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

//! Differencing and percentage change.
//!
//! [`Difference`] computes `x[t] - x[t - period]` and percentage change.

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

/// Compute differences and percentage changes.
///
/// # Example
///
/// ```rust
/// use featrs::time_series::diff::Difference;
/// use featrs::traits::{Fit, Transform};
///
/// let mut d = Difference::new(&["value"], 1, false);
/// # let df = polars::prelude::DataFrame::new(0usize, vec![]).unwrap();
/// // d.fit(df.clone(), target)?;
/// // let diffed = d.transform(df)?;
/// ```
pub struct Difference {
    fitted: bool,
    columns: Vec<String>,
    period: i64,
    pct_change: bool,
}

impl Difference {
    pub fn new(columns: &[&str], period: i64, pct_change: bool) -> Self {
        Self {
            fitted: false,
            columns: columns.iter().map(|s| s.to_string()).collect(),
            period,
            pct_change,
        }
    }

    /// Create a diff transformer (`x[t] - x[t-1]`).
    pub fn diff(columns: &[&str], period: i64) -> Self {
        Self::new(columns, period, false)
    }

    /// Create a pct_change transformer (`(x[t] - x[t-1]) / x[t-1]`).
    pub fn pct_change(columns: &[&str], period: i64) -> Self {
        Self::new(columns, period, true)
    }
}

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

    fn fit(&mut self, x: DataFrame, _y: DataFrame) -> Result<()> {
        if self.columns.is_empty() {
            return Err(Error::InvalidInput(
                "Difference: at least one column is required.".into(),
            ));
        }
        for col in &self.columns {
            if x.column(col.as_str()).is_err() {
                return Err(Error::InvalidInput(format!(
                    "Difference: column '{}' not found.",
                    col
                )));
            }
        }
        self.fitted = true;
        Ok(())
    }
}

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

    fn transform(&self, x: DataFrame) -> Result<DataFrame> {
        if !self.fitted {
            return Err(Error::NotFitted("Difference".into()));
        }
        let mut out = x.clone();

        for col in &self.columns {
            let s = out.column(col.as_str()).unwrap().clone();

            let suffix = if self.pct_change { "pct" } else { "diff" };
            let col_name = format!("{}_{}_{}", col, suffix, self.period);

            let orig = s.f64().map_err(|_| {
                Error::InvalidInput(format!("Difference: column '{}' is not f64", col))
            })?;
            let shifted = s.shift(self.period);
            let shifted = shifted.f64().map_err(|_| {
                Error::InvalidInput(format!("Difference: column '{}' is not f64", col))
            })?;

            let result: ChunkedArray<Float64Type> = orig
                .iter()
                .zip(shifted.iter())
                .map(|(a, b)| match (a, b) {
                    (Some(va), Some(vb)) => {
                        if self.pct_change {
                            if vb.abs() > f64::EPSILON {
                                Some((va - vb) / vb)
                            } else {
                                None
                            }
                        } else {
                            Some(va - vb)
                        }
                    }
                    _ => None,
                })
                .collect();

            out.with_column(
                result
                    .into_series()
                    .with_name(col_name.as_str().into())
                    .into(),
            )
            .map_err(|e| Error::Computation(e.to_string()))?;
        }

        Ok(out)
    }
}

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

    #[test]
    fn test_diff() {
        let vals = Column::from(Series::new("x".into(), &[10.0f64, 20.0, 30.0, 40.0]));
        let df = DataFrame::new(4, vec![vals]).unwrap();
        let mut d = Difference::diff(&["x"], 1);
        let y = df.clone();

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

        let diffed = result.column("x_diff_1").unwrap().f64().unwrap();
        assert!(diffed.get(0).is_none());
        assert_relative_eq!(diffed.get(1).unwrap(), 10.0, epsilon = 1e-6);
        assert_relative_eq!(diffed.get(2).unwrap(), 10.0, epsilon = 1e-6);
    }

    #[test]
    fn test_pct_change() {
        let vals = Column::from(Series::new("x".into(), &[100.0f64, 110.0, 121.0]));
        let df = DataFrame::new(3, vec![vals]).unwrap();
        let mut d = Difference::pct_change(&["x"], 1);
        let y = df.clone();

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

        let pct = result.column("x_pct_1").unwrap().f64().unwrap();
        assert!(pct.get(0).is_none());
        assert_relative_eq!(pct.get(1).unwrap(), 0.1, epsilon = 1e-6);
    }
}