Create Zeiden_HW2.r

Zeiden_HW2.r

@@ -0,0 +1,119 @@
+library(AppliedPredictiveModeling)
+library(caret)
+data(permeability)
+library("pls")
+#
+# (B)
+#
+
+ncol(fingerprints)
+
+nzv_cols <- nearZeroVar(fingerprints)
+if(length(nzv_cols) > 0) fingerprints <- fingerprints[, -nzv_cols]
+ncol(fingerprints) ## 388 predictors remain from the original 1107 predictors (719 removed)
+
+#
+# (C)
+#
+
+
+## 75% of the sample size
+smp_size <- floor(0.75 * nrow(fingerprints))
+
+## set the seed to make your partition reproductible
+set.seed(123)
+train_ind <- sample(seq_len(nrow(fingerprints)), size = smp_size)
+##permeability <- data.frame(permeability)
+train_ind_perm <- sample(seq_len(nrow(fingerprints)), size = smp_size)
+
+train_perm <- permeability[train_ind_perm,]
+test_perm <- permeability[-train_ind_perm,]
+
+train_fp <- fingerprints[train_ind, ]
+test_fp <- fingerprints[-train_ind, ]
+
+
+## PLS model
+
+ctrl <- trainControl(method = "cv", number = 10)
+
+df_train_fp <- data.frame(train_fp)
+
+#####
+
+plsFit = plsr(train_perm ~ . , data = df_train_fp, ncomp = 20)
+
+## try on test data (316)
+plsPred = predict(plsFit, df_train_fp, ncomp = 20)
+
+plsValue = data.frame(obs = train_perm, pred = plsPred[,,1])
+
+defaultSummary(plsValue) 
+
+
+plsModel <- train(x=train_fp, y = train_perm,
+                  method="pls",
+                  tuneLength=20,
+                  trControl = ctrl,
+                  preProc= c("center", "scale"))
+
+
+## Again, try 10 folds cross validation
+set.seed(123)
+pls <- train(x = df_train_fp, y = train_perm, method = "pls", tuneGrid = expand.grid(ncomp = 20), trControl = ctrl)
+pls
+
+testpls <- data.frame(obs = train_perm, pred = predict(pls, df_train_fp))
+
+defaultSummary(testpls)
+
+
+
+## Model tuning
+## Run PLS and PCR on solubility data and compare results with 10 folds CV
+
+set.seed(123)
+plsTune <- train(x = df_train_fp, y = train_perm, method = "pls", tuneGrid = expand.grid(ncomp = 1:20), trControl = ctrl)
+plsTune
+
+testResultsPLS <- data.frame(obs = train_perm, pred = predict(plsTune, df_train_fp))
+
+defaultSummary(testResultsPLS)
+
+set.seed(123)
+pcrTune <- train(x = df_train_fp, y = train_perm, method = "pcr", tuneGrid = expand.grid(ncomp = 1:35), trControl = ctrl)
+pcrTune                  
+
+plsResamples <- plsTune$results
+plsResamples$Model <- "PLS"
+pcrResamples <- pcrTune$results
+pcrResamples$Model <- "PCR"
+plsPlotData <- rbind(plsResamples, pcrResamples)
+
+xyplot(RMSE ~ ncomp,
+       data = plsPlotData,
+       #aspect = 1,
+       xlab = "# Components",
+       ylab = "RMSE (Cross-Validation)",       
+       col = c("blue","red"),
+       auto.key = list(columns = 2),
+       groups = Model,
+       type = c("o", "g"))
+
+# variable importance - PLS  Fig. 6.14
+plsImp <- varImp(plsTune, scale = FALSE)
+plot(plsImp, top = 25, scales = list(y = list(cex = .95)))
+
+# variable importance - PCR
+pcrImp <- varImp(pcrTune, scale = FALSE)
+plot(pcrImp, top = 25, scales = list(y = list(cex = .95)))
+
+###################
+
+
+# exercise 7.4
+
+
+
+
+