rustyml 0.11.0

A high-performance machine learning & deep learning library in pure Rust, offering ML algorithms and neural network support
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

#![cfg(feature = "neural_network")]

use ndarray::{Array2, Array3, Array4, Array5};
use rustyml::neural_network::layer::activation_layer::relu::ReLU;
use rustyml::neural_network::layer::activation_layer::sigmoid::Sigmoid;
use rustyml::neural_network::layer::activation_layer::softmax::Softmax;
use rustyml::neural_network::layer::convolution_layer::PaddingType;
use rustyml::neural_network::layer::convolution_layer::conv_2d::Conv2D;
use rustyml::neural_network::layer::dense::Dense;
use rustyml::neural_network::layer::flatten::Flatten;
use rustyml::neural_network::layer::pooling_layer::max_pooling_2d::MaxPooling2D;
use rustyml::neural_network::loss_function::categorical_cross_entropy::CategoricalCrossEntropy;
use rustyml::neural_network::loss_function::mean_squared_error::MeanSquaredError;
use rustyml::neural_network::neural_network_trait::Layer;
use rustyml::neural_network::optimizer::adam::Adam;
use rustyml::neural_network::optimizer::sgd::SGD;
use rustyml::neural_network::sequential::Sequential;

#[test]
fn test_flatten_in_sequential() {
    // Create a simple 4D input tensor: [batch_size, channels, height, width]
    // Batch size=2, 3 channels, each 4x4 pixels
    let x = Array4::ones((2, 3, 4, 4)).into_dyn();

    // Create target tensor - assuming final output is a single value
    let y = Array2::ones((2, 1)).into_dyn();

    // Build model: convolution layer -> max pooling layer -> flatten layer -> fully connected layer
    let mut model = Sequential::new();
    model
        .add(
            Conv2D::new(
                6,                  // Number of filters
                (3, 3),             // Kernel size
                vec![2, 3, 4, 4],   // Input shape
                (1, 1),             // Stride
                PaddingType::Valid, // No padding
                ReLU::new(),        // ReLU activation function
            )
            .unwrap(),
        )
        .add(
            MaxPooling2D::new(
                (2, 2),           // Pooling window size
                vec![2, 6, 2, 2], // Input shape (after convolution)
                None,             // Use default stride
            )
            .unwrap(),
        )
        .add(Flatten::new(vec![2, 6, 1, 1]).unwrap()) // Flatten layer
        .add(Dense::new(6, 1, Sigmoid::new()).unwrap()) // Fully connected layer
        .compile(SGD::new(0.01).unwrap(), MeanSquaredError::new());

    // Print model structure
    model.summary();

    // Train the model
    model.fit(&x, &y, 3).unwrap();

    // Use predict for forward propagation prediction
    let prediction = model.predict(&x).unwrap();
    println!("CNN+Flatten prediction results: {:?}", prediction);

    // Check if output shape is correct
    assert_eq!(prediction.shape(), &[2, 1]);
}

#[test]
fn test_flatten_only_model() {
    // Create 4D input tensor
    let x = Array4::from_shape_fn((2, 3, 4, 4), |(b, c, h, w)| {
        (b * 100 + c * 10 + h + w) as f32 / 10.0
    })
    .into_dyn();

    // Target tensor - already flattened format
    let flattened_size = 3 * 4 * 4; // channels * height * width

    // Create model with only Flatten layer
    let mut model = Sequential::new();
    model
        .add(Flatten::new(vec![2, 3, 4, 4]).unwrap())
        .compile(SGD::new(0.01).unwrap(), MeanSquaredError::new());

    // Print model structure
    model.summary();

    // Perform forward propagation
    let output = model.predict(&x).unwrap();

    // Check output shape
    assert_eq!(output.shape(), &[2, flattened_size]);
}

#[test]
fn test_multiple_layers_with_flatten() {
    // Create 4D input tensor
    let x = Array4::ones((2, 1, 8, 8)).into_dyn();

    // Create target tensor
    let y = Array2::ones((2, 10)).into_dyn();

    // Build model with multiple layers including two convolution layers, one pooling layer,
    // one flatten layer and two dense layers
    let mut model = Sequential::new();
    model
        .add(
            Conv2D::new(
                8,                 // 8 filters
                (3, 3),            // 3x3 kernel
                vec![2, 1, 8, 8],  // Input shape
                (1, 1),            // Stride
                PaddingType::Same, // Same padding
                ReLU::new(),       // ReLU activation
            )
            .unwrap(),
        )
        .add(
            Conv2D::new(
                16,                 // 16 filters
                (3, 3),             // 3x3 kernel
                vec![2, 8, 8, 8],   // Input shape (after first convolution)
                (1, 1),             // Stride
                PaddingType::Valid, // No padding
                ReLU::new(),        // ReLU activation
            )
            .unwrap(),
        )
        .add(
            MaxPooling2D::new(
                (2, 2),            // 2x2 pooling window
                vec![2, 16, 6, 6], // Input shape (after second convolution)
                Some((2, 2)),      // 2x2 stride
            )
            .unwrap(),
        )
        .add(Flatten::new(vec![2, 16, 3, 3]).unwrap()) // Flatten layer
        .add(Dense::new(16 * 3 * 3, 20, ReLU::new()).unwrap()) // Fully connected layer
        .add(Dense::new(20, 10, Softmax::new()).unwrap()) // Output layer
        .compile(
            Adam::new(0.001, 0.9, 0.999, 1e-8).unwrap(),
            CategoricalCrossEntropy::new(),
        );

    // Print model structure
    model.summary();

    // Train the model
    model.fit(&x, &y, 3).unwrap();

    // Forward propagation prediction
    let prediction = model.predict(&x).unwrap();
    println!(
        "Complex CNN with Flatten prediction results: {:?}",
        prediction
    );

    // Check output shape
    assert_eq!(prediction.shape(), &[2, 10]);
}

#[test]
fn test_flatten_3d() {
    let input = Array3::ones((2, 10, 5)).into_dyn(); // batch=2, features=10, length=5
    let mut flatten = Flatten::new(vec![2, 10, 5]).unwrap();

    let output = flatten.forward(&input).unwrap();
    assert_eq!(output.shape(), &[2, 50]); // 10 * 5 = 50

    // Test backward pass
    let grad_output = Array2::ones((2, 50)).into_dyn();
    let grad_input = flatten.backward(&grad_output).unwrap();
    assert_eq!(grad_input.shape(), input.shape());
}

#[test]
fn test_flatten_4d() {
    let input = Array4::ones((2, 3, 4, 4)).into_dyn(); // batch=2, channels=3, height=4, width=4
    let mut flatten = Flatten::new(vec![2, 3, 4, 4]).unwrap();

    let output = flatten.forward(&input).unwrap();
    assert_eq!(output.shape(), &[2, 48]); // 3 * 4 * 4 = 48

    // Test backward pass
    let grad_output = Array2::ones((2, 48)).into_dyn();
    let grad_input = flatten.backward(&grad_output).unwrap();
    assert_eq!(grad_input.shape(), input.shape());
}

#[test]
fn test_flatten_5d() {
    let input = Array5::ones((2, 3, 4, 8, 8)).into_dyn(); // batch=2, channels=3, depth=4, height=8, width=8
    let mut flatten = Flatten::new(vec![2, 3, 4, 8, 8]).unwrap();

    let output = flatten.forward(&input).unwrap();
    assert_eq!(output.shape(), &[2, 768]); // 3 * 4 * 8 * 8 = 768

    // Test backward pass
    let grad_output = Array2::ones((2, 768)).into_dyn();
    let grad_input = flatten.backward(&grad_output).unwrap();
    assert_eq!(grad_input.shape(), input.shape());
}