GERGM:用于估计广义指数随机图模型的R包

# GERGM -- Master: [](https://travis-ci.org/matthewjdenny/GERGM) [](https://CRAN.R-project.org/package=GERGM)   An R package to estimate Generalized Exponential Random Graph Models. To get started, **[check out this vignette!](http://www.mjdenny.com/getting_started_with_GERGM.html)** **PLEASE REPORT ANY BUGS OR ERRORS TO <mdenny@psu.edu>**. ## News **[05/15/18]** Major estimation updates with version 0.13.x. * I have implemented the convex hull initialization method of Hummel et al (2012) as the default option in the package. This method is often vastly more efficient and effective at initializing the model parameter for Metropolis Hastings, and can result in a 99%+ reduction in model runtime in some situations. * The covariate parameter estiamtion proceedure has been sped up by reimplementing in C++. * The gergm() function now skips MPLE after the first iteration of covariate parameter estimation and uses the previous theta values instead. This often dramatically speeds up estimation, but can be controlled with a logical argument. * For large networks, or networks yielding a very low MH acceptance rate, the `sample_edges_at_a_time` option allows the user to propose blocks of edges at a time in the MH updates. This can be used to optimally tune the model acceptance rate. **[04/13/17]** New estimation functionality, bug fixes. * A bug fix was added for the initialization of the covariate parameter estimates which will lead to faster convergence and more stable estimation. * New option to estimate structural features (no covariates currently) for rowwise marginal distributions. See the `distribution_estimator` argument. * More updates to the documentation to make it easier to read. **[03/12/17]** A number of minor updates to the GERGM package (mostly in the documentation) and bump to version 0.11.0 on CRAN: * The **["getting started with GERGM" vignette](http://www.mjdenny.com/getting_started_with_GERGM.html)** has been updated to include an example using for correlation matrix estimation and well as a section addressing several common errors users tend to encounter. Please check out this section before you email the maintainers with estimation issues. * A number of corections and updates have been made to the function documentation. **[09/17/16]** A major update to the package has been pushed to the public GERGM repo. Here are some highlights: * **correlation networks**: We have added functionality to estimate GERGMS on correlation networks via the `beta_correlation_model` parameter. * **updated conditional edge prediction** The package now offers updated functions for conditional edge prediction and comparison to a null model in terms of edgewise MSE. * **[a "getting started with GERGM" vignette](http://www.mjdenny.com/getting_started_with_GERGM.html)** now includes lots more information on estimating a simple model and generating diagnostic plots. **[08/06/16]** A major update to all aspects of the package has been pushed to the public GERGM repo and will shortly be up on CRAN. **Note that this version does break compatibility with estimation results generated by the previous version of the package.** Here are some highlights: * **parallelization**: We have added a number of different methods of parallelization to the package, both in initialization, and in statistic calculation. * **conditional edge prediction**: The package now provides functionality to predict edge values from a GERGM fit. * **updated diagnostic plots**: The package now calculates additional statistics and provides node degree density plots. * **node subset statistics**: All endogenous statistics (ttriads, etc.) can now be calculated for subsets of nodes in the network, greatly expanding the number of possible specifications. * **explicit intercept specification**: You are now required to explicitly include an intercept term in models. * **slackR integration**: You can now have the `gergm()` function send you updates about estimation to a designated slack channel. This is as close to having the package text message users as we could get. Useful for keeping up to date on long running models. * **estimation without endogenous statistics**: You may now estimate models without any endogenous statistics (covariate effects only), while still having access to all goodness of fit diagnostics. ## Model Overview An R package which implements the Generalized Exponential Random Graph Model (GERGM) with an extension to estimation via Metropolis Hastings. The relevant papers detailing the model can be found at the links below: * Bruce A. Desmarais, and Skyler J. Cranmer, (2012). "Statistical inference for valued-edge networks: the generalized exponential random graph model". PloS One. [[Available Here](http://dx.plos.org/10.1371/journal.pone.0030136)] * James D. Wilson, Matthew J. Denny, Shankar Bhamidi, Skyler Cranmer, and Bruce Desmarais (2017). "Stochastic weighted graphs: Flexible model specification and simulation". Social Networks, 49, 37–47. [[Available Here](http://doi.org/10.1016/j.socnet.2016.11.002)] * Matthew J. Denny (2016). "The Importance of Generative Models for Assessing Network Structure". [[Available Here](http://ssrn.com/abstract=2798493)] To maximize translation across related methods, we have followed many naming conventions used in the [statnet](https://CRAN.R-project.org/package=statnet) suite of packages, and in the [ergm](https://CRAN.R-project.org/package=ergm) package particularly. ## Installation ### Requirements for using C++ code with R See the **Requirements for using C++ code with R** section in the following tutorial: [Using C++ and R code Together with Rcpp](http://www.mjdenny.com/Rcpp_Intro.html). You will likely need to install either `Xcode` or `Rtools` depending on whether you are using a Mac or Windows machine before you can use the package. ### Installing The Package The easiest way to do this is to install the package from CRAN via the standard `install.packages` command: install.packages("GERGM") This will take care of some weird compilation issues that can arise, and is the best option for most people. If you want the most current development version of the package (available here), you will need to start by making sure you have Hadley Wickham's devtools package installed. install.packages("devtools") Now we can install from Github using the following line: devtools::install_github("matthewjdenny/GERGM") I have had success installing this way on most major operating systems with R 3.2.0+ installed, but if you do not have the latest version of R installed, or run into some install errors (please email if you do), it should work as long as you install the dependencies first with the following block of code: install.packages( pkgs = c("BH","RcppArmadillo","ggplot2","methods", "stringr","igraph", "plyr", "parallel", "coda", "vegan", "scales", "RcppParallel","slackr"), dependencies = TRUE) Once the `GERGM` package is installed, you may access its functionality as you would any other package by calling: library(GERGM) If all went well, check out the `?GERGM` help file to see a full working example with info on how the data should look. ## Basic Usage To use this package, first load in the network you wish to use as a (square) matrix, following the example provided below. You may then use the `gergm()` function to estimate a model using any combination of the following statistics: `out2stars(alpha = 1)`, `in2stars(alpha = 1)`, `ctriads(alpha = 1)`, `mutual(alpha = 1)`, `ttriads(alpha = 1)`, `absdiff(covariate = "MyCov")`, `sender(covariate = "MyCov")`, `reciever(covariate = "MyCov")`, `nodematch(covariate)`, `nodemix(covariate, base = "MyBa