polars_arrow/legacy/kernels/rolling/no_nulls/
variance.rs

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
use polars_error::polars_ensure;

use super::*;

pub(super) struct SumSquaredWindow<'a, T> {
    slice: &'a [T],
    sum_of_squares: T,
    last_start: usize,
    last_end: usize,
    // if we don't recompute every 'n' iterations
    // we get a accumulated error/drift
    last_recompute: u8,
}

impl<'a, T: NativeType + IsFloat + std::iter::Sum + AddAssign + SubAssign + Mul<Output = T>>
    RollingAggWindowNoNulls<'a, T> for SumSquaredWindow<'a, T>
{
    fn new(slice: &'a [T], start: usize, end: usize, _params: Option<RollingFnParams>) -> Self {
        let sum = slice[start..end].iter().map(|v| *v * *v).sum::<T>();
        Self {
            slice,
            sum_of_squares: sum,
            last_start: start,
            last_end: end,
            last_recompute: 0,
        }
    }

    unsafe fn update(&mut self, start: usize, end: usize) -> Option<T> {
        // if we exceed the end, we have a completely new window
        // so we recompute
        let recompute_sum = if start >= self.last_end || self.last_recompute > 128 {
            self.last_recompute = 0;
            true
        } else {
            self.last_recompute += 1;
            // remove elements that should leave the window
            let mut recompute_sum = false;
            for idx in self.last_start..start {
                // SAFETY:
                // we are in bounds
                let leaving_value = self.slice.get_unchecked(idx);

                if T::is_float() && !leaving_value.is_finite() {
                    recompute_sum = true;
                    break;
                }

                self.sum_of_squares -= *leaving_value * *leaving_value;
            }
            recompute_sum
        };

        self.last_start = start;

        // we traverse all values and compute
        if T::is_float() && recompute_sum {
            self.sum_of_squares = self
                .slice
                .get_unchecked(start..end)
                .iter()
                .map(|v| *v * *v)
                .sum::<T>();
        } else {
            for idx in self.last_end..end {
                let entering_value = *self.slice.get_unchecked(idx);
                self.sum_of_squares += entering_value * entering_value;
            }
        }
        self.last_end = end;
        Some(self.sum_of_squares)
    }
}

// E[(xi - E[x])^2]
// can be expanded to
// E[x^2] - E[x]^2
pub struct VarWindow<'a, T> {
    mean: MeanWindow<'a, T>,
    sum_of_squares: SumSquaredWindow<'a, T>,
    ddof: u8,
}

impl<
        'a,
        T: NativeType
            + IsFloat
            + Float
            + std::iter::Sum
            + AddAssign
            + SubAssign
            + Div<Output = T>
            + NumCast
            + One
            + Zero
            + PartialOrd
            + Sub<Output = T>,
    > RollingAggWindowNoNulls<'a, T> for VarWindow<'a, T>
{
    fn new(slice: &'a [T], start: usize, end: usize, params: Option<RollingFnParams>) -> Self {
        Self {
            mean: MeanWindow::new(slice, start, end, None),
            sum_of_squares: SumSquaredWindow::new(slice, start, end, None),
            ddof: match params {
                None => 1,
                Some(pars) => {
                    let RollingFnParams::Var(pars) = pars else {
                        unreachable!("expected Var params");
                    };
                    pars.ddof
                },
            },
        }
    }

    unsafe fn update(&mut self, start: usize, end: usize) -> Option<T> {
        let count: T = NumCast::from(end - start).unwrap();
        let sum_of_squares = self.sum_of_squares.update(start, end).unwrap_unchecked();
        let mean = self.mean.update(start, end).unwrap_unchecked();

        let denom = count - NumCast::from(self.ddof).unwrap();
        if denom <= T::zero() {
            None
        } else if end - start == 1 {
            Some(T::zero())
        } else {
            let out = (sum_of_squares - count * mean * mean) / denom;
            // variance cannot be negative.
            // if it is negative it is due to numeric instability
            if out < T::zero() {
                Some(T::zero())
            } else {
                Some(out)
            }
        }
    }
}

pub fn rolling_var<T>(
    values: &[T],
    window_size: usize,
    min_periods: usize,
    center: bool,
    weights: Option<&[f64]>,
    params: Option<RollingFnParams>,
) -> PolarsResult<ArrayRef>
where
    T: NativeType
        + Float
        + IsFloat
        + std::iter::Sum
        + AddAssign
        + SubAssign
        + Div<Output = T>
        + NumCast
        + One
        + Zero
        + Sub<Output = T>,
{
    let offset_fn = match center {
        true => det_offsets_center,
        false => det_offsets,
    };
    match weights {
        None => rolling_apply_agg_window::<VarWindow<_>, _, _>(
            values,
            window_size,
            min_periods,
            offset_fn,
            params,
        ),
        Some(weights) => {
            // Validate and standardize the weights like we do for the mean. This definition is fine
            // because frequency weights and unbiasing don't make sense for rolling operations.
            let mut wts = no_nulls::coerce_weights(weights);
            let wsum = wts.iter().fold(T::zero(), |acc, x| acc + *x);
            polars_ensure!(
                wsum != T::zero(),
                ComputeError: "Weighted variance is undefined if weights sum to 0"
            );
            wts.iter_mut().for_each(|w| *w = *w / wsum);
            super::rolling_apply_weights(
                values,
                window_size,
                min_periods,
                offset_fn,
                compute_var_weights,
                &wts,
            )
        },
    }
}

#[cfg(test)]
mod test {
    use super::*;

    #[test]
    fn test_rolling_var() {
        let values = &[1.0f64, 5.0, 3.0, 4.0];

        let out = rolling_var(values, 2, 2, false, None, None).unwrap();
        let out = out.as_any().downcast_ref::<PrimitiveArray<f64>>().unwrap();
        let out = out.into_iter().map(|v| v.copied()).collect::<Vec<_>>();
        assert_eq!(out, &[None, Some(8.0), Some(2.0), Some(0.5)]);

        let testpars = Some(RollingFnParams::Var(RollingVarParams { ddof: 0 }));
        let out = rolling_var(values, 2, 2, false, None, testpars).unwrap();
        let out = out.as_any().downcast_ref::<PrimitiveArray<f64>>().unwrap();
        let out = out.into_iter().map(|v| v.copied()).collect::<Vec<_>>();
        assert_eq!(out, &[None, Some(4.0), Some(1.0), Some(0.25)]);

        let out = rolling_var(values, 2, 1, false, None, None).unwrap();
        let out = out.as_any().downcast_ref::<PrimitiveArray<f64>>().unwrap();
        let out = out.into_iter().map(|v| v.copied()).collect::<Vec<_>>();
        // we cannot compare nans, so we compare the string values
        assert_eq!(
            format!("{:?}", out.as_slice()),
            format!("{:?}", &[None, Some(8.0), Some(2.0), Some(0.5)])
        );
        // test nan handling.
        let values = &[-10.0, 2.0, 3.0, f64::nan(), 5.0, 6.0, 7.0];
        let out = rolling_var(values, 3, 3, false, None, None).unwrap();
        let out = out.as_any().downcast_ref::<PrimitiveArray<f64>>().unwrap();
        let out = out.into_iter().map(|v| v.copied()).collect::<Vec<_>>();
        // we cannot compare nans, so we compare the string values
        assert_eq!(
            format!("{:?}", out.as_slice()),
            format!(
                "{:?}",
                &[
                    None,
                    None,
                    Some(52.333333333333336),
                    Some(f64::nan()),
                    Some(f64::nan()),
                    Some(f64::nan()),
                    Some(1.0)
                ]
            )
        );
    }
}