gtensor 1.0.0

Reverse-mode autodifferentiation of computational graphs with tensors and more for machine learning.
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


use gtensor as gt;

use rand::Rng;

// batch size
const N: usize = 10;

fn main() {
    // load a train and test dataset
    let train = load_dataset();
    let test = load_dataset();

    // draw the datasets for demonstration purposes
    draw_dataset(&train, "examples/class-train-dataset.bmp");
    draw_dataset(&test, "examples/class-test-dataset.bmp");

    // construct the tape and configure the batch size to N. 
    let mut tape = build_tape();
    tape.set_batch_size(N);

    // declare tensors to store error and gradient data.
    let mut error = gt::Tensor::from_fill([N], 0.0);
    let mut grad = gt::Tensor::from_fill([N,1], 0.0);

    // Use the training dataset to train the network (a.k.a tape).
    // each epoch is a full iteration of the dataset.
    for epoch in 0..50 {
        let mut loss = 0.0;
        for (feature, label) in train.iter_batched(N) {
            // 'tape.forward' returns an RwLockReadGuard, so
            // we add a block here so the guard goes out of scope
            // before we call 'tape.backward'. If you remove the block,
            // the compiler will error on 'tape.backward'. 
            {
                // get the prediction from the network.
                let prediction = tape.forward(feature);

                gt::math::loss::mse(
                    label,
                    prediction.slice(),
                    &mut error,
                    &mut grad,
                );
            }

            // add the loss for each batch to the total loss
            error.iter().for_each(|x| loss += *x);

            // Execute the backward pass
            tape.backward(grad.slice())
        }

        // Print the epoch number and the average loss, 
        // dividing by the number of features in the dataset.
        println!("Epoch: {epoch}, loss: {}", loss / 200.)
    }

    // Use the testing dataset to test the tape. 
    let mut loss = 0.0;
    for (feature, label) in test.iter_batched(N) {
        let prediction = tape.forward(feature);

        gt::math::loss::mse(
            label,
            prediction.slice(),
            &mut error,
            &mut grad,
        );

        // add the loss for each batch to the total loss
        error.iter().for_each(|x| loss += *x);
    }

    println!("Test Loss: {}", loss / 200.);

    draw_prediction(&mut tape, &train, "examples/class-prediction.bmp");

    tape.save("examples/class").unwrap();
}

/// Create a dataset
fn load_dataset() -> gt::Dataset {
    // 2 inputs/features, 1 truth label
    let mut data = gt::Dataset::new([2], [1]);

    let mut rng = rand::thread_rng();

    for _ in 0..200 {
        // generate random points in the range (-1,1)
        let x = rng.gen_range(-1.0..1.0);
        let y = rng.gen_range(-1.0..1.0);

        // distance from (x,y) to the origin at (0,0).
        let d = f32::sqrt((x*x)+(y*y));
        
        // discretive the labels for classification.
        // if the distance is greater than 0.6, the
        // label is orange (1.0) and blue (0.0) otherwise.
        let d = if d > 0.6 { 1.0 } else { 0.0 };

        // load the feature/label into the dataset
        data.load_feature(&[x,y], &[d]);
    }

    data
}

/// Record Operators to the Tape.
fn build_tape() -> gt::Tape {
    let mut tape = gt::Tape::builder();

    // set the optimizer and initializer for the weights.
    tape.opt = gt::opt::momentum(0.04, 0.9);
    tape.init = gt::init::normal(0.5, 1.0);

    // input
    let x = tape.input([2]);
    
    // first layer (2 inputs, 4 neurons)
    // 1. declare weight parameters (2x4)
    // 2. declare bias parameters (4)
    // 3. matmul x * w (Nx2 * 2x4 = Nx4)
    // 4. add bias to the channels
    // 5. activate with tanh
    let w = tape.parameter([2,4]);
    let b = tape.parameter([4]);
    let x = gt::op::matmul(x, w);
    let x = gt::op::axis_add(x, b, 'C');
    let x = gt::op::tanh(x);

    // second layer (4 inputs, 4 neurons)
    // 1. declare weight parameters (4x4)
    // 2. declare bias parameters (4)
    // 3. matmul x * w (Nx4 * 4x4 = Nx4)
    // 4. add bias to the channels
    // 5. activate with tanh
    let w = tape.parameter([4,4]);
    let b = tape.parameter([4]);
    let x = gt::op::matmul(x,w);
    let x = gt::op::axis_add(x, b, 'C');
    let x = gt::op::tanh(x);

    // output layer (4 inputs, 1 neuron)
    // 1. declare weight parameters (4x1)
    // 2. declare bias parameters (1)
    // 3. matmul x * w (Nx4 * 4x1 = Nx1)
    // 4. add bias to the channels
    // 5. activate with tanh
    let w = tape.parameter([4,1]);
    let b = tape.parameter([1]);
    let x = gt::op::matmul(x, w);
    let x = gt::op::axis_add(x, b, 'C');
    let _ = gt::op::tanh(x);

    tape.finish()
}

/// Create a graphic of the dataset (what the neural net trains on).
pub fn draw_dataset(dataset: &gt::Dataset, name: &str) {
    let mut img = bmp::Image::new(200,200);

    for x in 0..200 {
        for y in 0..200 {
            img.set_pixel(x, y, bmp::Pixel::new(255,255,255))
        }
    }

    for (feature, label) in dataset.iter_batched(1) {
        let color = 
        if label[0] < 0.5 {
            bmp::Pixel::new(60, 165, 255)
        } else {
            bmp::Pixel::new(255, 165, 0)
        };

        let x = ((feature[0] + 1.0) * 100.) as u32;
        let y = ((feature[1] + 1.0) * 100.) as u32;

        img.set_pixel(x, y, color)
    }

    img.save(name).unwrap()
}

/// Draw the neural networks' predictions.
pub fn draw_prediction(tape: &mut gt::Tape, dataset: &gt::Dataset, name: &str) {
    let mut img = bmp::Image::new(200,200);

    // change the batch size to 1 for easier iteration. 
    tape.set_batch_size(1);

    // run the neural network for every pixel
    for x in 0..200 {
        for y in 0..200 {
            // convert x and y to the range (-1,1)
            let xf = (x as f32 / 100.) - 1.;
            let yf = (y as f32 / 100.) - 1.;

            // convert inputs to tensor (for tape)
            let input = gt::Tensor::from_slice([1,2], &[xf,yf]);

            // get the prediction
            let prediction = tape.forward(input.slice());   

            // the color indicated by the prediction.
            let color = 
            if prediction[0] < 0.5 {
                bmp::Pixel::new(60, 165, 255)
            } else {
                bmp::Pixel::new(255, 165, 0)
            };
            
            img.set_pixel(x, y, color)
        }
    }

    // draw the training dataset slightly darker
    for (feature, label) in dataset.iter_batched(1) {
        let color = 
        if label[0] < 0.5 {
            bmp::Pixel::new(30, 135, 225)
        } else {
            bmp::Pixel::new(225, 135, 0)
        };

        let x = ((feature[0] + 1.0) * 100.) as u32;
        let y = ((feature[1] + 1.0) * 100.) as u32;

        img.set_pixel(x, y, color)
    }

    img.save(name).unwrap();
}