entrenar 0.7.13

Training & Optimization library with autograd, LoRA, quantization, and model merging
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

//! Tests for TPE optimizer

#![allow(clippy::module_inception)]
#[cfg(test)]
mod tests {
    use crate::optim::hpo::error::HPOError;
    use crate::optim::hpo::tpe::TPEOptimizer;
    use crate::optim::hpo::types::{HyperparameterSpace, ParameterDomain};

    #[test]
    fn test_tpe_new() {
        let space = HyperparameterSpace::new();
        let tpe = TPEOptimizer::new(space);
        assert_eq!(tpe.n_trials(), 0);
        assert!(tpe.best_trial().is_none());
    }

    #[test]
    fn test_tpe_suggest_empty_space() {
        let space = HyperparameterSpace::new();
        let mut tpe = TPEOptimizer::new(space);
        let result = tpe.suggest();
        assert!(matches!(result, Err(HPOError::EmptySpace)));
    }

    #[test]
    fn test_tpe_suggest_startup() {
        let mut space = HyperparameterSpace::new();
        space.add("lr", ParameterDomain::Continuous { low: 1e-5, high: 1e-1, log_scale: true });

        let mut tpe = TPEOptimizer::new(space).with_startup(5);

        // First suggestions should work (startup phase)
        for _i in 0..5 {
            let trial = tpe.suggest().expect("operation should succeed");
            assert!(trial.config.contains_key("lr"));
        }
    }

    #[test]
    fn test_tpe_record_and_best() {
        let mut space = HyperparameterSpace::new();
        space.add("lr", ParameterDomain::Continuous { low: 0.0, high: 1.0, log_scale: false });

        let mut tpe = TPEOptimizer::new(space);

        let trial1 = tpe.suggest().expect("operation should succeed");
        tpe.record(trial1, 0.5, 10);

        let trial2 = tpe.suggest().expect("operation should succeed");
        tpe.record(trial2, 0.3, 10);

        assert_eq!(tpe.n_trials(), 2);
        let best = tpe.best_trial().expect("operation should succeed");
        assert_eq!(best.score, 0.3);
    }

    #[test]
    fn test_tpe_with_gamma() {
        let space = HyperparameterSpace::new();
        let tpe = TPEOptimizer::new(space).with_gamma(0.15);
        assert!((tpe.gamma - 0.15).abs() < 1e-10);
    }

    #[test]
    fn test_tpe_guided_sampling() {
        let mut space = HyperparameterSpace::new();
        space.add("x", ParameterDomain::Continuous { low: 0.0, high: 10.0, log_scale: false });

        let mut tpe = TPEOptimizer::new(space).with_startup(5);

        // Run startup phase
        for _i in 0..5 {
            let trial = tpe.suggest().expect("operation should succeed");
            // Lower x values get better scores
            let score = trial
                .config
                .get("x")
                .expect("key should exist")
                .as_float()
                .expect("key should exist");
            tpe.record(trial, score, 10);
        }

        // After startup, TPE should suggest values closer to 0
        // (where scores are better in our mock objective)
        assert_eq!(tpe.n_trials(), 5);
    }

    #[test]
    fn test_tpe_record_failed() {
        let mut space = HyperparameterSpace::new();
        space.add("lr", ParameterDomain::Continuous { low: 0.0, high: 1.0, log_scale: false });

        let mut tpe = TPEOptimizer::new(space);
        let trial = tpe.suggest().expect("operation should succeed");
        tpe.record_failed(trial);

        // Failed trials shouldn't count as completed
        assert_eq!(tpe.n_trials(), 0);
    }

    #[test]
    fn test_tpe_tpe_sampling_with_trials() {
        let mut space = HyperparameterSpace::new();
        space.add("lr", ParameterDomain::Continuous { low: 0.0, high: 1.0, log_scale: false });
        space.add("batch_size", ParameterDomain::Discrete { low: 8, high: 32 });
        space.add(
            "activation",
            ParameterDomain::Categorical { choices: vec!["relu".to_string(), "gelu".to_string()] },
        );

        let mut tpe = TPEOptimizer::new(space).with_startup(3);

        // Run startup phase
        for _ in 0..3 {
            let trial = tpe.suggest().expect("operation should succeed");
            let lr = trial
                .config
                .get("lr")
                .expect("key should exist")
                .as_float()
                .expect("key should exist");
            tpe.record(trial, lr, 10); // Score equals lr
        }

        // Now TPE sampling kicks in
        for _ in 0..5 {
            let trial = tpe.suggest().expect("operation should succeed");
            assert!(trial.config.contains_key("lr"));
            assert!(trial.config.contains_key("batch_size"));
            assert!(trial.config.contains_key("activation"));
            let lr = trial
                .config
                .get("lr")
                .expect("key should exist")
                .as_float()
                .expect("key should exist");
            tpe.record(trial, lr, 10);
        }

        assert_eq!(tpe.n_trials(), 8);
    }
}

#[cfg(test)]
mod sampling_tests {
    use super::super::sampling::{
        count_categorical, sample_ei_ratio_continuous, sample_ei_ratio_discrete,
    };
    use crate::optim::hpo::types::{ParameterValue, Trial};

    #[test]
    fn test_sample_ei_ratio_continuous_empty() {
        let mut rng = rand::rng();
        let value = sample_ei_ratio_continuous(&[], &[], 0.0, 1.0, 1.0, &mut rng);
        assert!((0.0..=1.0).contains(&value));
    }

    #[test]
    fn test_sample_ei_ratio_continuous_with_data() {
        let mut rng = rand::rng();
        let good = vec![0.1, 0.2, 0.15];
        let bad = vec![0.8, 0.9, 0.85];
        let value = sample_ei_ratio_continuous(&good, &bad, 0.0, 1.0, 0.1, &mut rng);
        assert!((0.0..=1.0).contains(&value));
    }

    #[test]
    fn test_sample_ei_ratio_discrete_empty() {
        let mut rng = rand::rng();
        let value = sample_ei_ratio_discrete(&[], &[], 1, 10, &mut rng);
        assert!((1..=10).contains(&value));
    }

    #[test]
    fn test_sample_ei_ratio_discrete_with_data() {
        let mut rng = rand::rng();
        let good = vec![2, 3, 2];
        let bad = vec![8, 9, 10];
        let value = sample_ei_ratio_discrete(&good, &bad, 1, 10, &mut rng);
        assert!((1..=10).contains(&value));
    }

    #[test]
    fn test_count_categorical_empty() {
        let trials: Vec<&Trial> = vec![];
        let choices = vec!["a".to_string(), "b".to_string()];
        let counts = count_categorical("opt", &trials, &choices);
        assert_eq!(counts, vec![0, 0]);
    }

    #[test]
    fn test_count_categorical_with_trials() {
        use std::collections::HashMap;

        let mut config1 = HashMap::new();
        config1.insert("opt".to_string(), ParameterValue::Categorical("a".to_string()));
        let trial1 = Trial::new(0, config1);

        let mut config2 = HashMap::new();
        config2.insert("opt".to_string(), ParameterValue::Categorical("b".to_string()));
        let trial2 = Trial::new(1, config2);

        let mut config3 = HashMap::new();
        config3.insert("opt".to_string(), ParameterValue::Categorical("a".to_string()));
        let trial3 = Trial::new(2, config3);

        let trials: Vec<&Trial> = vec![&trial1, &trial2, &trial3];
        let choices = vec!["a".to_string(), "b".to_string()];
        let counts = count_categorical("opt", &trials, &choices);
        assert_eq!(counts, vec![2, 1]);
    }
}

#[cfg(test)]
mod property_tests {
    use crate::optim::hpo::tpe::TPEOptimizer;
    use crate::optim::hpo::types::{HyperparameterSpace, ParameterDomain};
    use proptest::prelude::*;

    proptest! {
        #![proptest_config(ProptestConfig::with_cases(200))]

        #[test]
        fn prop_tpe_trials_increment(n_trials in 1usize..20) {
            let mut space = HyperparameterSpace::new();
            space.add("x", ParameterDomain::Continuous {
                low: 0.0,
                high: 1.0,
                log_scale: false,
            });

            let mut tpe = TPEOptimizer::new(space);
            for i in 0..n_trials {
                let trial = tpe.suggest().expect("operation should succeed");
                let score = (i as f64) / 10.0;
                tpe.record(trial, score, 10);
            }
            prop_assert_eq!(tpe.n_trials(), n_trials);
        }
    }
}