featrs 0.2.0

Feature engineering library for Rust, inspired by scikit-learn
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

//! Missing value imputation.
//!
//! [`SimpleImputer`] replaces missing (`None` / `null`) values with
//! a statistic computed from the non-missing values in each column.

use crate::traits::{Error, Fit, Result, Transform};
use polars::prelude::*;
use std::collections::HashMap;

/// Strategy for computing imputation values.
#[derive(Clone, Copy)]
pub enum Strategy {
    /// Replace missing values with the column mean.
    Mean,
    /// Replace missing values with the column median.
    Median,
    /// Replace missing values with the most frequent value in the column.
    MostFrequent,
    /// Replace missing values with a constant value.
    Constant(f64),
}

/// Impute missing values using basic statistics.
///
/// Only operates on [`Float64`](DataType::Float64) columns.
///
/// # Example
///
/// ```rust
/// use featrs::preprocessing::imputer::SimpleImputer;
/// use featrs::preprocessing::imputer::Strategy;
/// use featrs::traits::{Fit, Transform};
///
/// let mut imp = SimpleImputer::mean();
/// # let df = polars::prelude::DataFrame::new(0usize, vec![]).unwrap();
/// // imp.fit(df.clone(), target)?;
/// // let filled = imp.transform(df)?;
/// ```
pub struct SimpleImputer {
    fitted: bool,
    strategy: Strategy,
    fill_values: Option<HashMap<String, f64>>,
}

impl SimpleImputer {
    /// Create a new `SimpleImputer` with the given strategy.
    pub fn new(strategy: Strategy) -> Self {
        Self {
            fitted: false,
            strategy,
            fill_values: None,
        }
    }

    /// Impute using the column mean.
    pub fn mean() -> Self {
        Self::new(Strategy::Mean)
    }

    /// Impute using the column median.
    pub fn median() -> Self {
        Self::new(Strategy::Median)
    }

    /// Impute using the most frequent (mode) value.
    pub fn most_frequent() -> Self {
        Self::new(Strategy::MostFrequent)
    }

    /// Impute with a constant value.
    pub fn constant(value: f64) -> Self {
        Self::new(Strategy::Constant(value))
    }
}

impl Fit<DataFrame, DataFrame> for SimpleImputer {
    type Output = ();

    fn fit(&mut self, x: DataFrame, _y: DataFrame) -> Result<()> {
        if x.height() == 0 {
            return Err(Error::InvalidInput(
                "SimpleImputer.fit received a DataFrame with 0 rows. \
                 Provide data with at least 1 row."
                    .into(),
            ));
        }
        let mut fill_values = HashMap::new();

        for col in x.columns() {
            let name = col.name().to_string();
            if col.dtype() != &DataType::Float64 {
                continue;
            }
            let ca = col.f64().unwrap();
            let all_vals: Vec<f64> = ca.iter().flatten().collect();
            let has_missing = ca.iter().any(|v| v.is_none());

            if !has_missing {
                continue;
            }

            let fill = match self.strategy {
                Strategy::Mean => {
                    if all_vals.is_empty() {
                        return Err(Error::Computation(format!(
                            "SimpleImputer(Mean): column '{}' has no non-null values. \
                             Cannot compute the mean of an all-null column. \
                             Use Strategy::Constant instead.",
                            name
                        )));
                    }
                    all_vals.iter().sum::<f64>() / all_vals.len() as f64
                }
                Strategy::Median => {
                    let mut sorted = all_vals.clone();
                    sorted.sort_by(|a, b| a.partial_cmp(b).unwrap());
                    if sorted.is_empty() {
                        return Err(Error::Computation(format!(
                            "SimpleImputer(Median): column '{}' has no non-null values. \
                             Cannot compute the median of an all-null column. \
                             Use Strategy::Constant instead.",
                            name
                        )));
                    }
                    let mid = sorted.len() / 2;
                    if sorted.len().is_multiple_of(2) {
                        (sorted[mid - 1] + sorted[mid]) / 2.0
                    } else {
                        sorted[mid]
                    }
                }
                Strategy::MostFrequent => {
                    if all_vals.is_empty() {
                        return Err(Error::Computation(format!(
                            "SimpleImputer(MostFrequent): column '{}' has no non-null values. \
                             Cannot compute the mode of an all-null column. \
                             Use Strategy::Constant instead.",
                            name
                        )));
                    }
                    let mut freq: HashMap<u64, usize> = HashMap::new();
                    for &v in &all_vals {
                        *freq.entry(v.to_bits()).or_default() += 1;
                    }
                    let (max_key, _) = freq.into_iter().max_by_key(|&(_, c)| c).unwrap();
                    f64::from_bits(max_key)
                }
                Strategy::Constant(v) => v,
            };

            fill_values.insert(name, fill);
        }

        self.fill_values = Some(fill_values);
        self.fitted = true;
        Ok(())
    }
}

impl Transform<DataFrame> for SimpleImputer {
    type Output = DataFrame;

    fn transform(&self, x: DataFrame) -> Result<DataFrame> {
        if !self.fitted {
            return Err(Error::NotFitted(
                "SimpleImputer has not been fitted. \
                 Call .fit(dataframe, target) before .transform()."
                    .into(),
            ));
        }
        let fill_values = self.fill_values.as_ref().unwrap();
        let mut out = x.clone();

        for (name, fill) in fill_values {
            let s = out.column(name.as_str()).map_err(|e| {
                Error::InvalidInput(format!(
                    "SimpleImputer.transform: column '{}' not found. \
                     The imputer was fitted on columns: {:?}. {}",
                    name,
                    fill_values.keys().collect::<Vec<_>>(),
                    e
                ))
            })?;
            let ca = s.f64().map_err(|e| {
                Error::InvalidInput(format!(
                    "SimpleImputer.transform: column '{}' has dtype {}; expected Float64. {}",
                    name,
                    s.dtype(),
                    e
                ))
            })?;
            let filled: ChunkedArray<Float64Type> =
                ca.iter().map(|opt| opt.or(Some(*fill))).collect();
            out.replace(name.as_str(), filled.into_series().into())
                .unwrap();
        }

        Ok(out)
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use approx::assert_relative_eq;

    fn make_test_df() -> DataFrame {
        let a = Column::from(Series::new("x".into(), &[Some(1.0f64), None, Some(3.0)]));
        let b = Column::from(Series::new("y".into(), &[Some(10.0f64), Some(20.0), None]));
        DataFrame::new(3, vec![a, b]).unwrap()
    }

    #[test]
    fn test_imputer_mean() {
        let mut imp = SimpleImputer::mean();
        let df = make_test_df();
        let y = df.clone();

        imp.fit(df.clone(), y).unwrap();
        let result = imp.transform(df).unwrap();

        let x_vals: Vec<f64> = result
            .column("x")
            .unwrap()
            .f64()
            .unwrap()
            .iter()
            .flatten()
            .collect();
        assert_relative_eq!(x_vals[1], 2.0, epsilon = 1e-6);

        let y_vals: Vec<f64> = result
            .column("y")
            .unwrap()
            .f64()
            .unwrap()
            .iter()
            .flatten()
            .collect();
        assert_relative_eq!(y_vals[2], 15.0, epsilon = 1e-6);
    }

    #[test]
    fn test_imputer_constant() {
        let mut imp = SimpleImputer::constant(0.0);
        let df = make_test_df();
        let y = df.clone();

        imp.fit(df.clone(), y).unwrap();
        let result = imp.transform(df).unwrap();

        let x_vals: Vec<f64> = result
            .column("x")
            .unwrap()
            .f64()
            .unwrap()
            .iter()
            .flatten()
            .collect();
        assert_relative_eq!(x_vals[1], 0.0, epsilon = 1e-6);
    }
}