# `robust_pls` — Robust PLS (Partial Robust M-regression)
_Group_: **Robust / weighted** · _Registry tolerance_: `1e-08`
## Description
Robust PLS (Partial Robust M-regression, Serneels 2005)
From the `pls4all.sklearn.RobustPLSRegression` docstring:
> Robust PLS via Huber IRLS over weighted SIMPLS.
> **Registry note** — R `chemometrics::prm` (Serneels et al. 2005) — Partial Robust M-regression. pls4all defaults to PRM matching the R algorithm bit-for-bit (median centering, Fair weights on leverage + residual, univariate SIMPLS inner kernel, intercept = median(y - X@b)). The legacy Huber-IRLS over weighted SIMPLS path is reachable via ``cfg.robust_pls_legacy = 1``.
### Parameters
| Name | Type | Default | Notes |
|------|------|---------|-------|
| `n_components` | `int` | `2` | Number of latent components extracted (k). |
| `huber_k` | `float` | `1.345` | Huber threshold (in residual-stdev units) controlling IRLS reweighting; smaller = more robust. |
| `max_irls_iter` | `int` | `20` | Maximum IRLS reweighting iterations. |
## Explanations
### Bibliographic source
Serneels, S., Croux, C., Filzmoser, P. & Van Espen, P. J. (2005). *Partial Robust M-Regression*. Chemometrics and Intelligent Laboratory Systems 79(1–2), 55–64.
### Mathematical principle
Robust PLS performs a sequence of weighted PLS fits where the weights $w_i$ are reduced for samples with large current residuals — the iteratively-reweighted least-squares (IRLS) recipe.
At each iteration, the weight of sample $i$ is set from Huber's $\psi$-function applied to the standardised residual: $w_i = \psi(r_i / s) / r_i$ where $\psi(z) = z$ for $|z| \le k$ and $\psi(z) = k\,\operatorname{sign}(z)$ otherwise. $k = 1.345$ gives 95 % asymptotic efficiency at the Gaussian. The robust scale $s$ is typically the MAD of the residuals.
Convergence is rapid: 3–5 iterations typically suffice. Robust PLS down-weights — rather than removes — outliers, which is desirable when outlier-ness is a continuous concept (mild spectral artefacts) rather than binary (broken samples).
Compared to median-PLS variants, the M-regression form preserves the analytic structure of SIMPLS and offers smooth weighting; it also generalises to leverage-based weights (PRM with x-weights).
### Implementation
`n4m_estimators_robust_pls_fit`. Reference: CRAN `chemometrics::prm` (Serneels et al. authors). The exact weight schedule and scale estimator differ from `prm` so RMSE-rel parity is widened to ~2.0 to flag presence rather than enforce exact agreement.
R roxygen note (`sklearn_extra.R::robust_pls`):
> Robust PLS — formula entry point.
MATLAB header (`bindings/matlab/+pls4all/RobustPlsRegression.m`):
```text
pls4all.RobustPlsRegression Robust PLS via Huber IRLS.
```
### Usage
Every pls4all binding tab dispatches into the same C kernel; the external libraries listed at the bottom of the page are the parity references registered in `benchmarks.parity_timing.registry`. Switch tabs to read the same fit in your language. The R package now ships drop-in-compatible facades for the CRAN `pls` package (`plsr`, `pcr`, `mvr`) and for the `mdatools::pls(x, y, ...)` matrix idiom — those tabs appear only on the methods that have a meaningful equivalence.
**pls4all bindings**
::::{tab-set}
:class: pls4all-bindings
:::{tab-item} C ABI · libn4m
:sync: c
:class-label: lang-c
```c
/* C ABI — libn4m */
n4m_context_t* ctx = n4m_context_create();
n4m_config_t* cfg = n4m_config_create();
n4m_method_result_t* res = NULL;
n4m_estimators_robust_pls_fit(ctx, cfg, &x_view, &y_view, /* hyperparams */, &res);
/* … read coefficients / mask / scores via */
/* n4m_method_result_get_double_matrix / vector / scalar … */
n4m_method_result_destroy(res);
n4m_config_destroy(cfg);
n4m_context_destroy(ctx);
```
:::
:::{tab-item} Python · pls4all (raw)
:sync: python-raw
:class-label: lang-python
```python
import pls4all
from pls4all._methods import robust_pls_fit
with pls4all.Context() as ctx, pls4all.Config() as cfg:
res = robust_pls_fit(ctx, cfg, X, y, n_components=4)
# then: res.matrix("predictions"), res.matrix("coefficients"),
# res.vector("mask"), res.scalar("intercept"), …
```
:::
:::{tab-item} Python · pls4all.sklearn
:sync: python-sklearn
:class-label: lang-python
```python
from pls4all.sklearn import RobustPLSRegression
mdl = RobustPLSRegression(n_components=2, huber_k=1.345, max_irls_iter=20)
mdl.fit(X, y)
y_hat = mdl.predict(X_test)
```
:::
:::{tab-item} R · pls4all_method()
:sync: r-dispatcher
:class-label: lang-r
```r
library(pls4all)
# Unified low-level dispatcher (May 2026 R cleanup):
res <- pls4all_method("robust_pls", X, y,
n_components = 4L, params = list(huber_k = 4.0, max_irls_iter = 30L))
# res is a named list with MethodResult arrays/scalars.
# selected_indices / top_k_intervals are 1-based.
```
:::
:::{tab-item} R · pls4all (raw fn)
:sync: r-raw
:class-label: lang-r
```r
library(pls4all)
res <- robust_pls_fit(X, Y, n_components,
huber_k = 1.345, max_irls_iter = 20L)
yhat <- pls4all_predict(res, X_test)
```
:::
:::{tab-item} R · pls4all (formula+S3)
:sync: r-formula
:class-label: lang-r
```r
library(pls4all)
fit <- robust_pls(y ~ ., data = train, ncomp = 4L)
yhat <- predict(fit, newdata = test)
summary(fit)
```
:::
:::{tab-item} MATLAB · pls4all (MEX)
:sync: matlab-mex
:class-label: lang-matlab
```matlab
res = pls4all.robust_pls(X, y, 4);
% see header of bindings/matlab/+pls4all/robust_pls.m for full
% parameter surface:
% res = robust_pls(X, Y, n_components, huber_k, max_irls_iter)
yhat = predict(res, Xtest);
```
:::
:::{tab-item} MATLAB · pls4all (classdef)
:sync: matlab-classdef
:class-label: lang-matlab
```matlab
mdl = pls4all.fit("robust_pls", X, y, "NumComponents", 4);
yhat = predict(mdl, Xtest);
```
:::
::::
**Registry parity references** 📐
:::{card}
:class-card: external-refs
- 📐 **`ref.r_chemometrics`** (R · r) — `chemometrics` 0.7.x · strict (rmse_rel ≤ 1e-08) — R `chemometrics::prm` (Partial Robust M-regression). pls4all uses Huber IRLS over weighted SIMPLS; this is an M-estimator variant from the same family. Predictions diverge by O(0.5).
:::
### Benchmarks
Adaptive wall-clock per cell measured against [`full_matrix.csv`](../benchmarks/overview.md). Only backends that implement this method are listed; libraries without the method are omitted.
**Verdict** · ✓ ref / ≈ ref / ~ shape mark a reference-gate pass at strict / relaxed / qualitative tolerance · ✓ bind = pls4all binding agrees with the C++ baseline · ⇄ cross-check = documented by-design selector/RNG/model, noncanonical API/facade convention, or secondary oracle · ✗ divergent · ⚠ error · — not run. The fastest backend per column is marked 🏆.
**Reference gate**: strict — numeric equivalence (`rmse_rel_tol ≤ 1e-08`).
Rows tagged with **📐** are the canonical parity references for this method (declared in [`parity_timing.registry`](../benchmarks/methodology.md)). C++ and external rows show reference parity; pls4all language bindings show binding parity against the C++ backend. Hover the icon for role and tolerance band.
::::{tab-set}
:class: parity-tabs
:::{tab-item} 1 thread
:sync: threads-1
| Backend | Parity | 200×50 (ms) |
| C++ native · libn4m |
pls4all.cpp.blas+omp | ✓ ref 1e-14 | 2.07 ms |
| Python · pls4all |
pls4all.python | ✓ bind | 1.95 ms🏆 |
pls4all.sklearn | ✓ bind | 2.15 ms |
| R · pls4all |
pls4all.R | ✓ bind | 4.38 ms |
pls4all.R.formula | ✓ bind | 5.55 ms |
pls4all.R.mdatools | ✓ bind | 5.77 ms |
pls4all.R.pls | ✓ bind | 6.62 ms |
| R · external |
📐ref.r_chemometrics | source | 17.9 ms |
:::
:::{tab-item} 3 threads
:sync: threads-3
| Backend | Parity | 200×50 (ms) |
| C++ native · libn4m |
pls4all.cpp.blas+omp | ✓ ref 1e-14 | 2.02 ms |
| Python · pls4all |
pls4all.python | ✓ bind | 2.00 ms🏆 |
pls4all.sklearn | ✓ bind | 2.10 ms |
| R · pls4all |
pls4all.R | ✓ bind | 5.43 ms |
pls4all.R.formula | ✓ bind | 6.63 ms |
pls4all.R.mdatools | ✓ bind | 5.84 ms |
pls4all.R.pls | ✓ bind | 5.99 ms |
| R · external |
📐ref.r_chemometrics | source | 17.0 ms |
:::
:::{tab-item} 10 threads
:sync: threads-10
| Backend | Parity | 200×50 (ms) |
| C++ native · libn4m |
pls4all.cpp.blas+omp | ✓ ref 1e-14 | 2.05 ms |
| Python · pls4all |
pls4all.python | ✓ bind | 1.98 ms🏆 |
pls4all.sklearn | ✓ bind | 2.25 ms |
| R · pls4all |
pls4all.R | ✓ bind | 4.94 ms |
pls4all.R.formula | ✓ bind | 7.07 ms |
pls4all.R.mdatools | ✓ bind | 5.59 ms |
pls4all.R.pls | ✓ bind | 5.68 ms |
| R · external |
📐ref.r_chemometrics | source | 17.5 ms |
:::
::::
---
_See also_: [benchmark overview](../benchmarks/overview.md) · [methods index](index.md) · [interactive dashboard](../landing/dashboard.md)