fdars_core/explain_generic/
pdp.rs

1use crate::error::FdarError;
2use crate::explain::{ice_to_pdp, make_grid, FunctionalPdpResult};
3use crate::matrix::FdMatrix;
4
5use super::FpcPredictor;
6
7/// Generic partial dependence plot / ICE curves for any FPC-based model.
8///
9/// # Errors
10///
11/// Returns [`FdarError::InvalidDimension`] if `data` has zero rows or its
12/// column count does not match the model's FPCA mean length.
13/// Returns [`FdarError::InvalidParameter`] if `component >= ncomp` or
14/// `n_grid < 2`.
15#[must_use = "expensive computation whose result should not be discarded"]
16pub fn generic_pdp(
17    model: &dyn FpcPredictor,
18    data: &FdMatrix,
19    scalar_covariates: Option<&FdMatrix>,
20    component: usize,
21    n_grid: usize,
22) -> Result<FunctionalPdpResult, FdarError> {
23    let (n, m) = data.shape();
24    if n == 0 {
25        return Err(FdarError::InvalidDimension {
26            parameter: "data",
27            expected: "n > 0".into(),
28            actual: "0 rows".into(),
29        });
30    }
31    if m != model.fpca_mean().len() {
32        return Err(FdarError::InvalidDimension {
33            parameter: "data columns",
34            expected: model.fpca_mean().len().to_string(),
35            actual: m.to_string(),
36        });
37    }
38    if component >= model.ncomp() {
39        return Err(FdarError::InvalidParameter {
40            parameter: "component",
41            message: format!("component {} >= ncomp {}", component, model.ncomp()),
42        });
43    }
44    if n_grid < 2 {
45        return Err(FdarError::InvalidParameter {
46            parameter: "n_grid",
47            message: format!("n_grid must be >= 2, got {n_grid}"),
48        });
49    }
50    let ncomp = model.ncomp();
51    let scores = model.project(data);
52    let grid_values = make_grid(&scores, component, n_grid);
53
54    let p_scalar = scalar_covariates.map_or(0, crate::matrix::FdMatrix::ncols);
55    let mut ice_curves = FdMatrix::zeros(n, n_grid);
56    // Pre-allocate buffers outside the observation loop and reuse them
57    let mut obs_scores = vec![0.0; ncomp];
58    let mut obs_z = vec![0.0; p_scalar];
59    for i in 0..n {
60        for k in 0..ncomp {
61            obs_scores[k] = scores[(i, k)];
62        }
63        let obs_z_slice: Option<&[f64]> = if p_scalar > 0 {
64            if let Some(sc) = scalar_covariates {
65                for j in 0..p_scalar {
66                    obs_z[j] = sc[(i, j)];
67                }
68                Some(&obs_z)
69            } else {
70                None
71            }
72        } else {
73            None
74        };
75        for g in 0..n_grid {
76            obs_scores[component] = grid_values[g];
77            ice_curves[(i, g)] = model.predict_from_scores(&obs_scores, obs_z_slice);
78        }
79    }
80
81    let pdp_curve = ice_to_pdp(&ice_curves, n, n_grid);
82
83    Ok(FunctionalPdpResult {
84        grid_values,
85        pdp_curve,
86        ice_curves,
87        component,
88    })
89}
fdars_core/explain_generic/pdp.rs

fdars_core/explain_generic/
pdp.rs
