ferray-stats 0.3.5

Statistical functions, reductions, sorting, and histograms for ferray
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253

// Property-based tests for ferray-stats
//
// Tests mathematical invariants of statistical functions using proptest.

// Property tests sample integer sizes and lift into `f64` for invariant
// checks; some invariants assert exact float equality (idempotence,
// permutation invariance) by design.
#![allow(
    clippy::cast_possible_truncation,
    clippy::cast_possible_wrap,
    clippy::cast_precision_loss,
    clippy::cast_sign_loss,
    clippy::cast_lossless,
    clippy::float_cmp
)]

use ferray_core::Array;
use ferray_core::dimension::Ix1;

use ferray_stats::sorting::{SortKind, sort};
use ferray_stats::{argmax, argmin, cumsum, max, mean, min, std_, sum, unique, var};

use proptest::prelude::*;

// ---------------------------------------------------------------------------
// Helpers
// ---------------------------------------------------------------------------

fn arr1(data: Vec<f64>) -> Array<f64, Ix1> {
    let n = data.len();
    Array::<f64, Ix1>::from_vec(Ix1::new([n]), data).unwrap()
}

fn scalar_val(a: &Array<f64, ferray_core::IxDyn>) -> f64 {
    *a.iter().next().unwrap()
}

proptest! {
    #![proptest_config(ProptestConfig::with_cases(256))]

    // -----------------------------------------------------------------------
    // 1. mean(constant_array) == constant
    // -----------------------------------------------------------------------
    #[test]
    fn prop_mean_constant(c in -100.0f64..100.0, n in 1usize..=50) {
        let a = arr1(vec![c; n]);
        let m = mean(&a, None).unwrap();
        let val = scalar_val(&m);
        prop_assert!((val - c).abs() < 1e-10, "mean of constant {} array = {}", c, val);
    }

    // -----------------------------------------------------------------------
    // 2. var(constant_array) == 0
    // -----------------------------------------------------------------------
    #[test]
    fn prop_var_constant(c in -100.0f64..100.0, n in 2usize..=50) {
        let a = arr1(vec![c; n]);
        let v = var(&a, None, 0).unwrap();
        let val = scalar_val(&v);
        prop_assert!(val.abs() < 1e-10, "var of constant {} array = {}", c, val);
    }

    // -----------------------------------------------------------------------
    // 3. std(constant_array) == 0
    // -----------------------------------------------------------------------
    #[test]
    fn prop_std_constant(c in -100.0f64..100.0, n in 2usize..=50) {
        let a = arr1(vec![c; n]);
        let s = std_(&a, None, 0).unwrap();
        let val = scalar_val(&s);
        prop_assert!(val.abs() < 1e-10, "std of constant {} array = {}", c, val);
    }

    // -----------------------------------------------------------------------
    // 4. sum(a) == len(a) * mean(a) (approximately)
    // -----------------------------------------------------------------------
    #[test]
    fn prop_sum_eq_len_times_mean(data in proptest::collection::vec(-100.0f64..100.0, 1..=50)) {
        let n = data.len();
        let a = arr1(data);
        let s = sum(&a, None).unwrap();
        let m = mean(&a, None).unwrap();
        let s_val = scalar_val(&s);
        let m_val = scalar_val(&m);
        let expected = n as f64 * m_val;
        let diff = (s_val - expected).abs();
        let scale = s_val.abs().max(1.0);
        prop_assert!(diff / scale < 1e-10, "sum={} != len*mean={}", s_val, expected);
    }

    // -----------------------------------------------------------------------
    // 5. sort is idempotent: sort(sort(a)) == sort(a)
    // -----------------------------------------------------------------------
    #[test]
    fn prop_sort_idempotent(data in proptest::collection::vec(-100.0f64..100.0, 1..=50)) {
        let a = arr1(data);
        let sorted1 = sort(&a, None, SortKind::Stable).unwrap();
        let sorted2 = sort(&sorted1, None, SortKind::Stable).unwrap();
        let v1: Vec<f64> = sorted1.iter().copied().collect();
        let v2: Vec<f64> = sorted2.iter().copied().collect();
        prop_assert_eq!(v1, v2);
    }

    // -----------------------------------------------------------------------
    // 6. sort output is monotonically non-decreasing
    // -----------------------------------------------------------------------
    #[test]
    fn prop_sort_monotone(data in proptest::collection::vec(-100.0f64..100.0, 2..=50)) {
        let a = arr1(data);
        let sorted = sort(&a, None, SortKind::Stable).unwrap();
        let vals: Vec<f64> = sorted.iter().copied().collect();
        for i in 1..vals.len() {
            prop_assert!(
                vals[i] >= vals[i - 1],
                "sort not monotone at index {}: {} < {}",
                i, vals[i], vals[i - 1]
            );
        }
    }

    // -----------------------------------------------------------------------
    // 7. unique output has no duplicates
    // -----------------------------------------------------------------------
    #[test]
    fn prop_unique_no_duplicates(data in proptest::collection::vec(-20.0f64..20.0, 1..=50)) {
        // Round to integers so we get some duplicates
        let rounded: Vec<f64> = data.iter().map(|x| x.round()).collect();
        let a = arr1(rounded);
        let result = unique(&a, false, false, false).unwrap();
        let vals: Vec<f64> = result.values.iter().copied().collect();
        for i in 1..vals.len() {
            prop_assert!(
                vals[i] != vals[i - 1],
                "unique has duplicates at index {}: {} == {}",
                i, vals[i], vals[i - 1]
            );
        }
    }

    // -----------------------------------------------------------------------
    // 8. unique output is sorted
    // -----------------------------------------------------------------------
    #[test]
    fn prop_unique_sorted(data in proptest::collection::vec(-50.0f64..50.0, 1..=50)) {
        let rounded: Vec<f64> = data.iter().map(|x| x.round()).collect();
        let a = arr1(rounded);
        let result = unique(&a, false, false, false).unwrap();
        let vals: Vec<f64> = result.values.iter().copied().collect();
        for i in 1..vals.len() {
            prop_assert!(
                vals[i] >= vals[i - 1],
                "unique not sorted: {} < {} at index {}",
                vals[i], vals[i - 1], i
            );
        }
    }

    // -----------------------------------------------------------------------
    // 9. argmin indexes the minimum value
    // -----------------------------------------------------------------------
    #[test]
    fn prop_argmin_indexes_min(data in proptest::collection::vec(-100.0f64..100.0, 1..=50)) {
        let a = arr1(data.clone());
        let ami = argmin(&a, None).unwrap();
        let idx = *ami.iter().next().unwrap() as usize;
        let mn = min(&a, None).unwrap();
        let min_val = scalar_val(&mn);
        prop_assert!(
            (data[idx] - min_val).abs() < 1e-15,
            "argmin index {} has value {} but min is {}",
            idx, data[idx], min_val
        );
    }

    // -----------------------------------------------------------------------
    // 10. argmax indexes the maximum value
    // -----------------------------------------------------------------------
    #[test]
    fn prop_argmax_indexes_max(data in proptest::collection::vec(-100.0f64..100.0, 1..=50)) {
        let a = arr1(data.clone());
        let amx = argmax(&a, None).unwrap();
        let idx = *amx.iter().next().unwrap() as usize;
        let mx = max(&a, None).unwrap();
        let max_val = scalar_val(&mx);
        prop_assert!(
            (data[idx] - max_val).abs() < 1e-15,
            "argmax index {} has value {} but max is {}",
            idx, data[idx], max_val
        );
    }

    // -----------------------------------------------------------------------
    // 11. cumsum last element equals sum
    // -----------------------------------------------------------------------
    #[test]
    fn prop_cumsum_last_eq_sum(data in proptest::collection::vec(-100.0f64..100.0, 1..=50)) {
        let a = arr1(data);
        let cs = cumsum(&a, None).unwrap();
        let s = sum(&a, None).unwrap();
        let cs_last = *cs.iter().last().unwrap();
        let s_val = scalar_val(&s);
        let diff = (cs_last - s_val).abs();
        let scale = s_val.abs().max(1.0);
        prop_assert!(
            diff / scale < 1e-10,
            "cumsum last = {}, sum = {}, diff = {}",
            cs_last, s_val, diff
        );
    }

    // -----------------------------------------------------------------------
    // 12. min(a) <= mean(a) <= max(a) for non-empty arrays
    // -----------------------------------------------------------------------
    #[test]
    fn prop_min_le_mean_le_max(data in proptest::collection::vec(-100.0f64..100.0, 1..=50)) {
        let a = arr1(data);
        let mn = min(&a, None).unwrap();
        let mx = max(&a, None).unwrap();
        let m = mean(&a, None).unwrap();
        let min_val = scalar_val(&mn);
        let max_val = scalar_val(&mx);
        let mean_val = scalar_val(&m);
        prop_assert!(
            min_val <= mean_val + 1e-10,
            "min {} > mean {}",
            min_val, mean_val
        );
        prop_assert!(
            mean_val <= max_val + 1e-10,
            "mean {} > max {}",
            mean_val, max_val
        );
    }

    // -----------------------------------------------------------------------
    // 13. sum of sorted == sum of original
    // -----------------------------------------------------------------------
    #[test]
    fn prop_sum_invariant_under_sort(data in proptest::collection::vec(-100.0f64..100.0, 1..=50)) {
        let a = arr1(data);
        let sorted = sort(&a, None, SortKind::Stable).unwrap();
        let s1 = sum(&a, None).unwrap();
        let s2 = sum(&sorted, None).unwrap();
        let v1 = scalar_val(&s1);
        let v2 = scalar_val(&s2);
        let scale = v1.abs().max(1.0);
        prop_assert!(
            (v1 - v2).abs() / scale < 1e-10,
            "sum changed after sort: {} vs {}",
            v1, v2
        );
    }
}