Package 'inet'

Title: Performing Inference on Networks with Regularization
Description: Performs inference with the lasso in Gaussian Graphical Models. The package consists of wrappers for functions from the hdi package.
Authors: Lourens Waldorp <[email protected]>, Jonas Haslbeck <[email protected]>
Maintainer: Jonas Haslbeck <[email protected]>
License: GPL (>= 2)
Version: 0.1.0
Built: 2025-03-10 03:33:27 UTC
Source: https://github.com/jmbh/inet

Help Index

Datasets included in inet package

Description

The package includes a dataset with measurements of 17 PTSD symptoms taken from 344 individuals. See McNally et al. (2015) for more details.

Author(s)

Jonas Haslbeck

References

McNally, R. J., Robinaugh, D. J., Wu, G. W., Wang, L., Deserno, M. K., & Borsboom, D. (2015). Mental disorders as causal systems: A network approach to posttraumatic stress disorder. Clinical Psychological Science, 3(6), 836-849.

Internal inet functions

Description

Internal inet functions.

Details

These are internal functions.

Author(s)

Jonas Haslbeck

Estimate GGM with nodewise regression and the lasso.

Description

Estimate a Gaussian Graphical Model with lasso-regularized nodewise regression, where the regularization parameter is selected with cross-validation. This is a wrapper around the function cv.glmnet() from the glmnet package.

Usage

lasso(data, pbar = TRUE, nfolds = 10, rulereg = "and")

Arguments

data

An n x p matrix containing the data, where n are cases and p are variables

pbar

If pbar = TRUE, a progress bar will be displayed.

nfolds

Specifies the number of folds used to select the regularization parameter in each of the p nodewise regressions.

rulereg

Specifies how parameter estimates should be combined across nodewise regressions. The options are the AND-rule (requiring both estimates to be significant) or the OR-rule (only requiring one estimate to be significant). Defaults to rulereg = "and".

Value

The function returns a list with the following entries:

est

A p x p matrix with point estimates for all partial correlations
select

A p x p indicator matrix indicating which edges have been selected to be present.
ints

A p-vector of estimated intercepts.

Author(s)

Jonas Haslbeck <[email protected]>

References

Friedman, J., Hastie, T., & Tibshirani, R. (2010). Regularization paths for generalized linear models via coordinate descent. Journal of statistical software, 33(1), 1.

Examples

## Not run: 

# Fit GGM to PTSD data
set.seed(1)
out <- lasso(data = ptsd_data)


## End(Not run)

Estimate GGMs with the desparsified lasso.

Description

Estimate Gaussian Graphical Models using the desparsified lasso. This is a wrapper around the function lasso.proj of the hdi package.

Usage

lasso_dsp(data, betainit = "cv lasso", ci.level = 0.95,
          correction = TRUE, pbar = TRUE, rulereg = "and")

Arguments

data

An n x p matrix containing the data, where n are cases and p are variables

betainit

Specifying how to estimate lasso solution in initial estimation. Either betainit = "cv lasso" (default) or betainit = "cv lasso". See the manual of the function lasso.proj of the hdi package for more info.

ci.level

Specifies the width of the confidence interval used for testing the null hypothesis that a parameter is different to zero. Defaults to ci.level = 0.95, which corresponds to a critical threshold of 0.05.

correction

If correction = TRUE, the Bonferroni-Holm correction will be applied to p-values on the level of nodewise regressions (see e.g., Hochberg, 1987).

pbar

If pbar = TRUE, a progress bar will be displayed.

rulereg

Specifies how parameter estimates should be combined across nodewise regressions. The options are the AND-rule (requiring both estimates to be significant) or the OR-rule (only requiring one estimate to be significant). Defaults to rulereg = "and".

Value

The function returns a list with the following entries:

est

A p x p matrix with point estimates for all partial correlations
est.signf

A p x p matrix with point estimates for all partial correlations with non-significant partial correlations being thresholded to zero.
signf

A p x p matrix indicating for each partial correlation whether it is significantly different to zero.
ci.lower

A p x p matrix indicating the lower confidence interval for each partial correlation.
ci.upper

A p x p matrix indicating the upper confidence interval for each partial correlation.

Author(s)

Jonas Haslbeck <[email protected]>; Lourens Waldorp <[email protected]>

References

Hochberg, Y., & Tamhane, A. C. (1987). Multiple comparison procedures. John Wiley & Sons, Inc..

Buehlmann, P., Kalisch, M., & Meier, L. (2014). High-dimensional statistics with a view toward applications in biology. Annual Review of Statistics and Its Application, 1, 255-278.

Examples

## Not run: 

# Fit GGM to PTSD data
set.seed(1)
out <- lasso_dsp(data = ptsd_data)


## End(Not run)

Estimate GGMs with the desparsified lasso using the bootstrap.

Description

Estimate Gaussian Graphical Models using the desparsified lasso using the bootstrap. This is a wrapper around the function lasso.proj of the hdi package.

Usage

lasso_dsp_boot(data, betainit = "cv lasso", ci.level = 0.95,
                   correction = TRUE, B = 1000, pbar = TRUE,
                   rulereg = "and")

Arguments

data

An n x p matrix containing the data, where n are cases and p are variables

betainit

Specifying how to estimate lasso solution in initial estimation. Either betainit = "cv lasso" (default) or betainit = "cv lasso". See the manual of the function lasso.proj of the hdi package for more info.

ci.level

Specifies the width of the confidence interval used for testing the null hypothesis that a parameter is different to zero. Defaults to ci.level = 0.95, which corresponds to a critical threshold of 0.05.

correction

If correction = TRUE, the Bonferroni-Holm correction will be applied to p-values on the level of nodewise regressions (see e.g., Hochberg, 1987).

B

The number of bootstrap samples used for estimation. Defaults to B=1000.

pbar

If pbar = TRUE, a progress bar will be displayed.

rulereg

Specifies how parameter estimates should be combined across nodewise regressions. The options are the AND-rule (requiring both estimates to be significant) or the OR-rule (only requiring one estimate to be significant). Defaults to rulereg = "and".

Value

The function returns a list with the following entries:

est

A p x p matrix with point estimates for all partial correlations
est.signf

A p x p matrix with point estimates for all partial correlations with non-significant partial correlations being thresholded to zero.
signf

A p x p matrix indicating for each partial correlation whether it is significantly different to zero.
ci.lower

A p x p matrix indicating the lower confidence interval for each partial correlation.
ci.upper

A p x p matrix indicating the upper confidence interval for each partial correlation.

Author(s)

Jonas Haslbeck <[email protected]>; Lourens Waldorp <[email protected]>

References

Hochberg, Y., & Tamhane, A. C. (1987). Multiple comparison procedures. John Wiley & Sons, Inc..

Bühlmann, P., Kalisch, M., & Meier, L. (2014). High-dimensional statistics with a view toward applications in biology. Annual Review of Statistics and Its Application, 1, 255-278.

Davison, A. C., & Hinkley, D. V. (1997). Bootstrap methods and their application (No. 1). Cambridge university press.

Examples

## Not run: 

# Fit GGM to PTSD data
set.seed(1)
out <- lasso_dsp_boot(data = ptsd_data)


## End(Not run)

Estimate GMM with inference via the multi-split method.

Description

Estimate Gaussian Graphical Models with inference base don the multi-split method. This is a wrapper of the function multi.split of the hdi package.

Usage

lasso_ms(data, B = 50, fraction = 0.5, ci.level = 0.95,
         correction = TRUE, pbar = TRUE, rulereg = "and")

Arguments

data

An n x p matrix containing the data, where n are cases and p are variables

B

The number of sample-splits. Defaults to B=50.

fraction

a number in (0,1), the fraction of data used at each sample split for the model selection process. The remaining data is used for calculating the p-values.

ci.level

Specifies the width of the confidence interval used for testing the null hypothesis that a parameter is different to zero. Defaults to ci.level = 0.95, which corresponds to a critical threshold of 0.05.

correction

If correction = TRUE, the Bonferroni-Holm correction will be applied to p-values on the level of nodewise regressions (see e.g., Hochberg, 1987).

pbar

If pbar = TRUE, a progress bar will be displayed.

rulereg

Specifies how parameter estimates should be combined across nodewise regressions. The options are the AND-rule (requiring both estimates to be significant) or the OR-rule (only requiring one estimate to be significant). Defaults to rulereg = "and".

Value

The function returns a list with the following entries:

est

A p x p matrix with point estimates for all partial correlations
est.signf

A p x p matrix with point estimates for all partial correlations with non-significant partial correlations being thresholded to zero.
signf

A p x p matrix indicating for each partial correlation whether it is significantly different to zero.
ci.lower

A p x p matrix indicating the lower confidence interval for each partial correlation.
ci.upper

A p x p matrix indicating the upper confidence interval for each partial correlation.

Author(s)

Jonas Haslbeck <[email protected]>; Lourens Waldorp <[email protected]>

References

Hochberg, Y., & Tamhane, A. C. (1987). Multiple comparison procedures. John Wiley & Sons, Inc..

Wasserman, L., & Roeder, K. (2009). High dimensional variable selection. Annals of statistics, 37(5A), 2178.

Meinshausen, N., Meier, L., & Bühlmann, P. (2009). P-values for high-dimensional regression. Journal of the American Statistical Association, 104(488), 1671-1681.

Examples

## Not run: 

# Fit GGM to PTSD data
set.seed(1)
out <- lasso_ms(data = ptsd_data)


## End(Not run)

Estimate GMM via nodewise regression and hypothesis tests.

Description

Estimate Gaussian Graphical Model with nodewise regression, selecting edges with standard hypothesis tests and the Bonferroni-Holm Correction.

Usage

OLS(data, pbar = TRUE, correction = TRUE,
    ci.level = 0.95, rulereg = "and")

Arguments

data

An n x p matrix containing the data, where n are cases and p are variables

pbar

If pbar = TRUE, a progress bar will be displayed.

correction

If correction = TRUE, the Bonferroni-Holm correction will be applied to p-values on the level of nodewise regressions (see e.g., Hochberg, 1987).

ci.level

Specifies the width of the confidence interval used for testing the null hypothesis that a parameter is different to zero. Defaults to ci.level = 0.95, which corresponds to a critical threshold of 0.05.

rulereg

Specifies how parameter estimates should be combined across nodewise regressions. The options are the AND-rule (requiring both estimates to be significant) or the OR-rule (only requiring one estimate to be significant). Defaults to rulereg = "and".

Value

The function returns a list with the following entries:

est

A p x p matrix with point estimates for all partial correlations
est.signf

A p x p matrix with point estimates for all partial correlations with non-significant partial correlations being thresholded to zero.
signf

A p x p matrix indicating for each partial correlation whether it is significantly different to zero.
ci.lower

A p x p matrix indicating the lower confidence interval for each partial correlation.
ci.upper

A p x p matrix indicating the upper confidence interval for each partial correlation.
ints

A p-vector of estimated intercepts.

Author(s)

Jonas Haslbeck <[email protected]>

References

Hochberg, Y., & Tamhane, A. C. (1987). Multiple comparison procedures. John Wiley & Sons, Inc..

Examples

## Not run: 

# Fit GGM to PTSD data
out <- OLS(data = ptsd_data)


## End(Not run)

Plot point estimates and confidence intervals

Description

Plot point estimates and confidence intervals for models estimated with the lasso_ms, lasso_dsp, lasso_dsp_boot and OLS functions.

Usage

## S3 method for class 'inet'
plot(x, labels = NULL, order = FALSE, subset = NULL,
          cex.labels = 0.80, cex.axis = 0.75, ...)

Arguments

x

The output object from either lasso_ms, lasso_dsp, lasso_dsp_boot or OLS.

labels

A p-vector of characters specifying the labels for variables.

order

If order = TRUE, the edges are listed in decreasing order based on the point estimate.

subset

Allows to only display a subset of the edges. For example, if subset=1:20 the first 20 edges are displayed. This is especially useful for larger networks, in which all edges are unlikely to fit into a single figure.

cex.labels

The font size of the edge labels.

cex.axis

The font size of the axes.

...

Additional arguments.

Author(s)

Jonas Haslbeck <[email protected]>

Examples

## Not run: 

# Fit GGM to PTSD data
set.seed(1)
out <- lasso_dsp(data = ptsd_data)

# Plot first 20 edges
plot(out, labels = colnames(ptsd_data),
     order=TRUE, subset = 1:20)


## End(Not run)