runn 0.1.1

Runn is a feature-rich, easy-to-use library for building, training, and evaluating feed-forward neural networks in Rust
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

use std::collections::HashMap;

use crate::{matrix::DMat, util};

use super::{MetricEvaluator, Metrics};

pub struct ClassificationMetrics {
    pub accuracy: f32,
    pub micro_precision: f32,
    pub micro_recall: f32,
    pub macro_f1_score: f32,
    pub micro_f1_score: f32,
    pub metrics_by_class: Vec<ClassMetrics>,
}

pub struct ClassMetrics {
    pub f1_score: f32,
    pub recall: f32,
    pub precision: f32,
}

impl ClassificationMetrics {
    pub fn display(&self) -> String {
        format!(
            "Classification Metrics: Accuracy:{:.4}, Micro Precision:{:.4}, Micro Recall:{:.4}, Macro F1 Score:{:.4}, Micro F1 Score:{:.4}\n  Metrics by Class:\n{}",
            self.accuracy * 100.0,
            self.micro_precision,
            self.micro_recall,
            self.macro_f1_score,
            self.micro_f1_score,
            self.metrics_by_class
                .iter()
                .enumerate()
                .map(|(i, metric)| {
                    format!(
                        "    Class {}:    Precision:{:.4}    Recall:{:.4}    F1 Score:{:.4}\n",
                        i, metric.precision, metric.recall, metric.f1_score
                    )
                })
                .collect::<String>()
        )
    }

    pub(crate) fn headers(&self) -> Vec<&'static str> {
        vec!["Accuracy"]
    }

    pub(crate) fn values_str(&self) -> Vec<String> {
        vec![format!("{:.5}", self.accuracy * 100.0)]
    }

    pub(crate) fn values(&self) -> Vec<f32> {
        vec![self.accuracy * 100.0]
    }
}

pub(crate) struct ClassificationEvaluator;

impl MetricEvaluator for ClassificationEvaluator {
    fn evaluate(&self, targets: &DMat, predictions: &DMat) -> Metrics {
        let classification_metrics = calculate_classification_metrics(targets, predictions);
        Metrics::Classification(classification_metrics)
    }
}

fn calculate_classification_metrics(targets: &DMat, predictions: &DMat) -> ClassificationMetrics {
    let (true_positives_map, false_positives_map, false_negatives_map) =
        calculate_confusion_matrix(targets, predictions);

    // Initialize sums for micro-average calculations
    let mut sum_tp = 0;
    let mut sum_fp = 0;
    let mut sum_fn = 0;

    // Initialize sums for macro-average calculations
    let mut sum_f1_macro = 0.0;

    // Calculate metrics for each class
    let num_classes = true_positives_map.len();
    let mut metrics_by_class = Vec::with_capacity(num_classes);

    for class in 0..num_classes {
        let tp = *true_positives_map.get(&class).unwrap_or(&0);
        let f_pos = *false_positives_map.get(&class).unwrap_or(&0);
        let f_neg = *false_negatives_map.get(&class).unwrap_or(&0);

        // Update sums for micro-average
        sum_tp += tp;
        sum_fp += f_pos;
        sum_fn += f_neg;

        let precision = calculate_precision(tp, f_pos);
        let recall = calculate_recall(tp, f_neg);
        let f1_score = calculate_f1_score(precision, recall);

        let metric = ClassMetrics {
            precision,
            recall,
            f1_score,
        };
        metrics_by_class.push(metric);

        sum_f1_macro += f1_score;
    }
    let accuracy = calculate_accuracy(targets, predictions);

    // Calculate macro-average F1 score
    let macro_f1 = sum_f1_macro / num_classes as f32;

    // Calculate micro-average F1 score
    let micro_precision = calculate_precision(sum_tp, sum_fp);
    let micro_recall = calculate_recall(sum_tp, sum_fn);
    let micro_f1 = calculate_f1_score(micro_precision, micro_recall);

    ClassificationMetrics {
        accuracy,
        micro_precision,
        micro_recall,
        macro_f1_score: macro_f1,
        micro_f1_score: micro_f1,
        metrics_by_class,
    }
}

/// Calculate accuracy by comparing max values in each row
fn calculate_accuracy(predictions: &DMat, targets: &DMat) -> f32 {
    let rows = predictions.rows();
    if rows == 0 || predictions.rows() != targets.rows() || predictions.cols() != targets.cols() {
        return 0.0;
    }

    let mut correct = 0;
    for i in 0..rows {
        let pred_max_idx = util::find_max_index_in_row(predictions, i);
        let target_max_idx = util::find_max_index_in_row(targets, i);
        if pred_max_idx == target_max_idx {
            correct += 1;
        }
    }

    correct as f32 / rows as f32
}

fn calculate_confusion_matrix(
    targets: &DMat, predictions: &DMat,
) -> (HashMap<usize, usize>, HashMap<usize, usize>, HashMap<usize, usize>) {
    let mut true_positives: HashMap<usize, usize> = HashMap::new();
    let mut false_positives: HashMap<usize, usize> = HashMap::new();
    let mut false_negatives: HashMap<usize, usize> = HashMap::new();

    let rows = targets.rows();
    let cols = targets.cols();

    for i in 0..rows {
        let true_idx = util::find_max_index_in_row(targets, i);
        let pred_idx = util::find_max_index_in_row(predictions, i);

        if true_idx == pred_idx {
            *true_positives.entry(true_idx).or_insert(0) += 1; // Correctly predicted class
        } else {
            *false_positives.entry(pred_idx).or_insert(0) += 1; // Incorrectly predicted as another class
            *false_negatives.entry(true_idx).or_insert(0) += 1; // Failed to predict the actual class
        }
    }

    // For classes not predicted at all or not having any positives, set to 0
    for c in 0..cols {
        true_positives.entry(c).or_insert(0);
        false_positives.entry(c).or_insert(0);
        false_negatives.entry(c).or_insert(0);
    }

    (true_positives, false_positives, false_negatives)
}

pub fn calculate_precision(true_positives: usize, false_positives: usize) -> f32 {
    if true_positives + false_positives == 0 {
        return 0.0;
    }
    true_positives as f32 / (true_positives + false_positives) as f32
}

pub fn calculate_recall(true_positives: usize, false_negatives: usize) -> f32 {
    if true_positives + false_negatives == 0 {
        return 0.0;
    }
    true_positives as f32 / (true_positives + false_negatives) as f32
}

pub fn calculate_f1_score(precision: f32, recall: f32) -> f32 {
    if precision + recall == 0.0 {
        return 0.0;
    }
    2.0 * (precision * recall) / (precision + recall)
}

// Tests for utility functions
#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_calculate_accuracy() {
        let predictions = DMat::new(2, 2, &[0.9f32, 0.1, 0.2, 0.8]);
        let targets = DMat::new(2, 2, &[1.0, 0.0, 0.0, 1.0]);
        let accuracy = calculate_accuracy(&predictions, &targets);
        assert!((accuracy - 1.0).abs() < 1e-6);
    }
}