Update Zeiden_HW2.r

Zeiden_HW2.r

@@ -6,8 +6,8 @@ library("pls")
 # (B)
 #
 
+## here we are filtering out columns with low frequencies 
 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)
@@ -17,30 +17,37 @@ ncol(fingerprints) ## 388 predictors remain from the original 1107 predictors (7
 #
 
 
+## split the data into a test and training data set
+## preprocess the data and tune a PLS model
+
 ## 75% of the sample size
+## this is to split the training and test sets up from the data set
 smp_size <- floor(0.75 * nrow(fingerprints))
 
 ## set the seed to make your partition reproductible
 set.seed(123)
+
+## breaking up the data into training and test data sets for the fingerprint and
+## permeability data set
+
 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)
-
+## permeability test and training data sets
 train_perm <- permeability[train_ind_perm,]
 test_perm <- permeability[-train_ind_perm,]
-
+## fingerprint test and training data set
 train_fp <- fingerprints[train_ind, ]
 test_fp <- fingerprints[-train_ind, ]
 
 
 ## PLS model
 
+## 10 folds cross validation
+
 ctrl <- trainControl(method = "cv", number = 10)
 
+## need to turn this into a df for the plsFit 
 df_train_fp <- data.frame(train_fp)
-
-#####
-
 plsFit = plsr(train_perm ~ . , data = df_train_fp, ncomp = 20)
 
 ## try on test data (316)
@@ -49,7 +56,7 @@ plsPred = predict(plsFit, df_train_fp, ncomp = 20)
 plsValue = data.frame(obs = train_perm, pred = plsPred[,,1])
 
 defaultSummary(plsValue) 
-
+## RMSE 9.5 Rsquared 0.614 MAE 6.60
 
 plsModel <- train(x=train_fp, y = train_perm,
                   method="pls",
@@ -80,6 +87,8 @@ testResultsPLS <- data.frame(obs = train_perm, pred = predict(plsTune, df_train_
 
 defaultSummary(testResultsPLS)
 
+## RMSE of 14.54 and Rsquared of 0.09 MAE 11.09
+
 set.seed(123)
 pcrTune <- train(x = df_train_fp, y = train_perm, method = "pcr", tuneGrid = expand.grid(ncomp = 1:35), trControl = ctrl)
 pcrTune                  
@@ -112,8 +121,3 @@ plot(pcrImp, top = 25, scales = list(y = list(cex = .95)))
 
 
 # exercise 7.4
-
-
-
-
-