ctrl C + ctrl V = critical error

assignment7.R

@@ -0,0 +1,86 @@
+library(tidyverse)
+library(caTools)
+library(rpart)
+library(party)
+library(rpart.plot)
+library(randomForest)
+library(zoo)
+
+
+testscores <- online_data
+colnames(testscores)
+# [1] "id"                  "post.test.score"     "pre.test.score"      "messages"            "forum.posts"        
+# [6] "av.assignment.score"  "level.up"     
+
+library(ggplot2)
+
+ggplot(testscores, aes(x = post.test.score, y = pre.test.score, color = messages)) +
+  geom_point()
+
+# more messages higher post.test.score
+
+prediction <- rpart(level.up ~ messages + forum.posts + pre.test.score, 
+                      method = 'class', data = testscores)
+prediction
+prp(prediction)
+
+testscores$pred <- predict(prediction, type = "prob")[,2]
+
+library(ROCR)
+
+pred.detail <- prediction(testscores$pred, testscores$level.up) 
+plot(performance(pred.detail, "tpr", "fpr"))
+abline(0, 1, lty = 2)
+
+unlist(slot(performance(Pred2,"auc"), "y.values"))
+
+##
+
+
+testscores$threshold.pred1 <- ifelse(testscores$pred >= 0.4, 1, 0)
+
+data$TPOS.model1 <- ifelse(testscores$level.up == "1" & testscores$threshold.pred1 == "1", 
+                              1, 0)
+testscores$FPOS.model1 <- ifelse(testscores$level.up == "0" & testscores$threshold.pred1 == "1",
+                               1, 0)
+testscores$FNEG.model1 <- ifelse(testscores$level.up == "1" & testscores$threshold.pred1 == "0",
+                               1, 0)
+
+#Now generate three diagnostics:
+
+testscores$accuracy.model1 <- mean(ifelse(testscores$level.up == testscores$threshold.pred1, 1, 0))
+
+testscores$precision.model1 <- sum(testscores$TPOS.model1)/(sum(data$TPOS.model1) 
+                                                     + sum(data$FPOS.model1))
+
+testscores$recall.model1 <- sum(testscores$TPOS.model1)/(sum(testscores$TPOS.model1) 
+                                                  + sum(testscores$FNEG.model1))
+
+#Finally, calculate Kappa for your model according to
+#First generate the table of comparisons
+
+table1 <- table(testscores$level.up, testscores$threshold.pred1)
+table1 <- table(testscores$level.up, testscores$threshold.pred1)
+
+#Convert to matrix
+matrix1 <- as.matrix(table1)
+kappa(matrix1, exact = TRUE)/kappa(matrix1)
+
+#Now choose a different threshold value and repeat these diagnostics. What conclusions can you draw about your two thresholds?
+testscores$threshold.pred2 <- ifelse(testscores$pred >= 0.8, 1, 0)
+testscores$TPOS.model2 <- ifelse(testscores$level.up == "1" & testscores$threshold.pred2 == "1", 
+                              1, 0)
+testscores$FPOS.model2 <- ifelse(testscores$level.up == "0" & testscores$threshold.pred2 == "1",
+                               1, 0)
+testscores$FNEG.model2 <- ifelse(testscores$level.up == "1" & testscores$threshold.pred2 == "0",
+                               1, 0)
+testscores$accuracy.model2 <- mean(ifelse(testscores$level.up == testscores$threshold.pred2, 1, 0))
+testscores$precision.model2 <- sum(testscores$TPOS.model2)/(sum(testscores$TPOS.model2) 
+                                                   + sum(testscores$FPOS.model2))
+
+testscores$recall.model2 <- sum(testscores$TPOS.model2)/(sum(testscores$TPOS.model2) 
+                                                + sum(testscores$FNEG.model2))
+
+table2 <- table(testscores$level.up, testscores$threshold.pred2)
+matrix2 <- as.matrix(table2)
+kappa(matrix2, exact = TRUE)/kappa(matrix2)