scirs2_series/feature_selection/
mod.rs

1//! Time series feature selection methods
2//!
3//! This module provides comprehensive feature selection methods specifically designed for time series data.
4//! It includes filter methods, wrapper methods, embedded methods, and time series specific approaches.
5
6use scirs2_core::ndarray::Array1;
7use std::collections::HashMap;
8
9// Re-export submodules
10pub mod embedded;
11pub mod filter;
12pub mod selector;
13pub mod time_series;
14pub mod wrapper;
15
16#[cfg(test)]
17mod tests;
18
19// Re-export main types and structs
20pub use embedded::EmbeddedMethods;
21pub use filter::FilterMethods;
22pub use selector::FeatureSelector;
23pub use time_series::TimeSeriesMethods;
24pub use wrapper::WrapperMethods;
25
26/// Feature selection result containing selected features and their scores
27#[derive(Debug, Clone)]
28pub struct FeatureSelectionResult {
29    /// Indices of selected features
30    pub selected_features: Vec<usize>,
31    /// Feature scores (higher is better)
32    pub feature_scores: Array1<f64>,
33    /// Selection method used
34    pub method: String,
35    /// Additional metadata
36    pub metadata: HashMap<String, f64>,
37}
38
39/// Configuration for feature selection methods
40#[derive(Debug, Clone)]
41pub struct FeatureSelectionConfig {
42    /// Number of features to select (None = automatic)
43    pub n_features: Option<usize>,
44    /// Scoring method for wrapper methods
45    pub scoring_method: ScoringMethod,
46    /// Cross-validation folds for wrapper methods
47    pub cv_folds: usize,
48    /// Significance level for statistical tests
49    pub alpha: f64,
50    /// Minimum correlation threshold for filter methods
51    pub correlation_threshold: f64,
52    /// Minimum variance threshold for variance-based filtering
53    pub variance_threshold: f64,
54    /// Maximum number of iterations for wrapper methods
55    pub max_iterations: usize,
56    /// Random seed for reproducibility
57    pub random_seed: Option<u64>,
58    /// Regularization parameter for embedded methods
59    pub regularization_alpha: f64,
60    /// Maximum lag for time series specific methods
61    pub max_lag: usize,
62    /// Seasonal period for seasonal feature selection
63    pub seasonal_period: Option<usize>,
64}
65
66impl Default for FeatureSelectionConfig {
67    fn default() -> Self {
68        Self {
69            n_features: None,
70            scoring_method: ScoringMethod::MeanSquaredError,
71            cv_folds: 5,
72            alpha: 0.05,
73            correlation_threshold: 0.1,
74            variance_threshold: 0.01,
75            max_iterations: 100,
76            random_seed: None,
77            regularization_alpha: 1.0,
78            max_lag: 10,
79            seasonal_period: None,
80        }
81    }
82}
83
84/// Scoring methods for feature selection
85#[derive(Debug, Clone)]
86pub enum ScoringMethod {
87    /// Mean squared error
88    MeanSquaredError,
89    /// Mean absolute error
90    MeanAbsoluteError,
91    /// R-squared
92    RSquared,
93    /// Akaike Information Criterion
94    AIC,
95    /// Bayesian Information Criterion
96    BIC,
97    /// Cross-validation score
98    CrossValidation,
99}
scirs2_series/feature_selection/mod.rs

scirs2_series/feature_selection/
mod.rs
