|
+
raise "All cell sizes should be equal" if n!=v.size
|
145
|
+
}
|
146
|
+
|
147
|
+
@a_means={}
|
148
|
+
@ds.to_multiset_by_split(a_var).each_vector(dep_var) {|k,v|
|
149
|
+
@a_means[k]=v.mean
|
150
|
+
}
|
151
|
+
@b_means={}
|
152
|
+
@ds.to_multiset_by_split(b_var).each_vector(dep_var) {|k,v|
|
153
|
+
@b_means[k]=v.mean
|
154
|
+
}
|
155
|
+
ss_a = n*_q*@ds[a_var].factors.inject(0) {|ac,v|
|
156
|
+
ac + (@a_means[v]-@x_general)**2
|
157
|
+
}
|
158
|
+
ss_b=n*_p*@ds[b_var].factors.inject(0) {|ac,v|
|
159
|
+
ac+(@b_means[v]-@x_general)**2
|
160
|
+
}
|
161
|
+
ss_within = @ds.collect(:row) { |row|
|
162
|
+
(row[dep_var]-@axb_means[[row[a_var],row[b_var]]])**2
|
163
|
+
}.sum
|
164
|
+
ss_axb = n*@axb_means.inject(0) {|ac,v|
|
165
|
+
j,k=v[0]
|
166
|
+
xjk=v[1]
|
167
|
+
ac+(xjk-@a_means[j]-@b_means[k]+@x_general)**2
|
168
|
+
}
|
169
|
+
|
170
|
+
df_a=_p-1
|
171
|
+
df_b=_q-1
|
172
|
+
df_within=(_p*_q)*(n-1)
|
173
|
+
|
174
|
+
opts_default={:name=>_("Anova Two-Way on %s") % @ds[dep_var].name,
|
175
|
+
:name_a=>@ds[a_var].name,
|
176
|
+
:name_b=>@ds[b_var].name,
|
177
|
+
:summary_descriptives=>true,
|
178
|
+
:summary_levene=>false}
|
179
|
+
|
180
|
+
@opts=opts_default.merge(opts).merge({:ss_a=>ss_a,:ss_b=>ss_b, :ss_axb=>ss_axb, :ss_within=>ss_within, :df_a=>df_a, :df_b=>df_b, :df_within=>df_within})
|
181
|
+
|
182
|
+
|
183
|
+
super(@opts)
|
184
|
+
end
|
185
|
+
def levene
|
186
|
+
Statsample::Test.levene(@vectors, :name=>_("Test of Homogeneity of variances (Levene)"))
|
187
|
+
end
|
188
|
+
def report_building(builder) #:nodoc:#
|
189
|
+
builder.section(:name=>@name) do |s|
|
190
|
+
if summary_descriptives
|
191
|
+
s.table(:header =>['']+@ds[a_var].factors.map {|a| @ds[a_var].index_of(a)}+[_("%s Mean") % @name_b]) do |t|
|
192
|
+
@ds[b_var].factors.each do |b|
|
193
|
+
t.row([@ds[b_var].index_of(b)]+@ds[a_var].factors.map {|a| "%0.3f" % @axb_means[[a,b]] } + ["%0.3f" % @b_means[b]])
|
194
|
+
end
|
195
|
+
t.row([_("%s Mean") % @name_a]+@ds[a_var].factors.map {|a| "%0.3f" % @a_means[a]}+ ["%0.3f" % @x_general])
|
196
|
+
end
|
197
|
+
end
|
198
|
+
if summary_levene
|
199
|
+
s.parse_element(levene)
|
200
|
+
end
|
201
|
+
report_building_table(s)
|
202
|
+
|
203
|
+
end
|
204
|
+
end
|
205
|
+
end
|
206
|
+
end
|
207
|
+
end
|
@@ -0,0 +1,406 @@
|
|
1
|
+
require 'statsample/bivariate/pearson'
|
2
|
+
module Statsample
|
3
|
+
# Diverse methods and classes to calculate bivariate relations
|
4
|
+
# Specific classes:
|
5
|
+
# * Statsample::Bivariate::Pearson : Pearson correlation coefficient (r)
|
6
|
+
# * Statsample::Bivariate::Tetrachoric : Tetrachoric correlation
|
7
|
+
# * Statsample::Bivariate::Polychoric : Polychoric correlation (using joint, two-step and polychoric series)
|
8
|
+
module Bivariate
|
9
|
+
autoload(:Polychoric, 'statsample/bivariate/polychoric')
|
10
|
+
autoload(:Tetrachoric, 'statsample/bivariate/tetrachoric')
|
11
|
+
class << self
|
12
|
+
# Covariance between two vectors
|
13
|
+
def covariance(v1,v2)
|
14
|
+
v1a,v2a=Statsample.only_valid_clone(v1,v2)
|
15
|
+
|
16
|
+
return nil if v1a.size==0
|
17
|
+
if Statsample.has_gsl?
|
18
|
+
GSL::Stats::covariance(v1a.to_gsl, v2a.to_gsl)
|
19
|
+
else
|
20
|
+
covariance_slow(v1a,v2a)
|
21
|
+
end
|
22
|
+
end
|
23
|
+
# Estimate the ML between two dichotomic vectors
|
24
|
+
def maximum_likehood_dichotomic(pred,real)
|
25
|
+
preda,reala=Statsample.only_valid_clone(pred,real)
|
26
|
+
sum=0
|
27
|
+
preda.each_index{|i|
|
28
|
+
sum+=(reala[i]*Math::log(preda[i])) + ((1-reala[i])*Math::log(1-preda[i]))
|
29
|
+
}
|
30
|
+
sum
|
31
|
+
end
|
32
|
+
|
33
|
+
def covariance_slow(v1,v2) # :nodoc:
|
34
|
+
v1a,v2a=Statsample.only_valid(v1,v2)
|
35
|
+
sum_of_squares(v1a,v2a) / (v1a.size-1)
|
36
|
+
end
|
37
|
+
def sum_of_squares(v1,v2)
|
38
|
+
v1a,v2a=Statsample.only_valid_clone(v1,v2)
|
39
|
+
v1a.reset_index!
|
40
|
+
v2a.reset_index!
|
41
|
+
m1=v1a.mean
|
42
|
+
m2=v2a.mean
|
43
|
+
(v1a.size).times.inject(0) {|ac,i| ac+(v1a[i]-m1)*(v2a[i]-m2)}
|
44
|
+
end
|
45
|
+
# Calculate Pearson correlation coefficient (r) between 2 vectors
|
46
|
+
def pearson(v1,v2)
|
47
|
+
v1a,v2a=Statsample.only_valid_clone(v1,v2)
|
48
|
+
return nil if v1a.size ==0
|
49
|
+
if Statsample.has_gsl?
|
50
|
+
GSL::Stats::correlation(v1a.to_gsl, v2a.to_gsl)
|
51
|
+
else
|
52
|
+
pearson_slow(v1a,v2a)
|
53
|
+
end
|
54
|
+
end
|
55
|
+
def pearson_slow(v1,v2) # :nodoc:
|
56
|
+
v1a,v2a=Statsample.only_valid_clone(v1,v2)
|
57
|
+
|
58
|
+
# Calculate sum of squares
|
59
|
+
ss=sum_of_squares(v1a,v2a)
|
60
|
+
ss.quo(Math::sqrt(v1a.sum_of_squares) * Math::sqrt(v2a.sum_of_squares))
|
61
|
+
end
|
62
|
+
alias :correlation :pearson
|
63
|
+
# Retrieves the value for t test for a pearson correlation
|
64
|
+
# between two vectors to test the null hipothesis of r=0
|
65
|
+
def t_pearson(v1,v2)
|
66
|
+
v1a,v2a=Statsample.only_valid_clone(v1,v2)
|
67
|
+
r=pearson(v1a,v2a)
|
68
|
+
if(r==1.0)
|
69
|
+
0
|
70
|
+
else
|
71
|
+
t_r(r,v1a.size)
|
72
|
+
end
|
73
|
+
end
|
74
|
+
# Retrieves the value for t test for a pearson correlation
|
75
|
+
# giving r and vector size
|
76
|
+
# Source : http://faculty.chass.ncsu.edu/garson/PA765/correl.htm
|
77
|
+
def t_r(r,size)
|
78
|
+
r * Math::sqrt(((size)-2).to_f / (1 - r**2))
|
79
|
+
end
|
80
|
+
# Retrieves the probability value (a la SPSS)
|
81
|
+
# for a given t, size and number of tails.
|
82
|
+
# Uses a second parameter
|
83
|
+
# * :both or 2 : for r!=0 (default)
|
84
|
+
# * :right, :positive or 1 : for r > 0
|
85
|
+
# * :left, :negative : for r < 0
|
86
|
+
|
87
|
+
def prop_pearson(t, size, tails=:both)
|
88
|
+
tails=:both if tails==2
|
89
|
+
tails=:right if tails==1 or tails==:positive
|
90
|
+
tails=:left if tails==:negative
|
91
|
+
|
92
|
+
n_tails=case tails
|
93
|
+
when :both then 2
|
94
|
+
else 1
|
95
|
+
end
|
96
|
+
t=-t if t>0 and (tails==:both)
|
97
|
+
cdf=Distribution::T.cdf(t, size-2)
|
98
|
+
if(tails==:right)
|
99
|
+
1.0-(cdf*n_tails)
|
100
|
+
else
|
101
|
+
cdf*n_tails
|
102
|
+
end
|
103
|
+
end
|
104
|
+
|
105
|
+
|
106
|
+
# Predicted time for pairwise correlation matrix, in miliseconds
|
107
|
+
# See benchmarks/correlation_matrix.rb to see mode of calculation
|
108
|
+
|
109
|
+
def prediction_pairwise(vars,cases)
|
110
|
+
((-0.518111-0.000746*cases+1.235608*vars+0.000740*cases*vars)**2) / 100
|
111
|
+
end
|
112
|
+
# Predicted time for optimized correlation matrix, in miliseconds
|
113
|
+
# See benchmarks/correlation_matrix.rb to see mode of calculation
|
114
|
+
|
115
|
+
def prediction_optimized(vars,cases)
|
116
|
+
((4+0.018128*cases+0.246871*vars+0.001169*vars*cases)**2) / 100
|
117
|
+
end
|
118
|
+
# Returns residual score after delete variance
|
119
|
+
# from another variable
|
120
|
+
#
|
121
|
+
def residuals(from,del)
|
122
|
+
r=Statsample::Bivariate.pearson(from,del)
|
123
|
+
froms, dels = from.vector_standarized, del.vector_standarized
|
124
|
+
nv=[]
|
125
|
+
froms.reset_index!
|
126
|
+
dels.reset_index!
|
127
|
+
froms.each_index do |i|
|
128
|
+
if froms[i].nil? or dels[i].nil?
|
129
|
+
nv.push(nil)
|
130
|
+
else
|
131
|
+
nv.push(froms[i]-r*dels[i])
|
132
|
+
end
|
133
|
+
end
|
134
|
+
Daru::Vector.new(nv)
|
135
|
+
end
|
136
|
+
# Correlation between v1 and v2, controling the effect of
|
137
|
+
# control on both.
|
138
|
+
def partial_correlation(v1,v2,control)
|
139
|
+
v1a,v2a,cona=Statsample.only_valid_clone(v1,v2,control)
|
140
|
+
rv1v2=pearson(v1a,v2a)
|
141
|
+
rv1con=pearson(v1a,cona)
|
142
|
+
rv2con=pearson(v2a,cona)
|
143
|
+
(rv1v2-(rv1con*rv2con)).quo(Math::sqrt(1-rv1con**2) * Math::sqrt(1-rv2con**2))
|
144
|
+
end
|
145
|
+
|
146
|
+
def covariance_matrix_optimized(ds)
|
147
|
+
x=ds.to_gsl
|
148
|
+
n=x.row_size
|
149
|
+
m=x.column_size
|
150
|
+
means=((1/n.to_f)*GSL::Matrix.ones(1,n)*x).row(0)
|
151
|
+
centered=x-(GSL::Matrix.ones(n,m)*GSL::Matrix.diag(means))
|
152
|
+
ss=centered.transpose*centered
|
153
|
+
s=((1/(n-1).to_f))*ss
|
154
|
+
s
|
155
|
+
end
|
156
|
+
|
157
|
+
# Covariance matrix.
|
158
|
+
# Order of rows and columns depends on Dataset#fields order
|
159
|
+
|
160
|
+
def covariance_matrix(ds)
|
161
|
+
vars,cases = ds.ncols, ds.nrows
|
162
|
+
if !ds.include_values?(*Daru::MISSING_VALUES) and Statsample.has_gsl? and prediction_optimized(vars,cases) < prediction_pairwise(vars,cases)
|
163
|
+
cm=covariance_matrix_optimized(ds)
|
164
|
+
else
|
165
|
+
cm=covariance_matrix_pairwise(ds)
|
166
|
+
end
|
167
|
+
cm.extend(Statsample::CovariateMatrix)
|
168
|
+
cm.fields = ds.vectors.to_a
|
169
|
+
cm
|
170
|
+
end
|
171
|
+
|
172
|
+
|
173
|
+
def covariance_matrix_pairwise(ds)
|
174
|
+
cache={}
|
175
|
+
vectors = ds.vectors.to_a
|
176
|
+
mat_rows = vectors.collect do |row|
|
177
|
+
vectors.collect do |col|
|
178
|
+
if (ds[row].type!=:numeric or ds[col].type!=:numeric)
|
179
|
+
nil
|
180
|
+
elsif row==col
|
181
|
+
ds[row].variance
|
182
|
+
else
|
183
|
+
if cache[[col,row]].nil?
|
184
|
+
cov=covariance(ds[row],ds[col])
|
185
|
+
cache[[row,col]]=cov
|
186
|
+
cov
|
187
|
+
else
|
188
|
+
cache[[col,row]]
|
189
|
+
end
|
190
|
+
end
|
191
|
+
end
|
192
|
+
end
|
193
|
+
|
194
|
+
Matrix.rows mat_rows
|
195
|
+
end
|
196
|
+
|
197
|
+
# Correlation matrix.
|
198
|
+
# Order of rows and columns depends on Dataset#fields order
|
199
|
+
def correlation_matrix(ds)
|
200
|
+
vars, cases = ds.ncols, ds.nrows
|
201
|
+
if !ds.include_values?(*Daru::MISSING_VALUES) and Statsample.has_gsl? and prediction_optimized(vars,cases) < prediction_pairwise(vars,cases)
|
202
|
+
cm=correlation_matrix_optimized(ds)
|
203
|
+
else
|
204
|
+
cm=correlation_matrix_pairwise(ds)
|
205
|
+
end
|
206
|
+
cm.extend(Statsample::CovariateMatrix)
|
207
|
+
cm.fields = ds.vectors.to_a
|
208
|
+
cm
|
209
|
+
end
|
210
|
+
|
211
|
+
def correlation_matrix_optimized(ds)
|
212
|
+
s=covariance_matrix_optimized(ds)
|
213
|
+
sds=GSL::Matrix.diagonal(s.diagonal.sqrt.pow(-1))
|
214
|
+
cm=sds*s*sds
|
215
|
+
# Fix diagonal
|
216
|
+
s.row_size.times {|i|
|
217
|
+
cm[i,i]=1.0
|
218
|
+
}
|
219
|
+
cm
|
220
|
+
end
|
221
|
+
def correlation_matrix_pairwise(ds)
|
222
|
+
cache={}
|
223
|
+
vectors = ds.vectors.to_a
|
224
|
+
cm = vectors.collect do |row|
|
225
|
+
vectors.collect do |col|
|
226
|
+
if row==col
|
227
|
+
1.0
|
228
|
+
elsif (ds[row].type!=:numeric or ds[col].type!=:numeric)
|
229
|
+
nil
|
230
|
+
else
|
231
|
+
if cache[[col,row]].nil?
|
232
|
+
r=pearson(ds[row],ds[col])
|
233
|
+
cache[[row,col]]=r
|
234
|
+
r
|
235
|
+
else
|
236
|
+
cache[[col,row]]
|
237
|
+
end
|
238
|
+
end
|
239
|
+
end
|
240
|
+
end
|
241
|
+
|
242
|
+
Matrix.rows cm
|
243
|
+
end
|
244
|
+
|
245
|
+
# Retrieves the n valid pairwise.
|
246
|
+
def n_valid_matrix(ds)
|
247
|
+
vectors = ds.vectors.to_a
|
248
|
+
m = vectors.collect do |row|
|
249
|
+
vectors.collect do |col|
|
250
|
+
if row==col
|
251
|
+
ds[row].reject_values(*Daru::MISSING_VALUES).size
|
252
|
+
else
|
253
|
+
rowa,rowb = Statsample.only_valid_clone(ds[row],ds[col])
|
254
|
+
rowa.size
|
255
|
+
end
|
256
|
+
end
|
257
|
+
end
|
258
|
+
|
259
|
+
Matrix.rows m
|
260
|
+
end
|
261
|
+
|
262
|
+
# Matrix of correlation probabilities.
|
263
|
+
# Order of rows and columns depends on Dataset#fields order
|
264
|
+
|
265
|
+
def correlation_probability_matrix(ds, tails=:both)
|
266
|
+
rows=ds.fields.collect do |row|
|
267
|
+
ds.fields.collect do |col|
|
268
|
+
v1a,v2a=Statsample.only_valid_clone(ds[row],ds[col])
|
269
|
+
(row==col or ds[row].type!=:numeric or ds[col].type!=:numeric) ? nil : prop_pearson(t_pearson(ds[row],ds[col]), v1a.size, tails)
|
270
|
+
end
|
271
|
+
end
|
272
|
+
Matrix.rows(rows)
|
273
|
+
end
|
274
|
+
|
275
|
+
# Spearman ranked correlation coefficient (rho) between 2 vectors
|
276
|
+
def spearman(v1,v2)
|
277
|
+
v1a,v2a = Statsample.only_valid_clone(v1,v2)
|
278
|
+
v1r,v2r = v1a.ranked, v2a.ranked
|
279
|
+
pearson(v1r,v2r)
|
280
|
+
end
|
281
|
+
# Calculate Point biserial correlation. Equal to Pearson correlation, with
|
282
|
+
# one dichotomous value replaced by "0" and the other by "1"
|
283
|
+
def point_biserial(dichotomous,continous)
|
284
|
+
ds = Daru::DataFrame.new({:d=>dichotomous,:c=>continous}).reject_values(*Daru::MISSING_VALUES)
|
285
|
+
raise(TypeError, "First vector should be dichotomous") if ds[:d].factors.size != 2
|
286
|
+
raise(TypeError, "Second vector should be continous") if ds[:c].type != :numeric
|
287
|
+
f0=ds[:d].factors.sort.to_a[0]
|
288
|
+
m0=ds.filter_vector(:c) {|c| c[:d] == f0}
|
289
|
+
m1=ds.filter_vector(:c) {|c| c[:d] != f0}
|
290
|
+
((m1.mean-m0.mean).to_f / ds[:c].sdp) * Math::sqrt(m0.size*m1.size.to_f / ds.nrows**2)
|
291
|
+
end
|
292
|
+
# Kendall Rank Correlation Coefficient (Tau a)
|
293
|
+
# Based on Hervé Adbi article
|
294
|
+
def tau_a(v1,v2)
|
295
|
+
v1a,v2a=Statsample.only_valid_clone(v1,v2)
|
296
|
+
n=v1.size
|
297
|
+
v1r,v2r=v1a.ranked,v2a.ranked
|
298
|
+
o1=ordered_pairs(v1r)
|
299
|
+
o2=ordered_pairs(v2r)
|
300
|
+
delta= o1.size*2-(o2 & o1).size*2
|
301
|
+
1-(delta * 2 / (n*(n-1)).to_f)
|
302
|
+
end
|
303
|
+
# Calculates Goodman and Kruskal’s Tau b correlation.
|
304
|
+
# Tb is an asymmetric P-R-E measure of association for nominal scales
|
305
|
+
# (Mielke, X)
|
306
|
+
#
|
307
|
+
# Tau-b defines perfect association as strict monotonicity. Although it
|
308
|
+
# requires strict monotonicity to reach 1.0, it does not penalize ties as
|
309
|
+
# much as some other measures.
|
310
|
+
# == Reference
|
311
|
+
# Mielke, P. GOODMAN–KRUSKAL TAU AND GAMMA.
|
312
|
+
# Source: http://faculty.chass.ncsu.edu/garson/PA765/assocordinal.htm
|
313
|
+
def tau_b(matrix)
|
314
|
+
v=pairs(matrix)
|
315
|
+
((v['P']-v['Q']).to_f / Math::sqrt((v['P']+v['Q']+v['Y'])*(v['P']+v['Q']+v['X'])).to_f)
|
316
|
+
end
|
317
|
+
# Calculates Goodman and Kruskal's gamma.
|
318
|
+
#
|
319
|
+
# Gamma is the surplus of concordant pairs over discordant pairs, as a
|
320
|
+
# percentage of all pairs ignoring ties.
|
321
|
+
#
|
322
|
+
# Source: http://faculty.chass.ncsu.edu/garson/PA765/assocordinal.htm
|
323
|
+
def gamma(matrix)
|
324
|
+
v=pairs(matrix)
|
325
|
+
(v['P']-v['Q']).to_f / (v['P']+v['Q']).to_f
|
326
|
+
end
|
327
|
+
# Calculate indexes for a matrix the rows and cols has to be ordered
|
328
|
+
def pairs(matrix)
|
329
|
+
# calculate concordant #p matrix
|
330
|
+
rs=matrix.row_size
|
331
|
+
cs=matrix.column_size
|
332
|
+
conc=disc=ties_x=ties_y=0
|
333
|
+
(0...(rs-1)).each do |x|
|
334
|
+
(0...(cs-1)).each do |y|
|
335
|
+
((x+1)...rs).each do |x2|
|
336
|
+
((y+1)...cs).each do |y2|
|
337
|
+
# #p sprintf("%d:%d,%d:%d",x,y,x2,y2)
|
338
|
+
conc+=matrix[x,y]*matrix[x2,y2]
|
339
|
+
end
|
340
|
+
end
|
341
|
+
end
|
342
|
+
end
|
343
|
+
(0...(rs-1)).each {|x|
|
344
|
+
(1...(cs)).each{|y|
|
345
|
+
((x+1)...rs).each{|x2|
|
346
|
+
(0...y).each{|y2|
|
347
|
+
# #p sprintf("%d:%d,%d:%d",x,y,x2,y2)
|
348
|
+
disc+=matrix[x,y]*matrix[x2,y2]
|
349
|
+
}
|
350
|
+
}
|
351
|
+
}
|
352
|
+
}
|
353
|
+
(0...(rs-1)).each {|x|
|
354
|
+
(0...(cs)).each{|y|
|
355
|
+
((x+1)...(rs)).each{|x2|
|
356
|
+
ties_x+=matrix[x,y]*matrix[x2,y]
|
357
|
+
}
|
358
|
+
}
|
359
|
+
}
|
360
|
+
(0...rs).each {|x|
|
361
|
+
(0...(cs-1)).each{|y|
|
362
|
+
((y+1)...(cs)).each{|y2|
|
363
|
+
ties_y+=matrix[x,y]*matrix[x,y2]
|
364
|
+
}
|
365
|
+
}
|
366
|
+
}
|
367
|
+
{'P'=>conc,'Q'=>disc,'Y'=>ties_y,'X'=>ties_x}
|
368
|
+
end
|
369
|
+
|
370
|
+
def ordered_pairs(vector)
|
371
|
+
d = vector.to_a
|
372
|
+
a = []
|
373
|
+
(0...(d.size-1)).each do |i|
|
374
|
+
((i+1)...(d.size)).each do |j|
|
375
|
+
a.push([d[i],d[j]])
|
376
|
+
end
|
377
|
+
end
|
378
|
+
a
|
379
|
+
end
|
380
|
+
=begin
|
381
|
+
def sum_of_codeviated(v1,v2)
|
382
|
+
v1a,v2a=Statsample.only_valid(v1,v2)
|
383
|
+
sum=0
|
384
|
+
(0...v1a.size).each{|i|
|
385
|
+
sum+=v1a[i]*v2a[i]
|
386
|
+
}
|
387
|
+
sum-((v1a.sum*v2a.sum) / v1a.size.to_f)
|
388
|
+
end
|
389
|
+
=end
|
390
|
+
# Report the minimum number of cases valid of a covariate matrix
|
391
|
+
# based on a dataset
|
392
|
+
def min_n_valid(ds)
|
393
|
+
min = ds.nrows
|
394
|
+
m = n_valid_matrix(ds)
|
395
|
+
for x in 0...m.row_size
|
396
|
+
for y in 0...m.column_size
|
397
|
+
min=m[x,y] if m[x,y] < min
|
398
|
+
end
|
399
|
+
end
|
400
|
+
min
|
401
|
+
end
|
402
|
+
end
|
403
|
+
end
|
404
|
+
end
|
405
|
+
|
406
|
+
|