flashlight_tensor/cpu/subtypes/
vector.rs

1use crate::tensor::*;
2
3impl<T: Default + Clone> Tensor<T>{
4    /// Get vector from Tensor on position
5    ///
6    /// # Example
7    /// ```
8    /// use flashlight_tensor::prelude::*;
9    ///
10    /// let data: Vec<f32> = vec!{1.0, 2.0, 3.0, 4.0, 5.0, 6.0};
11    /// let sizes: Vec<u32> = vec!{2, 3};
12    ///
13    /// let tensor = Tensor::from_data(&data, &sizes).unwrap();
14    ///
15    /// let vector = tensor.vector(&[0]).unwrap();
16    ///
17    /// let expected_data: Vec<f32> = vec!{1.0, 2.0, 3.0};
18    /// let expected_sizes: Vec<u32> = vec!{3};
19    ///
20    /// assert_eq!(vector.get_data(), &expected_data);
21    /// assert_eq!(vector.get_shape(), &expected_sizes);
22    /// ```
23    pub fn vector(&self, pos: &[u32]) -> Option<Tensor<T>>{
24        let self_dimensions = self.get_shape().len();
25        let selector_dimensions = pos.len();
26        if self_dimensions - selector_dimensions != 1{
27            return None;
28        }
29        
30        for i in 0..pos.len(){
31            if pos[i] >= self.get_shape()[i]{
32                return None;
33            }
34        }
35
36        let mut data_begin: u32 = 0;
37
38        let mut stride = self.get_shape()[0];
39
40        for i in 0..pos.len() {
41            data_begin += pos[pos.len() - 1 - i] * stride;
42            stride *= self.get_shape()[1+i];
43        }
44
45        let data_end: u32 = data_begin + self.get_shape().get(self.get_shape().len()-1).unwrap();
46
47        let data = self.get_data()[data_begin as usize..data_end as usize].to_vec();
48        let sizes = self.get_shape()[self.get_shape().len()-1..self.get_shape().len()].to_vec();
49
50        Tensor::from_data(&data, &sizes)
51    }
52}
53
54impl Tensor<f32>{
55    /// Get dot product from tensors if tensors have one dimenstion
56    /// and have same size
57    ///
58    /// # Example
59    /// ```
60    /// use flashlight_tensor::prelude::*;
61    ///
62    /// let data1: Vec<f32> = vec!{1.0, 2.0, 3.0};
63    /// let data2: Vec<f32> = vec!{3.0, 2.0, 1.0};
64    ///
65    /// let expected: f32 = 10.0; // 3.0 + 4.0 + 3.0
66    ///
67    /// let tensor1 = Tensor::from_data(&data1, &[3]).unwrap();
68    /// let tensor2 = Tensor::from_data(&data2, &[3]).unwrap();
69    ///
70    /// let result = tensor1.dot_product(&tensor2).unwrap();
71    ///
72    /// assert_eq!(result, expected);
73    /// ```
74    pub fn dot_product(&self, tens2: &Tensor<f32>) -> Option<f32>{
75        if self.get_shape().len() != 1{
76            return None;
77        }
78        if self.get_shape() != tens2.get_shape(){
79            return None;
80        }
81        
82        let mut dot: f32 = 0.0;
83        for i in 0..self.get_shape()[0] as u32{
84            dot += self.value(&[i]).unwrap() * tens2.value(&[i]).unwrap();
85        }
86
87        Some(dot)
88    }
89}
flashlight_tensor/cpu/subtypes/vector.rs

flashlight_tensor/cpu/subtypes/
vector.rs
