v/vlib/math/stats/stats_test.v

270 lines
10 KiB
V
Raw Normal View History

import math.stats
import math
fn test_freq() {
// Tests were also verified on Wolfram Alpha
data := [f64(10.0), f64(10.0), f64(5.9), f64(2.7)]
mut o := stats.freq(data, 10.0)
assert o == 2
o = stats.freq(data, 2.7)
assert o == 1
o = stats.freq(data, 15)
assert o == 0
}
fn tst_res(str1 string, str2 string) bool {
if (math.abs(str1.f64() - str2.f64())) < 1e-5 {
return true
}
return false
}
fn test_mean() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.mean(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '5.762500')
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.mean(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '17.650000')
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.mean(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '37.708000')
}
fn test_geometric_mean() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.geometric_mean(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '5.15993')
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.geometric_mean(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert o.str().eq('nan') || o.str().eq('-nan') || o.str().eq('-1.#IND00') || o == f64(0)
|| o.str().eq('-nan(ind)') // Because in math it yields a complex number
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.geometric_mean(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '25.064496')
}
fn test_harmonic_mean() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.harmonic_mean(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '4.626519')
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.harmonic_mean(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '9.134577')
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.harmonic_mean(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '16.555477')
}
fn test_median() {
// Tests were also verified on Wolfram Alpha
// Assumes sorted array
// Even
mut data := [f64(2.7), f64(4.45), f64(5.9), f64(10.0)]
mut o := stats.median(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '5.175000')
data = [f64(-3.0), f64(1.89), f64(4.4), f64(67.31)]
o = stats.median(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '3.145000')
data = [f64(7.88), f64(12.0), f64(54.83), f64(76.122)]
o = stats.median(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '33.415000')
// Odd
data = [f64(2.7), f64(4.45), f64(5.9), f64(10.0), f64(22)]
o = stats.median(data)
assert o == f64(5.9)
data = [f64(-3.0), f64(1.89), f64(4.4), f64(9), f64(67.31)]
o = stats.median(data)
assert o == f64(4.4)
data = [f64(7.88), f64(3.3), f64(12.0), f64(54.83), f64(76.122)]
o = stats.median(data)
assert o == f64(12.0)
}
fn test_mode() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(2.7), f64(2.7), f64(4.45), f64(5.9), f64(10.0)]
mut o := stats.mode(data)
assert o == f64(2.7)
data = [f64(-3.0), f64(1.89), f64(1.89), f64(1.89), f64(9), f64(4.4), f64(4.4), f64(9),
f64(67.31),
]
o = stats.mode(data)
assert o == f64(1.89)
// Testing greedy nature
data = [f64(2.0), f64(4.0), f64(2.0), f64(4.0)]
o = stats.mode(data)
assert o == f64(2.0)
}
fn test_rms() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.rms(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '6.362046')
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.rms(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '33.773393')
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.rms(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '47.452561')
}
fn test_population_variance() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.population_variance(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '7.269219')
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.population_variance(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '829.119550')
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.population_variance(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '829.852282')
}
fn test_sample_variance() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.sample_variance(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '9.692292')
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.sample_variance(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '1105.492733')
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.sample_variance(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '1106.469709')
}
fn test_population_stddev() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.population_stddev(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '2.696149')
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.population_stddev(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '28.794436')
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.population_stddev(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '28.807157')
}
fn test_sample_stddev() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.sample_stddev(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '3.113245')
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.sample_stddev(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '33.248951')
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.sample_stddev(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '33.263639')
}
fn test_mean_absdev() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.mean_absdev(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '2.187500')
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.mean_absdev(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '24.830000')
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.mean_absdev(data)
// Some issue with precision comparison in f64 using == operator hence serializing to string
assert tst_res(o.str(), '27.768000')
}
fn test_min() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.min(data)
assert o == f64(2.7)
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.min(data)
assert o == f64(-3.0)
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.min(data)
assert o == f64(7.88)
}
fn test_max() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.max(data)
assert o == f64(10.0)
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.max(data)
assert o == f64(67.31)
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.max(data)
assert o == f64(76.122)
}
fn test_range() {
// Tests were also verified on Wolfram Alpha
mut data := [f64(10.0), f64(4.45), f64(5.9), f64(2.7)]
mut o := stats.range(data)
assert o == f64(7.3)
data = [f64(-3.0), f64(67.31), f64(4.4), f64(1.89)]
o = stats.range(data)
assert o == f64(70.31)
data = [f64(12.0), f64(7.88), f64(76.122), f64(54.83)]
o = stats.range(data)
assert o == f64(68.242)
}
fn test_passing_empty() {
2020-04-26 13:49:31 +02:00
data := []f64{}
assert stats.freq(data, 0) == 0
assert stats.mean(data) == f64(0)
assert stats.geometric_mean(data) == f64(0)
assert stats.harmonic_mean(data) == f64(0)
assert stats.median(data) == f64(0)
assert stats.mode(data) == f64(0)
assert stats.rms(data) == f64(0)
assert stats.population_variance(data) == f64(0)
assert stats.sample_variance(data) == f64(0)
assert stats.population_stddev(data) == f64(0)
assert stats.sample_stddev(data) == f64(0)
assert stats.mean_absdev(data) == f64(0)
assert stats.min(data) == f64(0)
assert stats.max(data) == f64(0)
assert stats.range(data) == f64(0)
2020-04-26 13:49:31 +02:00
}