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
//! `dot`/`concatenate` for [`MaskedArray`].
use super::MaskedArray;
use crate::array::Array;
use crate::error::{NumRs2Error, Result};
use num_traits::Zero;
use std::ops::{Add, Mul};
impl<T: Clone + Add<Output = T> + Mul<Output = T> + Zero> MaskedArray<T> {
/// Dot product of two 1-D `MaskedArray`s, matching the scope of its
/// unmasked cousin ([`Array::dot`], which is likewise 1-D-only and
/// errors on any other shape).
///
/// A pairwise product is skipped whenever *either* operand is masked
/// at that position -- exactly the mask-propagation rule `Mul` uses
/// for element-wise multiplication -- and the skipped products are
/// then summed like [`MaskedArray::sum`]: `Ok(None)` only when every
/// pairwise product was skipped, `Ok(Some(_))` otherwise. `Err` is
/// reserved for a genuine shape problem (not 1-D, or mismatched
/// lengths), which `Option` alone cannot distinguish from "every pair
/// was masked".
///
/// Pinned against `numpy.ma`:
/// `ma.dot(ma.array([1.,2.,3.],mask=[F,T,F]), ma.array([4.,5.,6.]))
/// == 22.0` (`1*4 + 3*6`; the masked `2.0`@1 drops its term
/// entirely, not just its contribution as zero-times-something).
/// When every pairwise term is masked (e.g. both operands masked at
/// disjoint, complementary positions so each product has exactly one
/// masked factor), `numpy.ma` returns its masked constant; here that
/// is `Ok(None)`.
pub fn dot(&self, other: &Self) -> Result<Option<T>> {
let a_shape = self.shape();
let b_shape = other.shape();
if a_shape.len() != 1 || b_shape.len() != 1 {
return Err(NumRs2Error::DimensionMismatch(
"MaskedArray::dot requires 1D arrays".to_string(),
));
}
if a_shape[0] != b_shape[0] {
return Err(NumRs2Error::ShapeMismatch {
expected: a_shape,
actual: b_shape,
});
}
let a_data = crate::kernels::borrow::operand(self.get_data());
let a_mask = crate::kernels::borrow::operand(self.get_mask());
let b_data = crate::kernels::borrow::operand(other.get_data());
let b_mask = crate::kernels::borrow::operand(other.get_mask());
let a_data: &[T] = &a_data;
let a_mask: &[bool] = &a_mask;
let b_data: &[T] = &b_data;
let b_mask: &[bool] = &b_mask;
let mut sum = T::zero();
let mut any_valid = false;
for ((av, am), (bv, bm)) in a_data
.iter()
.zip(a_mask.iter())
.zip(b_data.iter().zip(b_mask.iter()))
{
if !*am && !*bm {
sum = sum + av.clone() * bv.clone();
any_valid = true;
}
}
Ok(if any_valid { Some(sum) } else { None })
}
}
impl<T: Clone> MaskedArray<T> {
/// Concatenate `MaskedArray`s along a single axis, mask-preserving:
/// the output data is `Array`'s own
/// [`crate::array_ops::joining::concatenate`] applied to every input's
/// `.data`, and the output mask is the exact same operation applied to
/// every input's `.mask` -- concatenation only rearranges *which*
/// elements sit where, so no new masking decision is needed the way a
/// reduction's "all-masked lane" rule is.
///
/// Only a single `usize` axis is supported (unlike
/// `array_ops::joining::concatenate`'s `impl Into<AxisArg>`, which
/// also accepts concatenating along several axes at once) -- out of
/// scope here. The result's fill value is the first array's.
///
/// Pinned against `numpy.ma`:
/// `ma.concatenate([ma.array([1.,2.],mask=[F,T]), ma.array([3.,4.],mask=[T,F])])`
/// has data-with-mask `[1.0, --, --, 4.0]`, mask
/// `[False, True, True, False]`.
pub fn concatenate(arrays: &[&MaskedArray<T>], axis: usize) -> Result<Self> {
if arrays.is_empty() {
return Err(NumRs2Error::InvalidOperation(
"no MaskedArrays to concatenate".to_string(),
));
}
let data_refs: Vec<&Array<T>> = arrays.iter().map(|m| m.get_data()).collect();
let mask_refs: Vec<&Array<bool>> = arrays.iter().map(|m| m.get_mask()).collect();
let data = crate::array_ops::joining::concatenate(&data_refs, axis)?;
let mask = crate::array_ops::joining::concatenate(&mask_refs, axis)?;
Ok(MaskedArray {
data,
mask,
fill_value: arrays[0].get_fill_value(),
})
}
}
#[cfg(test)]
mod tests {
use super::*;
fn ma(data: Vec<f64>, mask: Vec<bool>, shape: &[usize]) -> MaskedArray<f64> {
MaskedArray {
data: Array::from_vec_shape(data, shape).expect("valid shape"),
mask: Array::from_vec_shape(mask, shape).expect("valid shape"),
fill_value: 0.0,
}
}
#[test]
fn dot_skips_pairs_where_either_operand_is_masked() {
let a = ma(vec![1.0, 2.0, 3.0], vec![false, true, false], &[3]);
let b = ma(vec![4.0, 5.0, 6.0], vec![false, false, false], &[3]);
assert_eq!(a.dot(&b).expect("1D shapes match"), Some(22.0));
}
/// Each position has exactly one masked factor, so every pairwise
/// product is skipped: the whole dot product is masked (`None`), not
/// `0.0`.
#[test]
fn dot_is_none_when_every_pairwise_product_is_masked() {
let a = ma(vec![1.0, 2.0], vec![true, false], &[2]);
let b = ma(vec![4.0, 5.0], vec![false, true], &[2]);
assert_eq!(a.dot(&b).expect("1D shapes match"), None);
}
#[test]
fn dot_rejects_non_1d_input() {
let a = ma(vec![1.0, 2.0, 3.0, 4.0], vec![false; 4], &[2, 2]);
let b = ma(vec![1.0, 2.0, 3.0, 4.0], vec![false; 4], &[2, 2]);
assert!(a.dot(&b).is_err());
}
#[test]
fn dot_rejects_mismatched_length() {
let a = ma(vec![1.0, 2.0], vec![false, false], &[2]);
let b = ma(vec![1.0, 2.0, 3.0], vec![false, false, false], &[3]);
assert!(a.dot(&b).is_err());
}
#[test]
fn concatenate_preserves_masks_from_every_input() {
let a = ma(vec![1.0, 2.0], vec![false, true], &[2]);
let b = ma(vec![3.0, 4.0], vec![true, false], &[2]);
let r = MaskedArray::concatenate(&[&a, &b], 0).expect("compatible shapes");
assert_eq!(r.shape(), vec![4]);
assert_eq!(r.get_mask().to_vec(), vec![false, true, true, false]);
assert_eq!(r.filled(Some(-1.0)).to_vec(), vec![1.0, -1.0, -1.0, 4.0]);
}
#[test]
fn concatenate_empty_list_is_an_error() {
let empty: Vec<&MaskedArray<f64>> = vec![];
assert!(MaskedArray::concatenate(&empty, 0).is_err());
}
}