---
title: "Penalized Precision Matrix Estimation in grasps"
bibliography: ../inst/REFERENCES.bib
vignette: >
  %\VignetteIndexEntry{pen_est}
  %\VignetteEngine{quarto::html}
  %\VignetteEncoding{UTF-8}
knitr:
  opts_chunk: 
    collapse: false
    comment: '#>'
    echo: true
    fig.align: "center"
    message: false
    results: 'hide'
    warning: false
---


## Preliminary


Consider the following setting:


- **Gaussian graphical model (GGM) assumption:** \
  The data $X_{p \times p}$ consists of independent and identically distributed
  samples $X_1, \dots, X_n \sim N_p(\mu,\Sigma)$.


- **Disjoint group structure:** \
  The $p$ variables can be partitioned into disjoint groups.


- **Goal:** \
  Estimate the precision matrix $\Omega = \Sigma^{-1} = (\omega_{ij})_{p \times p}$.


## Sparse-Group Estimator


\begin{gather}
\hat{\Omega}(\lambda,\alpha,\gamma)
= {\arg\min}_{\Omega \succ 0} \left\{ -\log\det(\Omega) + \text{tr}(S\Omega)
+ \mathcal{P}_{\lambda,\alpha,\gamma}(\Omega) \right\},
\\[10pt]
\mathcal{P}_{\lambda,\alpha,\gamma}(\Omega)
= \alpha \mathcal{P}^\text{idv}_{\lambda,\gamma}(\Omega) + (1-\alpha) \mathcal{P}^\text{grp}_{\lambda,\gamma}(\Omega),
\\[10pt]
\mathcal{P}^\text{idv}_{\lambda,\gamma}(\Omega)
= \sum_{i,j} P_{\lambda,\gamma}(\lvert\omega_{ij}\rvert),
\\[5pt]
\mathcal{P}^\text{grp}_{\lambda,\gamma}(\Omega)
= \sum_{g,g^\prime} P_{\lambda,\gamma}(\lVert\Omega_{gg^\prime}\rVert_F).
\end{gather}


where:


- $S = n^{-1} \sum_{i=1}^n (X_i-\bar{X})(X_i-\bar{X})^\top$ is the empirical
  covariance matrix.


- $\lambda \geq 0$ is the global regularization parameter controlling overall
  shrinkage.


- $\alpha \in [0,1]$ is the mixing parameter controlling the balance between
  element-wise and block-wise penalties.


- $\gamma$ is the additional parameter for non-convex penalties, controlling
  the degree of nonconvexity (or concavity) of the penalty function.


- $\mathcal{P}_{\lambda,\alpha,\gamma}(\Omega)$ is a generic bi-level penalty
  template that combines element-wise and block-wise regularization, allowing
  convex or non-convex regularizers while preserving the intrinsic group
  structure among variables.


- $\mathcal{P}^\text{idv}_{\lambda,\gamma}(\Omega)$ is the element-wise
  individual penalty component.


- $\mathcal{P}^\text{grp}_{\lambda,\gamma}(\Omega)$ is the block-wise
  group penalty component.


- $P_{\lambda,\gamma}(\cdot)$ is the penalty function.


- $\Omega_{gg^\prime}$ is the submatrix of $\Omega$ with the rows from group $g$
  and columns from group $g^\prime$.


- The Frobenius norm $\lVert\Omega\rVert_F$ is defined as
  $\lVert\Omega\rVert_F = (\sum_{i,j} \lvert\omega_{ij}\rvert^2)^{1/2} = [\text{tr}(\Omega^\top\Omega)]^{1/2}$.


<div class="note">
**Note**:

+ The parameter $\gamma$ is only relevant for non-convex penalties. The Lasso
  penalty can be viewed as a special case in which $\gamma$ is not required. </div>


## Penalties


1. Lasso: Least absolute shrinkage and selection operator
[@tibshirani1996regression; @friedman2008sparse]


$$P_\lambda(\omega_{ij}) = \lambda\vert\omega_{ij}\vert.$$


2. Adaptive lasso [@zou2006adaptive; @fan2009network]


$$
P_{\lambda,\gamma}(\omega_{ij}) = \lambda\frac{\vert\omega_{ij}\vert}{v_{ij}},
$$
where $V = (v_{ij})_{d \times d} = (\vert\tilde{\omega}_{ij}\vert^\gamma)_{d \times d}$
is a matrix of adaptive weights, and $\tilde{\omega}_{ij}$ is the initial estimate
obtained using `penalty = "lasso"`.


3. Atan: Arctangent type penalty [@wang2016variable]


$$
P_{\lambda,\gamma}(\omega_{ij})
= \lambda(\gamma+\frac{2}{\pi})
\arctan\left(\frac{\vert\omega_{ij}\vert}{\gamma}\right),
\quad \gamma > 0.
$$


4. Exp: Exponential type penalty [@wang2018variable]


$$
P_{\lambda,\gamma}(\omega_{ij})
= \lambda\left[1-\exp\left(-\frac{\vert\omega_{ij}\vert}{\gamma}\right)\right],
\quad \gamma > 0.
$$


5. Lq [@frank1993statistical; @fu1998penalized; @fan2001variable]


$$
P_{\lambda,\gamma}(\omega_{ij}) = \lambda\vert\omega_{ij}\vert^\gamma,
\quad 0 < \gamma < 1.
$$


6. LSP: Log-sum penalty [@candes2008enhancing]


$$
P_{\lambda,\gamma}(\omega_{ij})
= \lambda\log\left(1+\frac{\vert\omega_{ij}\vert}{\gamma}\right),
\quad \gamma > 0.
$$


7. MCP: Minimax concave penalty [@zhang2010nearly]


$$
P_{\lambda,\gamma}(\omega_{ij})
= \begin{cases}
\lambda\vert\omega_{ij}\vert - \dfrac{\omega_{ij}^2}{2\gamma},
& \text{if } \vert\omega_{ij}\vert \leq \gamma\lambda, \\
\dfrac{1}{2}\gamma\lambda^2,
& \text{if } \vert\omega_{ij}\vert > \gamma\lambda.
\end{cases}
\quad \gamma > 1.
$$


8. SCAD: Smoothly clipped absolute deviation [@fan2001variable; @fan2009network]


$$
P_{\lambda,\gamma}(\omega_{ij})
= \begin{cases}
\lambda\vert\omega_{ij}\vert
& \text{if } \vert\omega_{ij}\vert \leq \lambda, \\
\dfrac{2\gamma\lambda\vert\omega_{ij}\vert-\omega_{ij}^2-\lambda^2}{2(\gamma-1)}
& \text{if } \lambda < \vert\omega_{ij}\vert < \gamma\lambda, \\
\dfrac{\lambda^2(\gamma+1)}{2}
& \text{if } \vert\omega_{ij}\vert \geq \gamma\lambda.
\end{cases}
\quad \gamma > 2.
$$


<div class="note">
**Note**:

+ For Lasso, which is convex, the additional parameter $\gamma$ is not required,
  and the penalty function $P_{\lambda,\gamma}(\cdot)$ simplifies to
  $P_\lambda(\cdot)$. </div>


## Illustrative Visualization


@fig-pen illustrates a comparison of various penalty functions $P(\omega)$
evaluated over a range of $\omega$ values. The main panel (right) provides
a wider view of the penalty functions' behavior for larger $\vert\omega\vert$,
while the inset panel (left) magnifies the region near zero $[-1, 1]$.


```{r fig.show='hide'}
library(grasps) ## for penalty computation
library(ggplot2) ## for visualization

penalties <- c("atan", "exp", "lasso", "lq", "lsp", "mcp", "scad")

pen_df <- compute_penalty(seq(-4, 4, by = 0.01), penalties, lambda = 1)
plot(pen_df, xlim = c(-1, 1), ylim = c(0, 1), zoom.size = 1) +
  guides(color = guide_legend(nrow = 2, byrow = TRUE))
```


```{r echo=FALSE, fig.show='hide'}
plot <- plot(pen_df, xlim = c(-1, 1), ylim = c(0, 1), zoom.size = 1) +
  guides(color = guide_legend(nrow = 2, byrow = TRUE))
```


::: {#fig-pen}
```{r echo=FALSE}
print(plot)
```
Illustrative penalty functions.
:::


@fig-deriv displays the derivative function $P^\prime(\omega)$ associated with
a range of penalty types.
The Lasso exhibits a constant derivative, corresponding to uniform shrinkage.
For MCP and SCAD, the derivatives are piecewise: initially equal to the Lasso
derivative, then decreasing over an intermediate region, and eventually dropping
to zero, indicating that large $\vert\omega\vert$ receive no shrinkage.
Other non-convex penalties show smoothly diminishing derivatives as
$\vert\omega\vert$ increases, reflecting their tendency to shrink small
$\vert\omega\vert$ strongly while exerting little to no shrinkage on large ones.


```{r fig.show='hide'}
deriv_df <- compute_derivative(seq(0, 4, by = 0.01), penalties, lambda = 1)
plot(deriv_df) +
  scale_y_continuous(limits = c(0, 1.5)) +
  guides(color = guide_legend(nrow = 2, byrow = TRUE))
```


```{r echo=FALSE, fig.show='hide'}
plot <- plot(deriv_df) +
  scale_y_continuous(limits = c(0, 1.5)) +
  guides(color = guide_legend(nrow = 2, byrow = TRUE))
```


::: {#fig-deriv}
```{r echo=FALSE}
print(plot)
```
Illustrative penalty derivatives.
:::


## Reference {-}