stats/
unsorted.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
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
use std::default::Default;
use std::iter::{FromIterator, IntoIterator};
use num_traits::ToPrimitive;

use {Commute, Partial};

/// Compute the exact median on a stream of data.
///
/// (This has time complexity `O(nlogn)` and space complexity `O(n)`.)
pub fn median<I>(it: I) -> Option<f64>
        where I: Iterator, <I as Iterator>::Item: PartialOrd + ToPrimitive {
    it.collect::<Unsorted<_>>().median()
}

/// Compute the exact mode on a stream of data.
///
/// (This has time complexity `O(nlogn)` and space complexity `O(n)`.)
///
/// If the data does not have a mode, then `None` is returned.
pub fn mode<T, I>(it: I) -> Option<T>
       where T: PartialOrd + Clone, I: Iterator<Item=T> {
    it.collect::<Unsorted<T>>().mode()
}

/// Compute the modes on a stream of data.
/// 
/// If there is a single mode, then only that value is returned in the `Vec`
/// however, if there multiple values tied for occuring the most amount of times
/// those values are returned.
/// 
/// ## Example
/// ```
/// use stats;
/// 
/// let vals = vec![1, 1, 2, 2, 3];
/// 
/// assert_eq!(stats::modes(vals.into_iter()), vec![1, 2]);
/// ```
/// This has time complexity `O(n)`
///
/// If the data does not have a mode, then an empty `Vec` is returned.
pub fn modes<T, I>(it: I) -> Vec<T>
       where T: PartialOrd + Clone, I: Iterator<Item=T> {
    it.collect::<Unsorted<T>>().modes()
}

fn median_on_sorted<T>(data: &[T]) -> Option<f64>
        where T: PartialOrd + ToPrimitive {
    Some(match data.len() {
        0 => return None,
        1 => data[0].to_f64().unwrap(),
        len if len % 2 == 0 => {
            let v1 = data[(len / 2) - 1].to_f64().unwrap();
            let v2 = data[len / 2].to_f64().unwrap();
            (v1 + v2) / 2.0
        }
        len => {
            data[len / 2].to_f64().unwrap()
        }
    })
}

fn mode_on_sorted<T, I>(it: I) -> Option<T>
        where T: PartialOrd, I: Iterator<Item=T> {
    // This approach to computing the mode works very nicely when the
    // number of samples is large and is close to its cardinality.
    // In other cases, a hashmap would be much better.
    // But really, how can we know this when given an arbitrary stream?
    // Might just switch to a hashmap to track frequencies. That would also
    // be generally useful for discovering the cardinality of a sample.
    let (mut mode, mut next) = (None, None);
    let (mut mode_count, mut next_count) = (0usize, 0usize);
    for x in it {
        if mode.as_ref().map(|y| y == &x).unwrap_or(false) {
            mode_count += 1;
        } else if next.as_ref().map(|y| y == &x).unwrap_or(false) {
            next_count += 1;
        } else {
            next = Some(x);
            next_count = 0;
        }

        if next_count > mode_count {
            mode = next;
            mode_count = next_count;
            next = None;
            next_count = 0;
        } else if next_count == mode_count {
            mode = None;
            mode_count = 0usize;
        }
    }
    mode
}

fn modes_on_sorted<T, I>(it: I) -> Vec<T>
        where T: PartialOrd, I: Iterator<Item=T> {

    let mut highest_mode = 1_u32;
    let mut modes: Vec<u32> = vec![];
    let mut values = vec![];
    let mut count = 0;
    for x in it {
        if values.len() == 0 {
            values.push(x);
            modes.push(1);
            continue
        }
        if x == values[count] {
            modes[count] += 1;
            if highest_mode < modes[count] {
                highest_mode = modes[count];
            }
        } else {
            values.push(x);
            modes.push(1);
            count += 1;
        }
    }
    modes.into_iter()
        .zip(values)
        .filter(|(cnt, _val)| *cnt == highest_mode && highest_mode > 1)
        .map(|(_, val)| val)
        .collect()
}

/// A commutative data structure for lazily sorted sequences of data.
///
/// The sort does not occur until statistics need to be computed.
///
/// Note that this works on types that do not define a total ordering like
/// `f32` and `f64`. When an ordering is not defined, an arbitrary order
/// is returned.
#[derive(Clone)]
pub struct Unsorted<T> {
    data: Vec<Partial<T>>,
    sorted: bool,
}

impl<T: PartialOrd> Unsorted<T> {
    /// Create initial empty state.
    pub fn new() -> Unsorted<T> {
        Default::default()
    }

    /// Add a new element to the set.
    pub fn add(&mut self, v: T) {
        self.dirtied();
        self.data.push(Partial(v))
    }

    /// Return the number of data points.
    pub fn len(&self) -> usize {
        self.data.len()
    }

    fn sort(&mut self) {
        if !self.sorted {
            self.data.sort();
        }
    }

    fn dirtied(&mut self) {
        self.sorted = false;
    }
}

impl<T: PartialOrd + Eq + Clone> Unsorted<T> {
    pub fn cardinality(&mut self) -> usize {
        self.sort();
        let mut set = self.data.clone();
        set.dedup();
        set.len()
    }
}

impl<T: PartialOrd + Clone> Unsorted<T> {
    /// Returns the mode of the data.
    pub fn mode(&mut self) -> Option<T> {
        self.sort();
        mode_on_sorted(self.data.iter()).map(|p| p.0.clone())
    }

    /// Returns the modes of the data.
    pub fn modes(&mut self) -> Vec<T> {
        self.sort();
        modes_on_sorted(self.data.iter())
            .into_iter()
            .map(|p| p.0.clone())
            .collect()
    }
}

impl<T: PartialOrd + ToPrimitive> Unsorted<T> {
    /// Returns the median of the data.
    pub fn median(&mut self) -> Option<f64> {
        self.sort();
        median_on_sorted(&*self.data)
    }
}

impl<T: PartialOrd> Commute for Unsorted<T> {
    fn merge(&mut self, v: Unsorted<T>) {
        self.dirtied();
        self.data.extend(v.data.into_iter());
    }
}

impl<T: PartialOrd> Default for Unsorted<T> {
    fn default() -> Unsorted<T> {
        Unsorted {
            data: Vec::with_capacity(1000),
            sorted: true,
        }
    }
}

impl<T: PartialOrd> FromIterator<T> for Unsorted<T> {
    fn from_iter<I: IntoIterator<Item=T>>(it: I) -> Unsorted<T> {
        let mut v = Unsorted::new();
        v.extend(it);
        v
    }
}

impl<T: PartialOrd> Extend<T> for Unsorted<T> {
    fn extend<I: IntoIterator<Item=T>>(&mut self, it: I) {
        self.dirtied();
        self.data.extend(it.into_iter().map(Partial))
    }
}

#[cfg(test)]
mod test {
    use super::{median, mode, modes};

    #[test]
    fn median_stream() {
        assert_eq!(median(vec![3usize, 5, 7, 9].into_iter()), Some(6.0));
        assert_eq!(median(vec![3usize, 5, 7].into_iter()), Some(5.0));
    }

    #[test]
    fn mode_stream() {
        assert_eq!(mode(vec![3usize, 5, 7, 9].into_iter()), None);
        assert_eq!(mode(vec![3usize, 3, 3, 3].into_iter()), Some(3));
        assert_eq!(mode(vec![3usize, 3, 3, 4].into_iter()), Some(3));
        assert_eq!(mode(vec![4usize, 3, 3, 3].into_iter()), Some(3));
        assert_eq!(mode(vec![1usize, 1, 2, 3, 3].into_iter()), None);
    }

    #[test]
    fn median_floats() {
        assert_eq!(median(vec![3.0f64, 5.0, 7.0, 9.0].into_iter()), Some(6.0));
        assert_eq!(median(vec![3.0f64, 5.0, 7.0].into_iter()), Some(5.0));
        assert_eq!(median(vec![1.0f64, 2.5, 3.0].into_iter()), Some(2.5));
    }

    #[test]
    fn mode_floats() {
        assert_eq!(mode(vec![3.0f64, 5.0, 7.0, 9.0].into_iter()), None);
        assert_eq!(mode(vec![3.0f64, 3.0, 3.0, 3.0].into_iter()), Some(3.0));
        assert_eq!(mode(vec![3.0f64, 3.0, 3.0, 4.0].into_iter()), Some(3.0));
        assert_eq!(mode(vec![4.0f64, 3.0, 3.0, 3.0].into_iter()), Some(3.0));
        assert_eq!(mode(vec![1.0f64, 1.0, 2.0, 3.0, 3.0].into_iter()), None);
    }

    #[test]
    fn modes_stream() {
        assert_eq!(modes(vec![3usize, 5, 7, 9].into_iter()), vec![]);
        assert_eq!(modes(vec![3usize, 3, 3, 3].into_iter()), vec![3]);
        assert_eq!(modes(vec![3usize, 3, 4, 4].into_iter()), vec![3, 4]);
        assert_eq!(modes(vec![4usize, 3, 3, 3].into_iter()), vec![3]);
        assert_eq!(modes(vec![1usize, 1, 2, 2].into_iter()), vec![1, 2]);
        let vec: Vec<u32> = vec![];
        assert_eq!(modes(vec.into_iter()), vec![]);
    }

    #[test]
    fn modes_floats() {
        assert_eq!(modes(vec![3_f64, 5.0, 7.0, 9.0].into_iter()), vec![]);
        assert_eq!(modes(vec![3_f64, 3.0, 3.0, 3.0].into_iter()), vec![3.0]);
        assert_eq!(modes(vec![3_f64, 3.0, 4.0, 4.0].into_iter()), vec![3.0, 4.0]);
        assert_eq!(modes(vec![1_f64, 1.0, 2.0, 3.0, 3.0].into_iter()), vec![1.0, 3.0]);
    }
}