R package

.Rproj.user/3992D3D4/sources/prop/4F1C9B07

@@ -0,0 +1 @@
+{}

.Rproj.user/3992D3D4/sources/prop/AA8888B6

@@ -0,0 +1,5 @@
+{
+    "tempName": "Untitled1",
+    "cursorPosition": "3,7",
+    "scrollLine": "0"
+}

.Rproj.user/3992D3D4/sources/prop/D1765424

@@ -0,0 +1,6 @@
+{
+    "tempName": "Untitled1",
+    "cursorPosition": "75,17",
+    "scrollLine": "48",
+    "last_setup_crc32": "9F8C73CAbb338d19"
+}

.Rproj.user/3992D3D4/sources/prop/FCD88EE5

@@ -1,5 +1,5 @@
 {
     "tempName": "Untitled1",
-    "cursorPosition": "23,27",
-    "scrollLine": "1"
+    "cursorPosition": "23,8",
+    "scrollLine": "0"
 }

.Rproj.user/3992D3D4/sources/prop/INDEX

@@ -25,3 +25,6 @@ H%3A%2FUbuntuRv2%2FIBGS%2FIBGS%2Fman%2Fac.ratio.Rd="EF1AA608"
 H%3A%2FUbuntuRv2%2FIBGS%2FIBGS%2Fman%2FexBIC.Rd="BAAC6886"
 H%3A%2FUbuntuRv2%2FIBGS%2Ftest.R="FCD88EE5"
 H%3A%2FUbuntuRv2%2FIBGS%2Ftest.Rout="43B20C7B"
+H%3A%2FUbuntuRv2%2FTCGS%2FIBGS.Rmd="D1765424"
+H%3A%2FUbuntuRv2%2FTCGS%2FIBGS.tex="4F1C9B07"
+H%3A%2FUbuntuRv2%2FTCGS%2Ftcg_ar.R="AA8888B6"

.Rproj.user/3992D3D4/sources/s-9F8C73CA/254CACE3

@@ -0,0 +1,25 @@
+{
+    "id": "254CACE3",
+    "path": "H:/UbuntuRv2/IBGS/IBGS/R/blockgibbs.R",
+    "project_path": "R/blockgibbs.R",
+    "type": "r_source",
+    "hash": "0",
+    "contents": "",
+    "dirty": false,
+    "created": 1637272959110.0,
+    "source_on_save": false,
+    "relative_order": 6,
+    "properties": {
+        "tempName": "Untitled2",
+        "cursorPosition": "46,37",
+        "scrollLine": "25"
+    },
+    "folds": "",
+    "lastKnownWriteTime": 1637239897,
+    "encoding": "UTF-8",
+    "collab_server": "",
+    "source_window": "",
+    "last_content_update": 1637239897,
+    "read_only": false,
+    "read_only_alternatives": []
+}

.Rproj.user/3992D3D4/sources/s-9F8C73CA/254CACE3-contents

@@ -0,0 +1,49 @@
+#' The iterated block Gibbs sampler algorithm
+#'
+#' @param y        the response variable
+#' @param x        the predictors
+#' @param H        the number of predictors in small groups, default is 30
+#' @param kapp     the number of selected predictors in first step, default is 20
+#' @param tau      the threshold to select the important predictors in second step, default is 0.9
+#' @param len      the half number of generated samples, default is 200
+#' @param k        the tuning parameter, default is 1
+#' @param gamma    the parameter for extended BIC, default is 0.5
+#' @param info     the selected model selection criterion from AIC, AICc, BIC and exBIC
+#' @param family   the type of model from linear, logistic, poisson
+#' @param n.iter   the number of iterations
+#' @param n.models the number of top selected models
+#'
+#' @return a list contains a summary of final result
+#' @export
+
+BlockGibbsSampler <- function(y, x, n.iter = 3, n.models =10, H = 30, kapp = 20,
+                              tau = 0.9, len = 200, k = 1, gamma = 0.5,
+                              info = c("AIC", "BIC", "AICc", "exBIC"),
+                              family = c("gaussian","poisson", "binomial")){
+  p <- dim(x)[2]
+  x.predictors <- colnames(x)
+  colnames(x)  <- 1:p
+
+  x1 <- x
+  x2 <- vector()
+
+  if(n.iter < 2){
+    result <- BlockGibbsSampler.step3(y, x1, x2, n.models, x.predictors, H,
+                                      kapp, tau, len, k, gamma, p, info, family)
+  }
+  else{
+    j <- 1
+    while(j < n.iter){
+      v.select <- BlockGibbsSampler.step2(y, x1, x2, H, kapp, tau, len, k,
+                                          gamma, p, info, family)
+      x1 <- x[, -v.select]
+      x2 <- x[, v.select]
+
+      j <- j+1
+    }
+    result <- BlockGibbsSampler.step3(y, x1, x2, n.models, x.predictors, H,
+                                      kapp, tau, len, k, gamma, p, info, family)
+  }
+  result$x.predictors <- x.predictors
+  return(result)
+}

.Rproj.user/3992D3D4/sources/s-9F8C73CA/462FD7FF

@@ -0,0 +1,26 @@
+{
+    "id": "462FD7FF",
+    "path": "H:/UbuntuRv2/TCGS/IBGS.Rmd",
+    "project_path": null,
+    "type": "r_markdown",
+    "hash": "2363041877",
+    "contents": "",
+    "dirty": false,
+    "created": 1637269087029.0,
+    "source_on_save": false,
+    "relative_order": 5,
+    "properties": {
+        "tempName": "Untitled1",
+        "cursorPosition": "75,17",
+        "scrollLine": "48",
+        "last_setup_crc32": "9F8C73CAbb338d19"
+    },
+    "folds": "",
+    "lastKnownWriteTime": 1637272918,
+    "encoding": "UTF-8",
+    "collab_server": "",
+    "source_window": "",
+    "last_content_update": 1637272918280,
+    "read_only": false,
+    "read_only_alternatives": []
+}

.Rproj.user/3992D3D4/sources/s-9F8C73CA/462FD7FF-contents

@@ -0,0 +1,78 @@
+---
+title: "An introduction to IBGS"
+author: "Lizhong Chen"
+date: "11/19/2021"
+output:
+  pdf_document: default
+  html_document:
+    df_print: paged
+---
+
+```{r setup, include=FALSE}
+knitr::opts_chunk$set(echo = TRUE)
+```
+
+## Introduction
+
+Here is a brief introduction to the R package "IBGS". The iterated block Gibbs sampler is a powerful variable selection method in high dimensions. It is important to identify the potential predictors which has significant influence on the response variable.
+
+The "IBGS" package works for the generalized linear models including linear, logistic and poisson models and uses AIC, AICc, BIC, and exBIC as the model selection criterion.
+
+There are two main goals for "IBGS" package:
+
+* Identify the important predictors with the marginal probability
+
+* Make a prediction based on the best selected model or model averaging using top selected models
+
+## Install
+
+To install and load the "IBGS" package, use this command
+```{r, echo=FALSE}
+cat("install.packages(\"IBGS_0.1.0.tar.gz\", repos = NULL)\nlibrary(IBGS)")
+```
+
+The "IBGS" package requires "doParallel" package. Please ensure that "doParallel" package is installed and loaded, then set up the register-cores.
+```{r, echo=FALSE}
+cat("library(doParallel)\nregisterDoParallel(ncores)\n")
+```
+where ncores is the number of CPU cores used in the calculation.
+
+## Main function
+
+The main function is 
+```{r, echo=FALSE}
+cat("BlockGibbsSampler <- function(y, x, n.iter = 3, n.models =10, H = 30, kapp = 20,
+                              tau = 0.9, len = 200, k = 1, gamma = 0.5,
+                              info = c(\"AIC\", \"BIC\", \"AICc\", \"exBIC\"),
+                              family = c(\"gaussian\",\"poisson\", \"binomial\"))")
+```
+For details, please use help document.
+```{r, echo=FALSE}
+cat("?BlockGibbsSampler")
+```
+
+## Example
+
+Here is an example to show how "IBGS" works. We use the tropical cyclone genesis data.
+```{r, echo=FALSE}
+load("tcg_ar.RData")
+```
+
+It has 1856 samples and 149 predictors.
+```{r, collapse=TRUE}
+dim(x)
+```
+We divide the data into train and test set and use the train set ($\approx 80\%$ of data) to run "IBGS" algorithm. Then we use the test set ($\approx 20\%$ of data) to verify the results.
+
+The simple following command is used
+```{r}
+cat("model <- BlockGibbsSampler(y_train, x_train, info = \"exBIC\", family = \"binomial\")")
+```
+We only need to specify the response, predictors, the model selection criterion and the model type. 
+### Variable ranking
+
+We can plot the marginal probabilities for those important predictors using plots.Gibbs command.
+```{r, collapse=TRUE}
+plots.Gibbs(model)
+```
+