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
use tea_core::prelude::*;
/// Trait for rolling window normalization operations on valid (non-None) elements.
pub trait RollingValidNorm<T: IsNone>: Vec1View<T> {
/// Calculates the rolling z-score (standard score) for valid elements within a window.
///
/// # Arguments
///
/// * `window` - The size of the rolling window.
/// * `min_periods` - The minimum number of observations in window required to have a value.
/// * `out` - Optional output buffer to store the results.
///
/// # Returns
///
/// A vector containing the rolling z-scores.
///
/// # Notes
///
/// The z-score is calculated as (x - mean) / standard_deviation.
/// If the standard deviation is zero or if there are fewer than `min_periods` valid observations,
/// the result will be NaN.
#[no_out]
fn ts_vzscore<O: Vec1<U>, U>(
&self,
window: usize,
min_periods: Option<usize>,
out: Option<O::UninitRefMut<'_>>,
) -> O
where
T::Inner: Number,
f64: Cast<U>,
{
let mut sum = 0.;
let mut sum2 = 0.;
let mut n = 0;
let min_periods = min_periods.unwrap_or(window / 2).min(window);
self.rolling_apply(
window,
|v_rm, v| {
let res = if v.not_none() {
n += 1;
let v = v.unwrap().f64();
sum += v;
sum2 += v * v;
if n >= min_periods {
let n_f64 = n.f64();
let mut var = sum2 / n_f64;
let mean = sum / n_f64;
var -= mean.powi(2);
if var > EPS {
(v - mean) / (var * n_f64 / (n - 1).f64()).sqrt()
} else {
f64::NAN
}
} else {
f64::NAN
}
} else {
f64::NAN
};
if let Some(v) = v_rm {
if v.not_none() {
let v = v.unwrap().f64();
n -= 1;
sum -= v;
sum2 -= v * v
};
}
res.cast()
},
out,
)
}
/// Calculates the rolling min-max normalization for valid elements within a window.
///
/// # Arguments
///
/// * `window` - The size of the rolling window.
/// * `min_periods` - The minimum number of observations in window required to have a value.
/// * `out` - Optional output buffer to store the results.
///
/// # Returns
///
/// A vector containing the rolling min-max normalized values.
///
/// # Notes
///
/// The min-max normalization is calculated as (x - min) / (max - min).
/// If max equals min or if there are fewer than `min_periods` valid observations,
/// the result will be NaN.
#[no_out]
fn ts_vminmaxnorm<O: Vec1<U>, U>(
&self,
window: usize,
min_periods: Option<usize>,
out: Option<O::UninitRefMut<'_>>,
) -> O
where
T::Inner: Number,
f64: Cast<U>,
{
let mut max = T::Inner::min_();
let mut max_idx = 0;
let mut min = T::Inner::max_();
let mut min_idx = 0;
let mut n = 0;
let min_periods = min_periods.unwrap_or(window / 2).min(window);
self.rolling_apply_idx(
window,
|start, end, v| {
if let Some(start) = start {
match (max_idx < start, min_idx < start) {
(true, false) => {
// max value is invalid, find max value again
max = T::Inner::min_();
for i in start..end {
let v = unsafe { self.uget(i) };
if v.not_none() {
let v = v.unwrap();
if v >= max {
(max, max_idx) = (v, i);
}
}
}
},
(false, true) => {
// min value is invalid, find min value again
min = T::Inner::max_();
for i in start..end {
let v = unsafe { self.uget(i) };
if v.not_none() {
let v = v.unwrap();
if v <= min {
(min, min_idx) = (v, i);
}
}
}
},
(true, true) => {
// both max and min value are invalid, find max and min value again
(max, min) = (T::Inner::min_(), T::Inner::max_());
for i in start..end {
let v = unsafe { self.uget(i) };
if v.not_none() {
let v = v.unwrap();
if v >= max {
(max, max_idx) = (v, i);
}
if v <= min {
(min, min_idx) = (v, i);
}
}
}
},
(false, false) => (), // we don't need to find max and min value again
}
}
// check if end position is max or min value
let res = if v.not_none() {
n += 1;
let v = v.unwrap();
if v >= max {
(max, max_idx) = (v, end);
}
if v <= min {
(min, min_idx) = (v, end);
}
if (n >= min_periods) & (max != min) {
((v - min).f64() / (max - min).f64()).cast()
} else {
f64::NAN.cast()
}
} else {
f64::NAN.cast()
};
if let Some(start) = start {
let v = unsafe { self.uget(start) };
if v.not_none() {
n -= 1;
}
}
res
},
out,
)
}
}
impl<T: IsNone, I: Vec1View<T>> RollingValidNorm<T> for I {}
#[cfg(test)]
mod tests {
use tea_core::testing::assert_vec1d_equal_numeric;
use super::*;
#[test]
fn test_ts_zscore() {
let data = vec![1., 2., 3., f64::NAN, 5., 6., 7., f64::NAN, 9., 10.];
let res: Vec<f64> = data.ts_vzscore(4, None);
let expect = vec![
f64::NAN,
0.707107,
1.0,
f64::NAN,
1.091089,
0.872872,
1.0,
f64::NAN,
1.091089,
0.872872,
];
// assert_eq!(res, expect);
assert_vec1d_equal_numeric(&res, &expect, Some(1e-5))
}
}