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
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313

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

use ndarray::Array5;
use rustyml::neural_network::layer::ReLU;
use rustyml::neural_network::layer::TrainingParameters;
use rustyml::neural_network::layer::activation_layer::sigmoid::Sigmoid;
use rustyml::neural_network::layer::activation_layer::tanh::Tanh;
use rustyml::neural_network::layer::convolution_layer::PaddingType;
use rustyml::neural_network::layer::convolution_layer::conv_3d::Conv3D;
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::rms_prop::RMSprop;
use rustyml::neural_network::optimizer::sgd::SGD;
use rustyml::neural_network::sequential::Sequential;

#[test]
fn test_conv3d_sequential_with_sgd() {
    // Create a 5D input tensor: [batch_size, channels, depth, height, width]
    let x = Array5::ones((2, 1, 8, 8, 8)).into_dyn();
    // Create a target tensor - assume output size is 6x6x6 (Valid padding)
    let y = Array5::ones((2, 3, 6, 6, 6)).into_dyn();

    // Build the model
    let mut model = Sequential::new();
    model
        .add(
            Conv3D::new(
                3,                   // filters
                (3, 3, 3),           // kernel_size
                vec![2, 1, 8, 8, 8], // input_shape
                (1, 1, 1),           // stride
                PaddingType::Valid,  // padding
                ReLU::new(),         // activation
            )
            .unwrap(),
        )
        .compile(SGD::new(0.01).unwrap(), MeanSquaredError::new());

    // Print the model structure
    model.summary();

    // Train the model
    let result = model.fit(&x, &y, 3);
    assert!(result.is_ok());

    // Make predictions
    let prediction = model.predict(&x).unwrap();
    assert_eq!(prediction.shape(), &[2, 3, 6, 6, 6]);

    // Verify that the predictions are non-negative (ReLU activation function)
    for value in prediction.iter() {
        assert!(*value >= 0.0);
    }
}

#[test]
fn test_conv3d_sequential_with_rmsprop() {
    // Create more complex training data
    let x = Array5::from_shape_fn((2, 2, 6, 6, 6), |(b, c, d, h, w)| {
        ((b * 2 + c * 3 + d + h + w) as f32).sin() * 0.5
    })
    .into_dyn();

    let y = Array5::zeros((2, 2, 4, 4, 4)).into_dyn(); // Valid padding output

    // Build the model
    let mut model = Sequential::new();
    model
        .add(
            Conv3D::new(
                2,                   // filters
                (3, 3, 3),           // kernel_size
                vec![2, 2, 6, 6, 6], // input_shape
                (1, 1, 1),           // stride
                PaddingType::Valid,  // padding
                Tanh::new(),         // activation
            )
            .unwrap(),
        )
        .compile(
            RMSprop::new(0.001, 0.9, 1e-8).unwrap(),
            MeanSquaredError::new(),
        );

    model.summary();

    // Train the model
    let result = model.fit(&x, &y, 4);
    assert!(result.is_ok());

    // Make predictions
    let prediction = model.predict(&x).unwrap();
    assert_eq!(prediction.shape(), &[2, 2, 4, 4, 4]);

    // Verify that Tanh output is within [-1, 1]
    for value in prediction.iter() {
        assert!(*value >= -1.0 && *value <= 1.0);
    }
}

#[test]
fn test_conv3d_different_strides() {
    let x = Array5::ones((1, 1, 10, 10, 10)).into_dyn();

    // Test with different stride values
    let stride_2_conv = Conv3D::new(
        1,
        (3, 3, 3),
        vec![1, 1, 10, 10, 10],
        (2, 2, 2), // stride = 2
        PaddingType::Valid,
        ReLU::new(),
    )
    .unwrap();

    let mut model = Sequential::new();
    model
        .add(stride_2_conv)
        .compile(SGD::new(0.01).unwrap(), MeanSquaredError::new());

    let prediction = model.predict(&x).unwrap();
    assert_eq!(prediction.shape(), &[1, 1, 4, 4, 4]);
}

#[test]
fn test_conv3d_multiple_channels() {
    // Test multi-channel input
    let x = Array5::from_shape_fn((2, 3, 6, 6, 6), |(b, c, d, h, w)| {
        (b + c + d + h + w) as f32 * 0.1
    })
    .into_dyn();

    let y = Array5::ones((2, 5, 4, 4, 4)).into_dyn();

    let mut model = Sequential::new();
    model
        .add(
            Conv3D::new(
                5,                   // filters
                (3, 3, 3),           // kernel_size
                vec![2, 3, 6, 6, 6], // input_shape (3 channels)
                (1, 1, 1),           // stride
                PaddingType::Valid,  // padding
                ReLU::new(),         // activation
            )
            .unwrap(),
        )
        .compile(SGD::new(0.01).unwrap(), MeanSquaredError::new());

    model.summary();

    let result = model.fit(&x, &y, 2);
    assert!(result.is_ok());

    let prediction = model.predict(&x).unwrap();
    assert_eq!(prediction.shape(), &[2, 5, 4, 4, 4]);
}

#[test]
fn test_conv3d_activation_functions() {
    let x = Array5::from_shape_fn((1, 1, 4, 4, 4), |(_, _, d, _, _)| {
        d as f32 - 2.0 // Generate negative values to test activation functions
    })
    .into_dyn();

    // Test ReLU activation function
    let mut relu_model = Sequential::new();
    relu_model
        .add(
            Conv3D::new(
                1,
                (2, 2, 2),
                vec![1, 1, 4, 4, 4],
                (1, 1, 1),
                PaddingType::Valid,
                ReLU::new(),
            )
            .unwrap(),
        )
        .compile(SGD::new(0.01).unwrap(), MeanSquaredError::new());

    let relu_output = relu_model.predict(&x).unwrap();
    // ReLU output should be non-negative
    for value in relu_output.iter() {
        assert!(*value >= 0.0);
    }

    // Test Sigmoid activation function
    let mut sigmoid_model = Sequential::new();
    sigmoid_model
        .add(
            Conv3D::new(
                1,
                (2, 2, 2),
                vec![1, 1, 4, 4, 4],
                (1, 1, 1),
                PaddingType::Valid,
                Sigmoid::new(),
            )
            .unwrap(),
        )
        .compile(SGD::new(0.01).unwrap(), MeanSquaredError::new());

    let sigmoid_output = sigmoid_model.predict(&x).unwrap();
    // Sigmoid output should be within [0, 1]
    for value in sigmoid_output.iter() {
        assert!(*value >= 0.0 && *value <= 1.0);
    }
}

#[test]
fn test_conv3d_parameter_count() {
    let conv3d = Conv3D::new(
        4,                   // filters
        (3, 3, 3),           // kernel_size
        vec![2, 2, 5, 5, 5], // input_shape (2 channels)
        (1, 1, 1),
        PaddingType::Valid,
        ReLU::new(),
    )
    .unwrap();

    // Parameter count = weights + bias = (4 * 2 * 3 * 3 * 3) + (1 * 4) = 216 + 4 = 220
    assert_eq!(conv3d.param_count(), TrainingParameters::Trainable(220));
}

#[test]
fn test_conv3d_same_padding() {
    let x = Array5::ones((1, 1, 8, 8, 8)).into_dyn();

    // Test with Same padding
    let mut model = Sequential::new();
    model
        .add(
            Conv3D::new(
                2,
                (3, 3, 3),
                vec![1, 1, 8, 8, 8],
                (1, 1, 1),
                PaddingType::Same,
                ReLU::new(),
            )
            .unwrap(),
        )
        .compile(SGD::new(0.01).unwrap(), MeanSquaredError::new());

    let prediction = model.predict(&x).unwrap();
    // With Same padding and stride 1, output should have the same spatial dimensions
    assert_eq!(prediction.shape(), &[1, 2, 8, 8, 8]);
}

#[test]
fn test_conv3d_with_adam() {
    // Test Conv3D with Adam optimizer
    let x = Array5::from_shape_fn((2, 1, 6, 6, 6), |(b, _, d, h, w)| {
        ((b + d + h + w) as f32) * 0.1
    })
    .into_dyn();

    let y = Array5::ones((2, 2, 4, 4, 4)).into_dyn();

    let mut model = Sequential::new();
    model
        .add(
            Conv3D::new(
                2,
                (3, 3, 3),
                vec![2, 1, 6, 6, 6],
                (1, 1, 1),
                PaddingType::Valid,
                ReLU::new(),
            )
            .unwrap(),
        )
        .compile(
            Adam::new(0.001, 0.9, 0.999, 1e-8).unwrap(),
            MeanSquaredError::new(),
        );

    model.summary();

    let result = model.fit(&x, &y, 3);
    assert!(result.is_ok());

    let prediction = model.predict(&x).unwrap();
    assert_eq!(prediction.shape(), &[2, 2, 4, 4, 4]);
}

#[test]
fn test_conv3d_asymmetric_stride() {
    // Test with asymmetric strides
    let x = Array5::ones((1, 1, 12, 8, 10)).into_dyn();

    let mut model = Sequential::new();
    model
        .add(
            Conv3D::new(
                1,
                (3, 3, 3),
                vec![1, 1, 12, 8, 10],
                (2, 1, 2), // Different strides for each dimension
                PaddingType::Valid,
                ReLU::new(),
            )
            .unwrap(),
        )
        .compile(SGD::new(0.01).unwrap(), MeanSquaredError::new());

    let prediction = model.predict(&x).unwrap();
    // Output: depth = (12-3)/2+1 = 5, height = (8-3)/1+1 = 6, width = (10-3)/2+1 = 4
    assert_eq!(prediction.shape(), &[1, 1, 5, 6, 4]);
}