tf_idf_vectorizer/utils/math/vector/
math.rs

1use std::{cmp::Ordering, ops::{AddAssign, MulAssign}, ptr};
2
3use num::Num;
4
5use super::ZeroSpVec;
6
7impl<N> ZeroSpVec<N> 
8where
9    N: Num + AddAssign + MulAssign
10{
11    /// ドット積を計算するメソッド
12    ///
13    /// # Arguments
14    /// * `other` - 他のベクトル
15    /// 
16    /// # Returns
17    /// * `N` - ドット積の結果
18    #[inline(always)]
19    pub fn dot<R>(&self, other: &Self) -> R
20    where
21        R: Num + AddAssign,
22        N: Into<R> + Copy,
23    {
24        debug_assert_eq!(
25            self.len(),
26            other.len(),
27            "Vectors must be of the same length to compute dot product."
28        );
29    
30        let mut result = R::zero();
31        let self_nnz = self.nnz();
32        let other_nnz = other.nnz();
33    
34        if self_nnz == 0 || other_nnz == 0 {
35            return result;
36        }
37    
38        unsafe {
39            let self_inds = std::slice::from_raw_parts(self.ind_ptr(), self_nnz);
40            let self_vals = std::slice::from_raw_parts(self.val_ptr(), self_nnz);
41            let other_inds = std::slice::from_raw_parts(other.ind_ptr(), other_nnz);
42            let other_vals = std::slice::from_raw_parts(other.val_ptr(), other_nnz);
43            
44            let mut i = 0;
45            let mut j = 0;
46            
47            while i < self_nnz && j < other_nnz {
48                match self_inds[i].cmp(&other_inds[j]) {
49                    Ordering::Equal => {
50                        result += self_vals[i].into() * other_vals[j].into();
51                        i += 1;
52                        j += 1;
53                    }
54                    Ordering::Less => i += 1,
55                    Ordering::Greater => j += 1,
56                }
57            }
58        }
59    
60        result
61    }
62
63    #[inline(always)]
64    pub fn norm_sq<R>(&self) -> R
65    where
66        R: Num + AddAssign + Copy,
67        N: Into<R> + Copy,
68    {
69        let mut result = R::zero();
70        let self_nnz = self.nnz();
71    
72        if self_nnz == 0 {
73            return result;
74        }
75    
76        unsafe {
77            let self_val_ptr = self.val_ptr();
78    
79            for i in 0..self_nnz {
80                let val: R = (*self_val_ptr.add(i)).into();
81                result += val * val;
82            }
83        }
84    
85        result
86    }
87
88    /// アダマール積を計算するメソッド
89    /// 
90    /// # Arguments
91    /// * `other` - 他のベクトル
92    /// 
93    /// # Returns
94    /// * `ZeroSpVec<N>` - アダマール積の結果
95    #[inline]
96    pub fn hadamard(&self, other: &Self) -> Self {
97        debug_assert_eq!(
98            self.len(),
99            other.len(),
100            "Vectors must be of the same length to compute hadamard product."
101        );
102
103        let min_nnz = self.nnz().min(other.nnz());
104        let mut result: ZeroSpVec<N> = ZeroSpVec::with_capacity(min_nnz);
105        result.len = self.len();
106
107        // nnz == 0 ならゼロ埋めで返す
108        if self.nnz() == 0 {
109            result.len = self.len();
110            return result;
111        }
112
113        unsafe {
114            let mut i = 0;
115            let mut j = 0;
116            while i < self.nnz() && j < other.nnz() {
117                let self_ind = ptr::read(self.ind_ptr().add(i));
118                let other_ind = ptr::read(other.ind_ptr().add(j));
119                if self_ind == other_ind {
120                    // 同じインデックスの要素を掛け算して加算
121                    let value = ptr::read(self.val_ptr().add(i)) * ptr::read(other.val_ptr().add(j));
122                    result.raw_push(self_ind, value);
123                    i += 1;
124                    j += 1;
125                } else if self_ind < other_ind {
126                    i += 1;
127                } else {
128                    j += 1;
129                }
130            }
131        }
132        result
133    }
134}
tf_idf_vectorizer/utils/math/vector/math.rs

tf_idf_vectorizer/utils/math/vector/
math.rs
