dendritic_regression/
elastic_net.rs

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
use dendritic_ndarray::ndarray::NDArray;
use dendritic_ndarray::ops::*;
use dendritic_metrics::loss::*;
use dendritic_autodiff::node::{Node, Value};
use dendritic_autodiff::regularizers::*; 
use dendritic_autodiff::ops::*; 
use std::fs;


pub struct ElasticNet {
    pub features: Value<NDArray<f64>>,
    pub outputs: Value<NDArray<f64>>,
    pub weights: Value<NDArray<f64>>, 
    pub bias: Value<NDArray<f64>>,
    lambda: Value<NDArray<f64>>,
    learning_rate: f64,
    loss_function: fn(
        y_true: &NDArray<f64>, 
        y_pred: &NDArray<f64>) -> Result<f64, String>
}


impl ElasticNet {

    /// Create new instance of elastic net regression
    pub fn new(
        features: &NDArray<f64>, 
        y: &NDArray<f64>,
        lambda: f64, 
        learning_rate: f64) -> Result<ElasticNet, String> {

        if learning_rate < 0.0 || learning_rate > 1.0 {
            return Err("Learning rate must be between 1 and 0".to_string());
        }

        let weights = NDArray::new(vec![features.shape().dim(1), 1]).unwrap();
        let bias = NDArray::new(vec![1, 1]).unwrap();
        let inputs = Value::new(features); 
        let outputs = Value::new(y);

        let lambda_value: NDArray<f64> = NDArray::array(
            vec![1, 1], vec![lambda]
        ).unwrap();

        Ok(Self {
            features: inputs.clone(),
            outputs: outputs.clone(),
            weights: Value::new(&weights),
            bias: Value::new(&bias),
            lambda: Value::new(&lambda_value),
            learning_rate: learning_rate,
            loss_function: mse,
        })
    }


    /// Predict features for elastic net regression
    pub fn predict(&mut self, inputs: NDArray<f64>) -> NDArray<f64> {

        self.features = Value::new(&inputs); 

        let mut linear = ScaleAdd::new(
            Dot::new(self.features.clone(), self.weights.clone()),
            self.bias.clone()
        );

        linear.forward(); 
        linear.value() 
    }

    /// Save parameters for elastic net regression
    pub fn save(&self, filepath: &str) -> std::io::Result<()> {

        let weights_file = format!("{}/weights", filepath);
        let bias_path = format!("{}/bias", filepath); 
        fs::create_dir_all(filepath)?;

        self.weights.val().save(&weights_file).unwrap();
        self.bias.val().save(&bias_path).unwrap();

        Ok(())
    }


    /// Load parameters for elastic net regression
    pub fn load(
        filepath: &str, 
        features: &NDArray<f64>, 
        y: &NDArray<f64>, 
        learning_rate: f64,
        lambda: f64) -> std::io::Result<ElasticNet> {

        let weights_file = format!("{}/weights", filepath);
        let bias_path = format!("{}/bias", filepath); 

        let inputs = Value::new(features); 
        let outputs = Value::new(y);
        let load_weights = NDArray::load(&weights_file).unwrap();
        let load_bias = NDArray::load(&bias_path).unwrap();

        let lambda_value: NDArray<f64> = NDArray::array(
            vec![1, 1], vec![lambda]
        ).unwrap();

        Ok(ElasticNet {
            features: inputs.clone(),
            outputs: outputs.clone(),
            weights: Value::new(&load_weights),
            bias: Value::new(&load_bias),
            lambda: Value::new(&lambda_value),
            learning_rate: learning_rate,
            loss_function: mse
        })

    }


    /// Train elastic net regression model
    pub fn train(&mut self, epochs: usize, log_output: bool) {

        let mut linear = ScaleAdd::new(
            Dot::new(self.features.clone(), self.weights.clone()),
            self.bias.clone()
        );

        let mut l1_reg = L1Regularization::new(
            self.weights.clone(),
            self.lambda.clone(),
            self.learning_rate
        );

        let mut l2_reg = L2Regularization::new(
            self.weights.clone(),
            self.lambda.clone(),
            self.learning_rate
        );

        for epoch in 0..epochs {

            linear.forward();
            l1_reg.forward();
            l2_reg.forward();

            let y_pred = linear.value();
            let loss = (self.loss_function)(&self.outputs.val(), &y_pred).unwrap();
            let error = y_pred.subtract(self.outputs.val()).unwrap();
            let l1_error = error.scale_add(l1_reg.value()).unwrap();
            let l2_error = error.scale_add(l2_reg.value()).unwrap();

            linear.backward(error.clone()); 
            l1_reg.backward(l1_error.clone());
            l2_reg.backward(l2_error);

            let w_grad = self.features.grad().scalar_mult(
                self.learning_rate/y_pred.size() as f64
            ).unwrap();

            let w_update = self.weights.val().subtract(w_grad).unwrap();
            let l1_reg_w = l1_reg.grad(); 
            let l2_reg_w = l2_reg.grad(); 
            let reg_w = l1_reg_w.add(l2_reg_w).unwrap(); 
            let dw = w_update.subtract(reg_w).unwrap();
            self.weights.set_val(&dw);

            /* update biases */
            let b_collapse = self.bias.grad().sum_axis(1).unwrap();
            let db = b_collapse.scalar_mult(
                self.learning_rate/y_pred.size() as f64
            ).unwrap();
            self.bias.set_val(&db);

            if log_output {
                println!("Epoch [{:?}/{:?}]: {:?}", epoch, epochs, loss);
            }

        }

    }


    /// Train elastic net regression model with batch gradient descent
    pub fn sgd(&mut self, epochs: usize, log_output: bool, batch_size: usize) {

        let mut loss: f64 = 0.0;
        let x_train_binding = self.features.val();
        let y_train_binding = self.outputs.val();
        let x_train = x_train_binding.batch(batch_size).unwrap();
        let y_train = y_train_binding.batch(batch_size).unwrap();

        let mut linear = ScaleAdd::new(
            Dot::new(self.features.clone(), self.weights.clone()),
            self.bias.clone()
        );

        let mut l1_reg = L1Regularization::new(
            self.weights.clone(),
            self.lambda.clone(),
            self.learning_rate
        );

        let mut l2_reg = L2Regularization::new(
            self.weights.clone(),
            self.lambda.clone(),
            self.learning_rate
        );

        for epoch in 0..epochs {

            let mut batch_index = 0;
            for batch in &x_train {

                self.features.set_val(&batch);
                self.outputs.set_val(&y_train[batch_index]);

                linear.forward();
                l1_reg.forward();
                l2_reg.forward();

                let y_pred = linear.value();
                loss = (self.loss_function)(&self.outputs.val(), &y_pred).unwrap();
                let error = y_pred.subtract(self.outputs.val()).unwrap();
                let l1_error = error.scale_add(l1_reg.value()).unwrap();
                let l2_error = error.scale_add(l2_reg.value()).unwrap();

                linear.backward(error.clone()); 
                l1_reg.backward(l1_error);
                l2_reg.backward(l2_error);

                let w_grad = self.features.grad().scalar_mult(
                    self.learning_rate/y_pred.size() as f64
                ).unwrap();

                let w_update = self.weights.val().subtract(w_grad).unwrap();
                let l1_reg_w = l1_reg.grad(); 
                let l2_reg_w = l2_reg.grad(); 
                let reg_w = l1_reg_w.add(l2_reg_w).unwrap(); 
                let dw = w_update.subtract(reg_w).unwrap();
                self.weights.set_val(&dw);

                /* update biases */
                let b_collapse = self.bias.grad().sum_axis(1).unwrap();
                let db = b_collapse.scalar_mult(
                    self.learning_rate/y_pred.size() as f64
                ).unwrap();
                self.bias.set_val(&db);

                batch_index += 1; 

            }

            if log_output {
                println!("Epoch [{:?}/{:?}]: {:?}", epoch, epochs, loss);
            }

        }

    }

}