parsnip 0.2.0

Data science metrics (presently categorical only) for 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
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
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341

use std::collections::HashMap;

/// Compute the gini impurity of a dataset. 
/// 
/// Returns a float, 0 representing a perfectly pure dataset. Normal distribution: ~0.33
/// 
/// By default, any empty dataset will return a gini of 1.0. This may be unexpected behaviour.
/// ```
/// use parsnip::gini;
/// assert_eq!(gini(&vec![0, 0, 0, 1]), 0.375);
/// ```
pub fn gini(data: &[u64]) -> f32 {
    if data.len() == 0 {
        return 1.0;
    } 
    fn p_squared(count: usize, len: f32) -> f32 {
        let p = count as f32 / len;
        return p * p;
    }
    let len = data.len() as f32;
    let mut count = HashMap::new();
    for &value in data {
        *count.entry(value).or_insert(0) += 1;
    }
    let counts: Vec<usize> = count.into_iter().map(|(_, c)| c).collect();
    let indiv : Vec<f32> = counts.iter().map(|x| p_squared(*x, len)).collect();
    let sum : f32 = indiv.iter().sum();
    return 1.0 - sum;
}

/// The categorical accuracy of a dataset
/// Returns a float where 1.0 is a perfectly accurate dataset
/// ```
/// use parsnip::categorical_accuracy;
/// let pred = vec![0, 0, 0 , 1, 2];
/// let actual = vec![1, 1, 1, 1, 2];
/// assert_eq!(categorical_accuracy(&pred, &actual), 0.4);
/// ```
pub fn categorical_accuracy(pred: &[u64], actual: &[u64]) -> f32 {
    assert_eq!(pred.len(), actual.len());
    let bools =  pred.iter().zip(actual).map(|(x,y)| x == y);
    let truthy : Vec<bool> =  bools.filter(|b| *b).collect();
    return truthy.len() as f32 / pred.len() as f32;
}

fn class_precision(pred: &[u64], actual: &[u64], class: u64) -> f32 {
    assert_eq!(pred.len(), actual.len());
    let true_positives_map = pred.iter().zip(actual).map(|(p, a)| p == a && *p == class);
    let true_positives = true_positives_map.filter(|b| *b).count() as f32;
    let all_positives = pred.iter().map(|p| *p == class).filter(|b| *b).count() as f32;
    if all_positives == 0.0 {
        return 0.0;
    }
    return true_positives / all_positives;
}

fn weighted_precision(pred: &[u64], actual: &[u64]) -> f32 {
    assert_eq!(pred.len(), actual.len());
    let mut classes : Vec<u64> = pred.into_iter().map(|x| *x).collect();
    let mut class_weights = HashMap::new();
    classes.sort();
    classes.dedup();
    for value in classes.clone() {
        class_weights.insert(value, actual.iter().filter(|a| **a == value).count() as f32 / actual.len() as f32);
    }
    return classes.iter().map(|c| class_precision(pred, actual, *c) * class_weights.get(c).unwrap()).sum();
}

fn macro_precision(pred: &[u64], actual: &[u64]) -> f32 {
    assert_eq!(pred.len(), actual.len());
    let mut classes : Vec<u64> = pred.into_iter().map(|x| *x).collect();
    let mut class_weights = HashMap::new();
    classes.sort();
    classes.dedup();
    for value in classes.clone() {
        class_weights.insert(value, 1.0 / actual.len() as f32);
    }
    return classes.iter().map(|c| class_precision(pred, actual, *c) / classes.len() as f32).sum();
}

/// The precision of a dataset
/// Returns a float where a 1.0 is a perfectly precise result set
/// 
/// Supports macro and weighted averages
/// ```
/// use parsnip::precision;
/// 
/// let actual = vec![0, 1, 2, 0, 1, 2];
/// let pred = vec![0, 2, 1, 0, 0, 1];
/// 
/// assert_eq!(precision(&pred, &actual, Some("macro".to_string())), 0.22222222);
/// ```
pub fn precision(pred: &[u64], actual: &[u64], average: Option<String>) -> f32 {
    match average {
        None => return macro_precision(pred, actual),
        Some(string) => match string.as_ref() {
            "macro" => return macro_precision(pred, actual),
            "weighted" => return weighted_precision(pred, actual),
            _ => panic!("invalid averaging type")
        }
    }
}

fn class_recall(pred: &[u64], actual: &[u64], class: u64) -> f32 {
    assert_eq!(pred.len(), actual.len());
    let true_positives_map = pred.iter().zip(actual).map(|(p, a)| p == a && *a == class);
    let true_positives = true_positives_map.filter(|b| *b).count() as f32;
    let tp_fn = actual.iter().map(|a| *a == class).filter(|b| *b).count() as f32;
    if tp_fn == 0.0 {
        return 0.0;
    }
    return true_positives / tp_fn;
}

fn weighted_recall(pred: &[u64], actual: &[u64]) -> f32 {
    assert_eq!(pred.len(), actual.len());
    let mut classes : Vec<u64> = pred.into_iter().map(|x| *x).collect();
    let mut class_weights = HashMap::new();
    classes.sort();
    classes.dedup();
    for value in classes.clone() {
        class_weights.insert(value, actual.iter().filter(|a| **a == value).count() as f32 / actual.len() as f32);
    }
    return classes.iter().map(|c| class_recall(pred, actual, *c) * class_weights.get(c).unwrap()).sum();
}

fn macro_recall(pred: &[u64], actual: &[u64]) -> f32 {
    assert_eq!(pred.len(), actual.len());
    let mut classes : Vec<u64> = pred.into_iter().map(|x| *x).collect();
    let mut class_weights = HashMap::new();
    classes.sort();
    classes.dedup();
    for value in classes.clone() {
        class_weights.insert(value, 1.0 / actual.len() as f32);
    }
    return classes.iter().map(|c| class_recall(pred, actual, *c) / classes.len() as f32).sum();
}

/// The recall of a dataset
/// Returns a float where a 1.0 is a perfectly recalled result set
/// 
/// Supports macro and weighted averages
/// ```
/// use parsnip::recall;
/// 
/// let actual = vec![0, 1, 2, 0, 1, 2];
/// let pred = vec![0, 2, 1, 0, 0, 1];
/// 
/// assert_eq!(recall(&pred, &actual, Some("macro".to_string())), 0.333333334);
/// ```
pub fn recall(pred: &[u64], actual: &[u64], average: Option<String>) -> f32 {
    match average {
        None => return macro_recall(pred, actual),
        Some(string) => match string.as_ref() {
            "macro" => return macro_recall(pred, actual),
            "weighted" => return weighted_recall(pred, actual),
            _ => panic!("invalid averaging type")
        }
    }
}

fn macro_f1(pred: &[u64], actual: &[u64]) -> f32 {
    let recall = macro_recall(pred, actual);
    let precision = macro_precision(pred, actual);
    return 2.0 * (recall * precision) / (recall + precision);
}

fn weighted_f1(pred: &[u64], actual: &[u64]) -> f32 {
    let recall = weighted_recall(pred, actual);
    let precision = weighted_precision(pred, actual);
    return 2.0 * (recall * precision) / (recall + precision);

}

/// The recall of a dataset
/// Returns an f1 score where 1 is perfect and 0 is atrocious.
/// 
/// Supports macro and weighted averages
/// ```
/// use parsnip::f1_score;
/// 
/// let actual = vec![0, 1, 2, 0, 1, 2];
/// let pred = vec![0, 2, 1, 0, 0, 1];
/// 
/// assert_eq!(f1_score(&pred, &actual, Some("macro".to_string())), 0.26666665);
/// assert_eq!(f1_score(&pred, &actual, Some("weighted".to_string())), 0.26666668);
/// ```
pub fn f1_score(pred: &[u64], actual: &[u64], average: Option<String>) -> f32 {
    match average {
        None => return macro_f1(pred, actual),
        Some(string) => match string.as_ref() {
            "macro" => return macro_f1(pred, actual),
            "weighted" => return weighted_f1(pred, actual),
            _ => panic!("invalid averaging type")
        }
    }
}

/// The recall of a dataset
/// Returns the hamming loss which is the percentage of items which are misclassified [0, 1]
/// 
/// Supports macro and weighted averages
/// ```
/// use parsnip::hamming_loss;
/// 
/// let actual = vec![0, 1, 2, 0, 0];
/// let pred = vec![0, 2, 1, 0, 1];
/// 
/// assert_eq!(hamming_loss(&pred, &actual), 0.6);
/// ```
pub fn hamming_loss(pred: &[u64], actual: &[u64]) -> f32 {
    return 1.0 - categorical_accuracy(pred, actual);
}

fn macro_fbeta_score(pred: &[u64], actual: &[u64], beta: f32) -> f32 {
    let precision = macro_precision(pred, actual);
    let recall = macro_recall(pred, actual);
    let top = (1.0 + beta * beta)  * (recall * precision);
    let bottom = (beta * beta * precision) + recall;
    return top / bottom;
}

fn weighted_fbeta_score(pred: &[u64], actual: &[u64], beta: f32) -> f32 {
    let precision = weighted_precision(pred, actual);
    let recall = weighted_recall(pred, actual);
    let top = (1.0 + beta * beta)  * (recall * precision);
    let bottom = (beta * beta * precision) + recall;
    return top / bottom;
}

/// The recall of a dataset
/// Returns the hamming loss which is the percentage of items which are misclassified [0, 1]
/// 
/// Supports macro and weighted averages
/// ```
/// use parsnip::fbeta_score;
/// 
/// let actual = vec![0, 1, 2, 0, 1, 2];
/// let pred = vec![0, 2, 1, 0, 0, 1];
/// 
/// assert_eq!(fbeta_score(&pred, &actual, 0.5, Some("macro".to_string())), 0.23809524);
/// assert_eq!(fbeta_score(&pred, &actual, 0.5, Some("weighted".to_string())), 0.23809527);
/// ```
pub fn fbeta_score(pred: &[u64], actual: &[u64], beta: f32, average: Option<String>) -> f32 {
    match average {
        None => return macro_fbeta_score(pred, actual, beta),
        Some(string) => match string.as_ref() {
            "macro" => return macro_fbeta_score(pred, actual, beta),
            "weighted" => return weighted_fbeta_score(pred, actual, beta),
            _ => panic!("invalid averaging type")
        }
    }
}

/// The recall of a dataset
/// Returns the hamming loss which is the percentage of items which are misclassified [0, 1]
/// 
/// Supports macro and weighted averages
/// ```
/// use parsnip::jaccard_similiarity_score;
/// 
/// let actual = vec![0, 2, 1, 3];
/// let pred = vec![0, 1, 2, 3];
/// 
/// assert_eq!(jaccard_similiarity_score(&pred, &actual), 0.5);
/// ```
pub fn jaccard_similiarity_score(pred: &[u64], actual: &[u64]) -> f32 {
    return categorical_accuracy(pred, actual);
}


#[cfg(test)]
mod tests {
    use super::*;
    #[test]
    fn test_gini() {
        let vec = vec![0, 0, 0, 1];
        assert_eq!(0.375, gini(&vec));
        let v2 = vec![0, 0];
        assert_eq!(0.0, gini(&v2));
        let mut v3 = vec![0];
        v3.pop();
        assert_eq!(1.0, gini(&v3));
    }

    #[test]
    fn test_categorical_accuracy() {
        let pred = vec![0, 1, 0, 1, 0, 1];
        let real = vec![0, 0, 0, 0, 1, 0];
        assert_eq!(0.33333334, categorical_accuracy(&pred, &real));
    }

    #[test]
    fn test_class_precision() {
        let actual = vec![0, 1, 2, 0, 1, 2];
        let pred = vec![0, 2, 1, 0, 0, 1];
        assert_eq!(0.6666667, class_precision(&pred, &actual, 0));
    }

    #[test]
    fn test_class_recall() {
        let actual = vec![0, 1, 2, 0, 0, 0];
        let pred = vec![0, 2, 1, 0, 0, 1];
        assert_eq!(0.75, class_recall(&pred, &actual, 0));
    }

    #[test]
    fn test_weighted_precision() {
        let actual = vec![0, 1, 2, 0, 1, 2];
        let pred = vec![0, 2, 1, 0, 0, 1];
        assert_eq!(0.22222224, weighted_precision(&pred, &actual));
    }

    #[test]
    fn test_macro_precision() {
        let actual = vec![0, 1, 2, 0, 1, 2];
        let pred = vec![0, 2, 1, 0, 0, 1];
        assert_eq!(0.22222222, macro_precision(&pred, &actual));
    }

        #[test]
    fn test_macro_recall() {
        let actual = vec![0, 1, 2, 0, 1, 2];
        let pred = vec![0, 2, 1, 0, 0, 1];
        assert_eq!(0.33333334, macro_recall(&pred, &actual));
    }

        #[test]
    fn test_weighted_recall() {
        let actual = vec![0, 1, 2, 0, 1, 2];
        let pred = vec![0, 2, 1, 0, 0, 1];
        assert_eq!(0.333333334, weighted_recall(&pred, &actual));
    }

    #[test]
    fn test_f1_score() {
        let actual = vec![0, 1, 2, 0, 1, 2];
        let pred = vec![0, 2, 1, 0, 0, 1];
        assert_eq!(f1_score(&pred, &actual, Some("macro".to_string())), 0.26666665);
    }
}