nove_tensor 0.1.2

An easy-to-use, lightweight deep learning library wrapped around Candle Tensor.
Documentation 
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

use nove::tensor::{DType, Device, Shape, Tensor};
use proptest::prelude::*;
use rand::{Rng, SeedableRng, rngs::StdRng};

/// Generate a random tensor and its corresponding data.
///
/// # Notes
/// * The dimension number of the tensor must be in the range [1, 4].
///
/// # Generic Type Parameters
/// * `S` - The element type of the vector. It supports `f32`, `f64`, `i64`, `u32`, `u8`.
///
/// # Arguments
/// * `shape` - The shape of the tensor.
/// * `dtype` - The data type of the tensor.
/// * `device` - The device of the tensor.
/// * `grad_enabled` - Whether the tensor requires gradient.
/// * `seed` - The seed for random number generator.
///
/// # Returns
/// * `(tensor, data)` - The tensor and its corresponding data.
pub fn generate_random_tensor_and_corresponding_data<S>(
    shape: &Shape,
    dtype: &DType,
    device: &Device,
    grad_enabled: bool,
    seed: u64,
) -> (Tensor, Vec<S>)
where
    S: candle_core::WithDType,
{
    // Check dimension number
    let num_dim = shape.rank();
    if num_dim > 4 {
        panic!(
            "Unsupported dimension number {} (only support [1, 4])",
            num_dim
        );
    }

    // Count total size
    let dims = shape.dims();
    let total_size = dims.iter().product::<usize>();

    // Generate random data
    let mut rng = StdRng::seed_from_u64(seed);
    let mut data: Vec<f32> = Vec::with_capacity(total_size);
    for _ in 0..total_size {
        data.push(rng.random());
    }

    // Create tensor from data
    let mut tensor = match num_dim {
        1 => Tensor::from_data(&data[..dims[0]], device, grad_enabled).unwrap(),
        2 => {
            let data_2d: Vec<Vec<f32>> = data[..dims[0] * dims[1]]
                .chunks(dims[1])
                .map(|c| c.to_vec())
                .collect();
            Tensor::from_data(data_2d, device, grad_enabled).unwrap()
        }
        3 => {
            let data_3d: Vec<Vec<Vec<f32>>> = data[..dims[0] * dims[1] * dims[2]]
                .chunks(dims[1] * dims[2])
                .map(|chunk| chunk.chunks(dims[2]).map(|c| c.to_vec()).collect())
                .collect();
            Tensor::from_data(data_3d, device, grad_enabled).unwrap()
        }
        4 => {
            let data_4d: Vec<Vec<Vec<Vec<f32>>>> = data
                .chunks(dims[1] * dims[2] * dims[3])
                .map(|chunk| {
                    chunk
                        .chunks(dims[2] * dims[3])
                        .map(|c| c.chunks(dims[3]).map(|d| d.to_vec()).collect())
                        .collect()
                })
                .collect();
            Tensor::from_data(data_4d, device, grad_enabled).unwrap()
        }
        _ => unreachable!(),
    };

    if *dtype != tensor.dtype().unwrap() {
        tensor = tensor.to_dtype(dtype).unwrap();
    }

    // Convert data to the specified dtype
    let data = match dtype {
        DType::F32 => {
            assert_eq!(std::any::TypeId::of::<S>(), std::any::TypeId::of::<f32>());
            let concrete_data: Vec<f32> = data;
            unsafe { std::mem::transmute(concrete_data) }
        }
        DType::F64 => {
            assert_eq!(std::any::TypeId::of::<S>(), std::any::TypeId::of::<f64>());
            let concrete_data: Vec<f64> = data.into_iter().map(|x| x as f64).collect();
            unsafe { std::mem::transmute(concrete_data) }
        }
        DType::I64 => {
            assert_eq!(std::any::TypeId::of::<S>(), std::any::TypeId::of::<i64>());
            let concrete_data: Vec<i64> = data.into_iter().map(|x| x as i64).collect();
            unsafe { std::mem::transmute(concrete_data) }
        }
        DType::U32 => {
            assert_eq!(std::any::TypeId::of::<S>(), std::any::TypeId::of::<u32>());
            let concrete_data: Vec<u32> = data.into_iter().map(|x| x as u32).collect();
            unsafe { std::mem::transmute(concrete_data) }
        }
        DType::U8 => {
            assert_eq!(std::any::TypeId::of::<S>(), std::any::TypeId::of::<u8>());
            let concrete_data: Vec<u8> = data.into_iter().map(|x| x as u8).collect();
            unsafe { std::mem::transmute(concrete_data) }
        }
        _ => panic!("Unsupported dtype {:?}", dtype),
    };

    (tensor, data)
}

/// Generate a random Shape with an appropriate size.
///
/// # Note
/// * The maximum dimension number is 4.
/// * The maximum size of each dimension is 100.
/// * The product of all dimensions is less than 10000.
///
/// # Arguments
/// * `seed` - The seed for random number generator.
///
/// # Returns
/// * `shape` - The random shape.
pub fn generate_random_shape_with_appropriate_size(seed: u64) -> Shape {
    let mut rng = StdRng::seed_from_u64(seed);
    let num_dim = rng.random_range(1..=4);
    let mut dims = Vec::with_capacity(num_dim);
    let mut total_size = 1;
    for _ in 0..num_dim {
        let max_dim = (10000 / total_size).min(100);
        let dim = rng.random_range(1..=max_dim);
        dims.push(dim);
        total_size *= dim;
    }
    Shape::from_dims(dims.as_slice())
}

proptest! {
    #[test]
    fn test_generate_random_tensor_and_corresponding_data_with_random_shape_and_random_dtype(
        shape_seed in prop::num::u64::ANY,
        dtype in prop::sample::select(vec![DType::U8, DType::U32, DType::I64, DType::F32,DType::F64]),
        grad_enabled in prop::bool::ANY,
        data_seed in prop::num::u64::ANY,
    ) {
        let shape = generate_random_shape_with_appropriate_size(shape_seed);
        let dtype = DType::from(dtype);
        let device = Device::cpu();

        match dtype {
            DType::U8 => {
                let (tensor, data) = generate_random_tensor_and_corresponding_data::<u8>(&shape, &dtype, &device, grad_enabled, data_seed);
                assert_eq!(tensor.shape().unwrap(), shape);
                assert_eq!(tensor.dtype().unwrap(), dtype);
                assert_eq!(tensor.device().unwrap(), device);
                assert_eq!(tensor.grad_enabled().unwrap(), grad_enabled);
                assert_eq!(tensor.to_vec::<u8>().unwrap(), data);
            }
            DType::U32 => {
                let (tensor, data) = generate_random_tensor_and_corresponding_data::<u32>(&shape, &dtype, &device, grad_enabled, data_seed);
                assert_eq!(tensor.shape().unwrap(), shape);
                assert_eq!(tensor.dtype().unwrap(), dtype);
                assert_eq!(tensor.device().unwrap(), device);
                assert_eq!(tensor.grad_enabled().unwrap(), grad_enabled);
                assert_eq!(tensor.to_vec::<u32>().unwrap(), data);
            }
            DType::I64 => {
                let (tensor, data) = generate_random_tensor_and_corresponding_data::<i64>(&shape, &dtype, &device, grad_enabled, data_seed);
                assert_eq!(tensor.shape().unwrap(), shape);
                assert_eq!(tensor.dtype().unwrap(), dtype);
                assert_eq!(tensor.device().unwrap(), device);
                assert_eq!(tensor.grad_enabled().unwrap(), grad_enabled);
                assert_eq!(tensor.to_vec::<i64>().unwrap(), data);
            }
            DType::F32 => {
                let (tensor, data) = generate_random_tensor_and_corresponding_data::<f32>(&shape, &dtype, &device, grad_enabled, data_seed);
                assert_eq!(tensor.shape().unwrap(), shape);
                assert_eq!(tensor.dtype().unwrap(), dtype);
                assert_eq!(tensor.device().unwrap(), device);
                assert_eq!(tensor.grad_enabled().unwrap(), grad_enabled);
                assert_eq!(tensor.to_vec::<f32>().unwrap(), data);
            }
            DType::F64 => {
                let (tensor, data) = generate_random_tensor_and_corresponding_data::<f64>(&shape, &dtype, &device, grad_enabled, data_seed);
                assert_eq!(tensor.shape().unwrap(), shape);
                assert_eq!(tensor.dtype().unwrap(), dtype);
                assert_eq!(tensor.device().unwrap(), device);
                assert_eq!(tensor.grad_enabled().unwrap(), grad_enabled);
                assert_eq!(tensor.to_vec::<f64>().unwrap(), data);
            }
            _ => {
                panic!("Unsupported dtype");
            }
        };
    }

    #[test]
    fn test_generate_random_shape_with_appropriate_size(
        seed in prop::num::u64::ANY,
    ) {
        let shape = generate_random_shape_with_appropriate_size(seed);
        let dims = shape.dims();
        let num_dim = dims.len();
        assert!(num_dim <= 4);

        let mut total_size: usize = 1;
        for &dim in dims {
            assert!(dim <= 100);
            total_size = total_size.saturating_mul(dim);
        }
        assert!(total_size <= 10000);
    }
}