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

//! Row-wise normalization.
//!
//! [`Normalizer`] scales each row (sample) to unit norm independently.
//! Supports L1, L2, and Max norms. Analogous to `sklearn.preprocessing.Normalizer`.

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

/// Normalize samples individually to unit norm.
///
/// Each row is divided by its norm so that the row has unit length.
/// Note: this is a **row-wise** operation, not column-wise like the scalers.
///
/// # Example
///
/// ```rust
/// use featrs::preprocessing::normalizer::Normalizer;
/// use featrs::preprocessing::normalizer::Norm;
/// use featrs::traits::{Fit, Transform};
///
/// let mut n = Normalizer::l2();
/// # let df = polars::prelude::DataFrame::new(0usize, vec![]).unwrap();
/// // n.fit(df.clone(), target)?;
/// // let normalized = n.transform(df)?;
/// ```
pub struct Normalizer {
    fitted: bool,
    norm: Norm,
}

/// The norm to use for normalization.
#[derive(Clone, Copy)]
pub enum Norm {
    /// L1 norm — sum of absolute values.
    L1,
    /// L2 norm — Euclidean (sqrt of sum of squares).
    L2,
    /// Max norm — maximum absolute value.
    Max,
}

impl Normalizer {
    /// Create a new `Normalizer` with the given norm.
    pub fn new(norm: Norm) -> Self {
        Self {
            fitted: false,
            norm,
        }
    }

    /// Normalize using the L1 norm (sum of absolute values).
    pub fn l1() -> Self {
        Self::new(Norm::L1)
    }

    /// Normalize using the L2 norm (Euclidean).
    pub fn l2() -> Self {
        Self::new(Norm::L2)
    }

    /// Normalize using the Max norm (maximum absolute value).
    pub fn max() -> Self {
        Self::new(Norm::Max)
    }

    fn row_norm(values: &[f64], norm: Norm) -> f64 {
        match norm {
            Norm::L1 => values.iter().map(|v| v.abs()).sum(),
            Norm::L2 => values.iter().map(|v| v * v).sum::<f64>().sqrt(),
            Norm::Max => values.iter().cloned().fold(0.0f64, f64::max),
        }
    }
}

impl Default for Normalizer {
    fn default() -> Self {
        Self::l2()
    }
}

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

    fn fit(&mut self, x: DataFrame, _y: DataFrame) -> Result<()> {
        if x.width() == 0 {
            return Err(Error::InvalidInput(
                "Normalizer.fit received a DataFrame with 0 columns. \
                 Provide at least one column."
                    .into(),
            ));
        }
        for col in x.columns() {
            if col.dtype() != &DataType::Float64 {
                return Err(Error::InvalidInput(format!(
                    "Normalizer.fit: column '{}' has dtype {}; expected Float64. \
                     Normalizer operates on f64 columns only.",
                    col.name(),
                    col.dtype()
                )));
            }
        }
        self.fitted = true;
        Ok(())
    }
}

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

    fn transform(&self, x: DataFrame) -> Result<DataFrame> {
        if !self.fitted {
            return Err(Error::NotFitted(
                "Normalizer has not been fitted. \
                 Call .fit(dataframe, target) before .transform()."
                    .into(),
            ));
        }

        let n_cols = x.width();
        let n_rows = x.height();
        let col_names: Vec<&str> = x.get_column_names().iter().map(|s| s.as_str()).collect();

        let mut col_data: Vec<Vec<f64>> = Vec::with_capacity(n_cols);
        for name in &col_names {
            let ca = x.column(name).unwrap().f64().unwrap();
            col_data.push(ca.iter().flatten().collect());
        }

        for i in 0..n_rows {
            let row_vals: Vec<f64> = col_data.iter().map(|col| col[i]).collect();
            let norm = Self::row_norm(&row_vals, self.norm);
            if norm > f64::EPSILON {
                for col in &mut col_data {
                    col[i] /= norm;
                }
            }
        }

        let mut out_cols: Vec<Column> = Vec::with_capacity(n_cols);
        for (j, name) in col_names.iter().enumerate() {
            let new_ca: ChunkedArray<Float64Type> =
                ChunkedArray::from_slice(name.to_string().as_str().into(), &col_data[j]);
            out_cols.push(new_ca.into_series().into());
        }

        DataFrame::new(n_rows, out_cols).map_err(|e| Error::Computation(e.to_string()))
    }
}

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

    fn make_test_df() -> DataFrame {
        let a = Column::from(Series::new("a".into(), &[3.0f64, 0.0, 1.0]));
        let b = Column::from(Series::new("b".into(), &[4.0f64, 0.0, 2.0]));
        DataFrame::new(3, vec![a, b]).unwrap()
    }

    #[test]
    fn test_l2_normalization() {
        let mut n = Normalizer::l2();
        let df = make_test_df();
        let y = df.clone();

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

        let col_a = result.column("a").unwrap().f64().unwrap();
        let col_b = result.column("b").unwrap().f64().unwrap();
        assert_relative_eq!(col_a.get(0).unwrap(), 0.6, epsilon = 1e-6);
        assert_relative_eq!(col_b.get(0).unwrap(), 0.8, epsilon = 1e-6);
        assert_relative_eq!(col_a.get(1).unwrap(), 0.0, epsilon = 1e-6);
    }

    #[test]
    fn test_l1_normalization() {
        let mut n = Normalizer::l1();
        let df = make_test_df();
        let y = df.clone();

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

        let col_a = result.column("a").unwrap().f64().unwrap();
        let col_b = result.column("b").unwrap().f64().unwrap();
        assert_relative_eq!(col_a.get(0).unwrap(), 3.0 / 7.0, epsilon = 1e-6);
        assert_relative_eq!(col_b.get(0).unwrap(), 4.0 / 7.0, epsilon = 1e-6);
    }

    #[test]
    fn test_max_normalization() {
        let mut n = Normalizer::max();
        let df = make_test_df();
        let y = df.clone();

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

        let col_a = result.column("a").unwrap().f64().unwrap();
        let col_b = result.column("b").unwrap().f64().unwrap();
        assert_relative_eq!(col_a.get(0).unwrap(), 0.75, epsilon = 1e-6);
        assert_relative_eq!(col_b.get(0).unwrap(), 1.0, epsilon = 1e-6);
    }
}