missoNet (version 1.5.1)

missoNet-package: missoNet: Multi-Task Regression and Conditional Network Estimation with Missing Responses

Description

missoNet fits a joint multivariate regression and conditional dependency (precision–matrix) model when some response entries are missing. The method estimates a sparse coefficient matrix \(B\) linking predictors \(X\) to multivariate responses \(Y\), together with a sparse inverse covariance \(\Theta\) for the residuals in \(Y = \mathbf{1}\mu^{\mathsf{T}} + XB + E\), \(E \sim \mathcal{N}(0, \Theta^{-1})\). Responses may contain missing values (e.g., MCAR/MAR); predictors must be finite. The package provides cross-validation, prediction, publication-ready plotting, and simple simulation utilities.

Main functions

missoNet

Fit models over user-specified penalty grids for \(\lambda_B\) and \(\lambda_\Theta\); returns estimated \(\mu\), \(B\), \(\Theta\), and metadata (grids, GoF).

cv.missoNet

Perform k-fold cross-validation over a penalty grid; stores est.min and (optionally) est.1se.beta, est.1se.theta.

plot.missoNet

S3 plotting method; heatmap or 3D scatter of CV error or GoF.

predict.missoNet

S3 prediction method; returns \(\hat{Y} = \mathbf{1}\hat{\mu}^{\mathsf{T}} + X_\mathrm{new}\hat{B}\) for a chosen solution.

generateData

Generate synthetic datasets with controllable dimensions, signal, and missingness mechanisms for benchmarking.

License

GPL-2.

Author

Maintainer: Yixiao Zeng [copyright holder]

Authors:

• Celia Greenwood [thesis advisor]

Details

Key features

• Joint estimation of \(B\) (regression) and \(\Theta\) (conditional network).

• \(\ell_1\)-regularization on both \(B\) and \(\Theta\) with user-controlled grids.

• K-fold cross-validation with optional 1-SE model selections.

• Heatmap and 3D surface visualizations for CV error or GoF across \((\lambda_B,\lambda_\Theta)\).

• Fast prediction for new data using stored solutions.

• Lightweight data generator for simulation studies.

Workflow

1. Fit a model across a grid of penalties with missoNet or select penalties via cv.missoNet.

2. Visualize the CV error/GoF surface with plot.missoNet.

3. Predict responses for new observations with predict.missoNet.

See Also

missoNet, cv.missoNet, plot.missoNet, predict.missoNet, generateData, and browseVignettes("missoNet") for tutorials.

Examples

sim <- generateData(n = 100, p = 8, q = 5, rho = 0.1, missing.type = "MCAR")

# \donttest{
fit <- missoNet(X = sim$X, Y = sim$Z)             # fit over a grid
plot(fit)                                         # GoF heatmap

cvfit <- cv.missoNet(X = sim$X, Y = sim$Z, kfold = 5, compute.1se = TRUE)
plot(cvfit, type = "scatter", plt.surf = TRUE)    # CV error surface
yhat  <- predict(cvfit, newx = sim$X, s = "lambda.min")
# }