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
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259

//! sklearn parity tests for classification metrics
//!
//! These tests verify that our metrics match sklearn reference values
//! to within 1e-6 precision.
//!
//! Reference values computed with sklearn 1.4.0:
//! ```python
//! from sklearn.metrics import (accuracy_score, precision_score,
//!                              recall_score, f1_score, confusion_matrix)
//! ```

#[cfg(test)]
mod tests {
    use crate::eval::classification::{
        confusion_matrix, Average, ConfusionMatrix, MultiClassMetrics,
    };

    #[test]
    fn test_sklearn_parity_accuracy() {
        // sklearn: accuracy_score([0, 0, 1, 1, 2, 2, 0, 1, 2],
        //                         [0, 1, 1, 2, 2, 0, 0, 1, 2]) = 0.6666666666666666
        let y_true = vec![0, 0, 1, 1, 2, 2, 0, 1, 2];
        let y_pred = vec![0, 1, 1, 2, 2, 0, 0, 1, 2];

        let cm = confusion_matrix(&y_pred, &y_true);
        let acc = cm.accuracy();

        assert!(
            (acc - 0.6666666666666666).abs() < 1e-6,
            "Accuracy {acc} does not match sklearn reference 0.6666666666666666"
        );
    }

    #[test]
    fn test_sklearn_parity_precision_macro() {
        // sklearn: precision_score(..., average='macro') = 0.6666666666666666
        let y_true = vec![0, 0, 1, 1, 2, 2, 0, 1, 2];
        let y_pred = vec![0, 1, 1, 2, 2, 0, 0, 1, 2];

        let metrics = MultiClassMetrics::from_predictions(&y_pred, &y_true);
        let p_macro = metrics.precision_avg(Average::Macro);

        assert!(
            (p_macro - 0.6666666666666666).abs() < 1e-6,
            "Macro precision {p_macro} does not match sklearn reference"
        );
    }

    #[test]
    fn test_sklearn_parity_recall_macro() {
        // sklearn: recall_score(..., average='macro') = 0.6666666666666666
        let y_true = vec![0, 0, 1, 1, 2, 2, 0, 1, 2];
        let y_pred = vec![0, 1, 1, 2, 2, 0, 0, 1, 2];

        let metrics = MultiClassMetrics::from_predictions(&y_pred, &y_true);
        let r_macro = metrics.recall_avg(Average::Macro);

        assert!(
            (r_macro - 0.6666666666666666).abs() < 1e-6,
            "Macro recall {r_macro} does not match sklearn reference"
        );
    }

    #[test]
    fn test_sklearn_parity_f1_macro() {
        // sklearn: f1_score(..., average='macro') = 0.6666666666666666
        let y_true = vec![0, 0, 1, 1, 2, 2, 0, 1, 2];
        let y_pred = vec![0, 1, 1, 2, 2, 0, 0, 1, 2];

        let metrics = MultiClassMetrics::from_predictions(&y_pred, &y_true);
        let f1_macro = metrics.f1_avg(Average::Macro);

        assert!(
            (f1_macro - 0.6666666666666666).abs() < 1e-6,
            "Macro F1 {f1_macro} does not match sklearn reference"
        );
    }

    #[test]
    fn test_sklearn_parity_micro_averages() {
        // sklearn: For this dataset, micro = macro = 0.6666666666666666
        let y_true = vec![0, 0, 1, 1, 2, 2, 0, 1, 2];
        let y_pred = vec![0, 1, 1, 2, 2, 0, 0, 1, 2];

        let metrics = MultiClassMetrics::from_predictions(&y_pred, &y_true);

        let p_micro = metrics.precision_avg(Average::Micro);
        let r_micro = metrics.recall_avg(Average::Micro);
        let f1_micro = metrics.f1_avg(Average::Micro);

        assert!(
            (p_micro - 0.6666666666666666).abs() < 1e-6,
            "Micro precision {p_micro} does not match sklearn reference"
        );
        assert!(
            (r_micro - 0.6666666666666666).abs() < 1e-6,
            "Micro recall {r_micro} does not match sklearn reference"
        );
        assert!(
            (f1_micro - 0.6666666666666666).abs() < 1e-6,
            "Micro F1 {f1_micro} does not match sklearn reference"
        );
    }

    #[test]
    fn test_sklearn_parity_weighted_averages() {
        // sklearn: weighted averages = 0.6666666666666666 for balanced classes
        let y_true = vec![0, 0, 1, 1, 2, 2, 0, 1, 2];
        let y_pred = vec![0, 1, 1, 2, 2, 0, 0, 1, 2];

        let metrics = MultiClassMetrics::from_predictions(&y_pred, &y_true);

        let p_weighted = metrics.precision_avg(Average::Weighted);
        let r_weighted = metrics.recall_avg(Average::Weighted);
        let f1_weighted = metrics.f1_avg(Average::Weighted);

        assert!(
            (p_weighted - 0.6666666666666666).abs() < 1e-6,
            "Weighted precision {p_weighted} does not match sklearn reference"
        );
        assert!(
            (r_weighted - 0.6666666666666666).abs() < 1e-6,
            "Weighted recall {r_weighted} does not match sklearn reference"
        );
        assert!(
            (f1_weighted - 0.6666666666666666).abs() < 1e-6,
            "Weighted F1 {f1_weighted} does not match sklearn reference"
        );
    }

    #[test]
    fn test_sklearn_parity_imbalanced() {
        // Test with imbalanced classes
        // y_true = [0, 0, 0, 0, 0, 1, 1, 2]
        // y_pred = [0, 0, 0, 1, 1, 1, 0, 2]
        //
        // Manual calculation:
        // Class 0: TP=3, FP=1, FN=2 -> P=0.75, R=0.6, F1=0.6667
        // Class 1: TP=1, FP=2, FN=1 -> P=0.333, R=0.5, F1=0.4
        // Class 2: TP=1, FP=0, FN=0 -> P=1, R=1, F1=1
        // Macro F1 = (0.6667 + 0.4 + 1.0) / 3 = 0.6889
        let y_true = vec![0, 0, 0, 0, 0, 1, 1, 2];
        let y_pred = vec![0, 0, 0, 1, 1, 1, 0, 2];

        let cm = confusion_matrix(&y_pred, &y_true);
        let metrics = MultiClassMetrics::from_predictions(&y_pred, &y_true);

        // Accuracy: 5 correct out of 8
        assert!(
            (cm.accuracy() - 0.625).abs() < 1e-6,
            "Accuracy {} does not match reference 0.625",
            cm.accuracy()
        );

        // Macro F1: (0.6667 + 0.4 + 1.0) / 3 = 0.6889
        let f1_macro = metrics.f1_avg(Average::Macro);
        assert!(
            (f1_macro - 0.6888888888888888).abs() < 1e-6,
            "Macro F1 {f1_macro} does not match reference 0.6889"
        );
    }

    #[test]
    fn test_sklearn_parity_binary() {
        // Binary classification parity
        // y_true = [0, 0, 1, 1, 0, 1, 0, 1]
        // y_pred = [0, 1, 1, 0, 0, 1, 1, 1]
        //
        // Manual calculation:
        // Class 0: TP=2, FP=1, FN=2 -> P=2/3=0.6667, R=2/4=0.5, F1=0.5714
        // Class 1: TP=3, FP=2, FN=1 -> P=3/5=0.6, R=3/4=0.75, F1=0.6667
        // Macro F1 = (0.5714 + 0.6667) / 2 = 0.6190
        let y_true = vec![0, 0, 1, 1, 0, 1, 0, 1];
        let y_pred = vec![0, 1, 1, 0, 0, 1, 1, 1];

        let cm = confusion_matrix(&y_pred, &y_true);
        let metrics = MultiClassMetrics::from_predictions(&y_pred, &y_true);

        // Accuracy: 5 correct out of 8
        assert!(
            (cm.accuracy() - 0.625).abs() < 1e-6,
            "Accuracy {} does not match reference 0.625",
            cm.accuracy()
        );

        // Macro F1: (0.5714 + 0.6667) / 2 = 0.6190
        let f1_macro = metrics.f1_avg(Average::Macro);
        assert!(
            (f1_macro - 0.6190476190476191).abs() < 1e-6,
            "Macro F1 {f1_macro} does not match reference 0.6190"
        );
    }

    #[test]
    fn test_multiclass_metrics_average_none() {
        let y_pred = vec![0, 1, 1, 2, 0];
        let y_true = vec![0, 1, 0, 2, 1];
        let metrics = MultiClassMetrics::from_predictions(&y_pred, &y_true);

        // Average::None should behave like Macro (returns single value)
        let f1_none = metrics.f1_avg(Average::None);
        let f1_macro = metrics.f1_avg(Average::Macro);
        assert!((f1_none - f1_macro).abs() < 1e-6);
    }

    #[test]
    fn test_multiclass_metrics_empty_support() {
        // Test with all predictions correct - verify weighted average doesn't panic with 0 support
        let y_pred = vec![0, 0, 0];
        let y_true = vec![0, 0, 0];
        let metrics = MultiClassMetrics::from_predictions(&y_pred, &y_true);

        let weighted = metrics.f1_avg(Average::Weighted);
        assert!((weighted - 1.0).abs() < 1e-6);
    }

    #[test]
    fn test_confusion_matrix_from_confusion_matrix() {
        let cm = ConfusionMatrix::new(3);
        let metrics = MultiClassMetrics::from_confusion_matrix(&cm);

        // Empty matrix - all metrics should be 0
        assert_eq!(metrics.n_classes, 3);
        for i in 0..3 {
            assert!((metrics.precision[i] - 0.0).abs() < 1e-6);
            assert!((metrics.recall[i] - 0.0).abs() < 1e-6);
            assert!((metrics.f1[i] - 0.0).abs() < 1e-6);
        }
    }

    #[test]
    fn test_macro_average_empty_values() {
        let metrics = MultiClassMetrics {
            precision: vec![],
            recall: vec![],
            f1: vec![],
            support: vec![],
            n_classes: 0,
        };

        assert!((metrics.precision_avg(Average::Macro) - 0.0).abs() < 1e-6);
        assert!((metrics.recall_avg(Average::Macro) - 0.0).abs() < 1e-6);
        assert!((metrics.f1_avg(Average::Macro) - 0.0).abs() < 1e-6);
    }

    #[test]
    fn test_weighted_average_zero_support() {
        let metrics = MultiClassMetrics {
            precision: vec![0.5, 0.5],
            recall: vec![0.5, 0.5],
            f1: vec![0.5, 0.5],
            support: vec![0, 0], // Zero support
            n_classes: 2,
        };

        // Zero total support should return 0
        assert!((metrics.precision_avg(Average::Weighted) - 0.0).abs() < 1e-6);
    }
}