Upgrading ruby supported versions (#81)

* ci: Run CI on supported stable ruby versions.

* fix: Coerce some denominators to rational when calculating inverse beta.

This change fixes a segmentation error produced by BigDecimal gem on
ruby 3.1.2, where it fails to calculate a big float. It was "fixed"
in BigDecimal 3.1.2, but this version might not be compatible with older
ruby versions.

Reported here: https://bugs.ruby-lang.org/issues/18604 & here: ruby/bigdecimal#220
Potential fix here: ruby/bigdecimal@e236c2e

* spec: Add missing test for BigDecimal and goodness of fit.

.github/workflows/ruby.yml

@@ -3,22 +3,22 @@ name: Ruby
 on: [push]
 
 jobs:
- build:
+ build: # Latest ruby
 
  runs-on: ubuntu-latest
 
  steps:
  - uses: actions/checkout@v3
- - name: Set up Ruby 2.6
+ - name: Set up Ruby 3.1
  uses: ruby/setup-ruby@v1.120.0
  with:
- ruby-version: 2.6.8
+ ruby-version: 3.1.2
  - name: Build and test with Rake
  run: |
  gem install bundler
  bundle install --jobs 2 --retry 1
  bundle exec rake
- 
+
  build_2_7:
 
  runs-on: ubuntu-latest
@@ -28,13 +28,13 @@ jobs:
  - name: Set up Ruby 2.7
  uses: ruby/setup-ruby@v1.120.0
  with:
- ruby-version: 2.7.4
+ ruby-version: 2.7.6
  - name: Build and test with Rake
  run: |
  gem install bundler
  bundle install --jobs 2 --retry 1
  bundle exec rake
- 
+
  build_3_0:
 
  runs-on: ubuntu-latest
@@ -44,7 +44,7 @@ jobs:
  - name: Set up Ruby 3.0
  uses: ruby/setup-ruby@v1.120.0
  with:
- ruby-version: 3.0.2
+ ruby-version: 3.0.4
  - name: Build and test with Rake
  run: |
  gem install bundler

README.md

@@ -5,11 +5,9 @@
 A basic ruby gem that implements some statistical methods, functions and concepts to be used in any ruby environment without depending on any mathematical software like `R`, `Matlab`, `Octave` or similar.
 
 Unit test runs under the following ruby versions:
-* Ruby 2.5.1.
-* Ruby 2.6.0.
-* Ruby 2.6.3.
-* Ruby 2.6.5.
-* Ruby 2.7.
+* Ruby 2.7.6.
+* Ruby 3.0.4.
+* Ruby 3.1.2.
 
 We got the inspiration from the folks at [JStat](https://github.com/jstat/jstat) and some interesting lectures about [Keystroke dynamics](http://www.biometric-solutions.com/keystroke-dynamics.html).
 

lib/math.rb

@@ -104,9 +104,9 @@ def self.incomplete_beta_function(x, alp, bet)
 
  d = 1.0 + numerator * d
  d = tiny if d.abs < tiny
- d = 1.0 / d
+ d = 1.0 / d.to_r
 
- c = 1.0 + numerator / c
+ c = 1.0 + numerator / c.to_r
  c = tiny if c.abs < tiny
 
  cd = (c*d).freeze

spec/statistics/bigdecimal_spec.rb

@@ -117,6 +117,31 @@
  expect(result[:null]).to be true
  expect(result[:alternative]).to be false
  end
+
+ # The following test is based on the numbers reported in https://github.com/estebanz01/ruby-statistics/issues/78
+ # which give us a minimum test case scenario where the integral being solved with simpson's rule
+ # uses zero iterations, raising errors.
+ it 'performs a goodness of fit test with values that generates small chi statistics' do
+ observed_counts = [
+ BigDecimal(481, 1), BigDecimal(483, 1),
+ BigDecimal(482, 1), BigDecimal(488, 1),
+ BigDecimal(478, 1), BigDecimal(471, 1),
+ BigDecimal(477, 1), BigDecimal(479, 1),
+ BigDecimal(475, 1), BigDecimal(462, 1)
+ ]
+
+ expected = BigDecimal(477, 1)
+
+ result = {}
+
+ expect do
+ result = StatisticalTest::ChiSquaredTest.goodness_of_fit(0.01, expected, observed_counts)
+ end.not_to raise_error
+
+ expect(result[:p_value].round(4)).to eq(0.9995)
+ expect(result[:null]).to be true
+ expect(result[:alternative]).to be false
+ end
  end
 
  context 'when bigdecimal is used in F tests' do