ProjectChallenge.Rmd

---
title: "Project Challenge"
output: pdf_document
author: "Franssila Fanni, Suihkonen Sini, Savolainen Outi"
date: "December 9, 2021"
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```
\newpage
Used libraries
```{r, message=F}
library(ggfortify)
library(randomForest)
library(caret)
library(ggplot2)
```

Read data

```{r}
set.seed(42)

npf <- read.csv("npf_train.csv")
npf_test <- read.csv("npf_test.csv")

rownames(npf) <- npf[,"date"]
rownames(npf_test) <- npf_test[,"date"]
vars <- colnames(npf)[sapply(colnames(npf),
                             function(s) nchar(s)>5 && substr(s,nchar(s)-4,nchar(s))==".mean")]

npf <- npf[,c(vars,"class4")]
npf_test <- npf_test[,c(vars,"class4")]
## strip the trailing ".mean" to make the variable names prettier
colnames(npf)[1:length(vars)] <- sapply(colnames(npf)[1:length(vars)],
                                        function(s) substr(s,1,nchar(s)-5))

colnames(npf_test)[1:length(vars)] <- sapply(colnames(npf_test)[1:length(vars)],
                                        function(s) substr(s,1,nchar(s)-5))
vars <- colnames(npf)[1:length(vars)]
```



```{r}
npf.pcA2 <- prcomp(scale(npf[,vars]))
npf.pcA2.withtest <- prcomp(scale(rbind(npf[,vars],npf_test[,vars])))
autoplot(npf.pcA2, data=npf, colour="class4", loadings=T, loadings.label=T)
```


Binary accuracy on random forest classifier with different sized training/validation sets:
## Binary accuracy (class 2)


## Accuracy of the estimate of accuracy

```{r}

```

## Perplexity

```{r}

```

## Multi-class accuracy (class4)

Multiclass accuracy on the random forest classifier:

```{r}
set.seed(42)
# Calculates the accuracy to each class and the total accuracy
npf.pcA2 <- prcomp(npf[,vars], center=T, scale=T)

idx <- sample.int(nrow(npf),229)
training_set <- npf[ idx,]
validation_set <- npf[-idx,]
train.pc <- data.frame(npf.pcA2$x[idx,1:14])
train.pc$class4 <- npf[idx,]$class4
validate.pc <- data.frame(npf.pcA2$x[-idx,1:14])
validate.pc$class4 <- npf[-idx,]$class4
ctrl <- trainControl(method = "repeatedcv",
                     number = 10,
                     repeats = 10,
                     classProbs = TRUE,
                     selectionFunction = "best"
                     )
rFClass4 <- train(factor(class4) ~ .,
                    method="rf",
                    data=train.pc,
                    trControl=ctrl)
probs4 <- predict(rFClass4, newdata = validate.pc, type = "prob")
pred <- predict(rFClass4, newdata = validate.pc)
confusionMatrix(pred, factor(validate.pc$class4))

```


```{r}
train.pc <- data.frame(npf.pcA2.withtest$x[1:458,1:14])
train.pc$class4 <- npf$class4

test.pc <- data.frame(npf.pcA2.withtest$x[459:(458+nrow(npf_test)),1:14])
test.pc$class4 <- npf_test$class4
ctrl <- trainControl(method = "repeatedcv",
                     number = 10,
                     repeats = 10,
                     classProbs = TRUE,
                     savePredictions = "final")
rFClass4 <- train(factor(class4) ~ .,
                    method="rf",
                    data=train.pc,
                    trControl=ctrl)

ggplot(rFClass4)
pred_test4 <- predict(rFClass4, newdata = test.pc)

probs_test4 <- predict(rFClass4, newdata = test.pc, type = "prob")

confusionMatrix(pred_test4, factor(test.pc$class4))
length(which(pred_test4=="nonevent"))
length(which(pred_test4=="Ia"))
length(which(pred_test4=="Ib"))
length(which(pred_test4=="II"))
```
```{r}
trellis.par.set(caretTheme())
densityplot(rFClass4, pch = "|")
```

```{r}
# Code to produce the answer-csv
# The first column "class4" in the answers.csv file is our 
# prediction for the day, where class4 is Ia, Ib, II, or nonevent.
# The second column "p" is our prediction for probability Pr(class2=event)


# Creates the csv and adds the first line
# Change the string here to our guess of the accuracy
write.table(0.75,
             file="./answers.csv",
             append = F,
             sep=',',
             row.names=F,
             col.names=F)

# Write column names to the file
write.table(data.frame("class4","p"),
            file="./answers.csv",
            append = T,
            sep=',',
            row.names=F,
            col.names=F)

# testing testing
#setwd()
#probs_test4 <- data.frame(c(0.1,0.2,0.5,0.1,0.5),c(0.6,0.05,0.1,0.2,0.05),c(0.1,0.7,0.2,0.2,0.15),c(0.2,0.05,0.2,0.5,0.3))
#colnames(probs_test4)<- c("Ia","Ib","II","nonevent")

# Assume the class probabilities for each row are in probs_test4
classes_test4 <- colnames(probs_test4)[max.col(probs_test4, ties.method = "first")]

# Write the class predictions and probabilities
write.table(data.frame(classes_test4, (probs_test4$Ia+probs_test4$Ib+probs_test4$II)),
            file="./answers.csv",
            append = T,
            sep=',',
            row.names=F,
            col.names=F)

```