Feature selection comparison using simulated data from an erdos-renyi graph structure

Schrick-Noah_Homework-6.R

@@ -14,6 +14,7 @@ source("Schrick-Noah_Ridge-LASSO-Regression.R")
 source("Schrick-Noah_Simulated-Data.R")
 bundled_data <- create_data()
 
+run_comparison <- function(bundled_data){
   ### LASSO
   unpen_beta <- unpen_coeff(bundled_data$train.X, bundled_data$train.y)
   lasso.df <- data.frame(att=c("intercept", colnames(bundled_data$train.X)), 
@@ -94,76 +95,13 @@ legend(x="topleft",
          lty=c(1,1,1,1,1),
          col=c("blue", "red", "green", "bisque4", "purple"),
          cex=1)
+}
 
-
+run_comparison(bundled_data)
 ## b. Repeat comparison using a graph with clusters
-if (!require("igraph")) install.packages("igraph")
+source("Schrick-Noah_graphs.R")
-library(igraph)
+bundled_graph_data <- sim_graph_data()
-if (!require("Matrix")) install.packages("Matrix")
+run_comparison(bundled_graph_data)
-library(Matrix) # bdiag 
-
-npc <-25 # nodes per cluster
-n_clust <- 4 # 4 clusters with 25 nodes each
-
-# no clusters
-g0 <- erdos.renyi.game(npc*n_clust, 0.2)
-plot(g0)
-
-matlist = list()
-for (i in 1:n_clust){ 
-  matlist[[i]] = get.adjacency(erdos.renyi.game(npc, 0.2))
-}
-
-# merge clusters into one matrix
-mat_clust <- bdiag(matlist) # create block-diagonal matrix
-
-## the following two things might not be necessary
-
-# check for loner nodes, connected to nothing, and join them to something
-k <- rowSums(mat_clust) 
-node_vector <- seq(1,npc*n_clust)
-for (i in node_vector){
-  if (k[i]==0){ # if k=0, connect to something random
-    j <- sample(node_vector[-i],1)
-    mat_clust[i,j] <- 1
-    mat_clust[j,i] <- 1
-  }
-}
-
-node_colors <- c(rep("red",npc), rep("green",npc), rep("blue",npc), rep("orange",npc))
-g1 <- graph_from_adjacency_matrix(mat_clust, mode="undirected", diag=F)
-plot(g1, vertex.color=node_colors)
-
-### Dataset with g1
-dataset.graph <- npdro::createSimulation2(num.samples=num.samples,
-                                    num.variables=num.variables,
-                                    pct.imbalance=0.5,
-                                    pct.signals=0.2,
-                                    main.bias=0.5,
-                                    interaction.bias=1,
-                                    hi.cor=0.95,
-                                    lo.cor=0.2,
-                                    mix.type="main-interactionScalefree",
-                                    label="class",
-                                    sim.type="mixed",
-                                    pct.mixed=0.5,
-                                    pct.train=0.5,
-                                    pct.holdout=0.5,
-                                    pct.validation=0,
-                                    plot.graph=F,
-                                    graph.structure = g1,
-                                    verbose=T)
-
-train.graph <- dataset.graph$train  #150x101
-test.graph <- dataset.graph$holdout 
-validation.graph <- dataset.graph$validation
-dataset.graph$signal.names
-colnames(train.graph)
-
-# separate the class vector from the predictor data matrix
-train.graph.X <- train.graph[, -which(colnames(train.graph) == "class")]
-train.graph.y <- train.graph[, "class"]
-train.graph.y.01 <- as.numeric(train.graph.y)-1
 
 ## c. Use npdro and igraph to create knn
 my.k <- 3  # larger k, fewer clusters

Schrick-Noah_graphs.R

@@ -0,0 +1,43 @@
+source("Schrick-Noah_Simulated-Data.R")
+
+if (!require("igraph")) install.packages("igraph")
+library(igraph)
+if (!require("Matrix")) install.packages("Matrix")
+library(Matrix) # bdiag 
+
+sim_graph_data <- function(){
+  npc <-25 # nodes per cluster
+  n_clust <- 4 # 4 clusters with 25 nodes each
+  
+  # no clusters
+  g0 <- erdos.renyi.game(npc*n_clust, 0.2)
+  plot(g0)
+  
+  matlist = list()
+  for (i in 1:n_clust){ 
+    matlist[[i]] = get.adjacency(erdos.renyi.game(npc, 0.2))
+  }
+  
+  # merge clusters into one matrix
+  mat_clust <- bdiag(matlist) # create block-diagonal matrix
+  
+  ## the following two things might not be necessary
+  # check for loner nodes, connected to nothing, and join them to something
+  k <- rowSums(mat_clust) 
+  node_vector <- seq(1,npc*n_clust)
+  for (i in node_vector){
+    if (k[i]==0){ # if k=0, connect to something random
+      j <- sample(node_vector[-i],1)
+      mat_clust[i,j] <- 1
+      mat_clust[j,i] <- 1
+    }
+  }
+  
+  node_colors <- c(rep("red",npc), rep("green",npc), rep("blue",npc), rep("orange",npc))
+  g1 <- graph_from_adjacency_matrix(mat_clust, mode="undirected", diag=F)
+  plot(g1, vertex.color=node_colors)
+  
+  ### Dataset with g1
+  bundled_graph_data <- create_data(graph.structure=g1)
+  return(bundled_graph_data)
+}