GradePredictionMain.Rmd

---
title: "Predict Student's Final Grade"
output:
  html_document:
    df_print: paged
---

<h2>
Created for the <a href="https://sites.google.com/ncsu.edu/csedm-dc-2021/home">2nd CSEDM Data Challenge</a>
</h2>

<h3>
  Installing libraries
</h3>
<p>This step is only necessary the first time the file is run on a machine.</p>
```{r}
# install.packages("tidyverse")
# install.packages("BBmisc")
# install.packages("caret")
# install.packages("ggvis")
#install.packages("randomForest")
#install.packages("caTools")
```

<h3> Importing libraries </h3>
```{r, message=FALSE}
library(tidyverse)
library(BBmisc)
library(caret)
library(ggvis)
library(randomForest)
library(caTools)
```

<h3>Loading All Data</h3>
<p>personal_path will need to be changed to match local machine filepath</p>
```{r, message=FALSE}
personal_path <- "/Users/katiepell/school/cs496"
base_dir <- "cs496-csedm-main"
train_path <- "S19_Release_6_28_21.zip/Train" # Path to training data
test_path <- "S19_Release_6_28_21.zip/Test" # Path to testing data
subject_path <- "Data/LinkTables/Subject.csv"
early_path <- "early.csv"
late_path <- "late.csv"

train_id_grade_path <- paste(personal_path, base_dir, train_path, subject_path, sep="/")
test_id_grade_path <- paste(personal_path, base_dir, test_path, subject_path, sep="/")

train_id_grade_data <- read_csv(train_id_grade_path)
test_id_grade_data <- read_csv(test_id_grade_path)

train_early_path <- paste(personal_path, base_dir, train_path, early_path, sep = "/")
test_early_path <- paste(personal_path, base_dir, test_path, early_path, sep = "/")

train_early_data <- read_csv(train_early_path)
test_early_data <- read_csv(test_early_path)

train_late_path <- paste(personal_path, base_dir, train_path, late_path, sep = "/")
test_late_path <- paste(personal_path, base_dir, test_path, late_path, sep = "/")

train_late_data <- read_csv(train_late_path)
test_late_data <- read_csv(test_late_path)

```

<p>Function to find each student's average number of attempts</p>
```{r, message=FALSE}
find_avg_attempts = function(student_instance, early_dataset){
  
  # This is get the all data related to the current student
  first_half_data <- filter(early_dataset, SubjectID == student_instance)
  
  # find avg number of attempts
  avg_attempts <- median(first_half_data$Attempts)
}
```

```{r, message=FALSE}
# Function to find each student's percent correct eventually
find_percent_CorrectEventually = function(student_instance, early_dataset){
  
  # This is get the all data related to the current student
  first_half_data <- filter(early_dataset, SubjectID == student_instance) # All of this student's problem entries from the the early part of the semester
  
  # find percent of time student gets the right answer eventually
  PerCorrectEventually <- mean(first_half_data$CorrectEventually) # Average number correct for all of this student's problems that are attempted
}

```

<p>
Feature Extraction on the training data
</p>

```{r, message=FALSE}
# ***** I THINK THIS NEEDS TO BE "MERGED" WITH THE APPROPRIATE OTHER DATASET... *****
merged_data_train <- merge(train_id_grade_data, train_early_data, by= "SubjectID")

# Add a column with each student's average number of attempts
agg_train_id_grade_data <- merged_data_train %>%
  rowwise() %>%
  mutate(avg_attempts = find_avg_attempts(SubjectID, train_early_data))

# Add a column with each student's percent correct eventually
agg_train_id_grade_data <- agg_train_id_grade_data %>%
  rowwise() %>%
  mutate(per_CorrectEventually = find_percent_CorrectEventually(SubjectID, train_early_data))
```

```{r, message=FALSE}
# Change names for simplicity
names(agg_train_id_grade_data)[names(agg_train_id_grade_data) == "X-Grade"] <- "grade"
names(agg_train_id_grade_data)[names(agg_train_id_grade_data) == "SubjectID"] <- "id"
names(agg_train_id_grade_data)[names(agg_train_id_grade_data) == "AssignmentID"] <- "assignment_id"
names(agg_train_id_grade_data)[names(agg_train_id_grade_data) == "ProblemID"] <- "problem_id"
```

```{r, message=FALSE}
# Format test data similar to above format of training data
merged_data_test <- merge(test_id_grade_data, test_early_data, by= "SubjectID")

agg_test_id_grade_data <- merged_data_test %>% # Add the attempt column to the test data
  rowwise() %>%
  mutate(avg_attempts = find_avg_attempts(SubjectID, test_early_data))

agg_test_id_grade_data <- agg_test_id_grade_data %>% # Add the percentCorrect column to the test data
  rowwise() %>%
  mutate(per_CorrectEventually = find_percent_CorrectEventually(SubjectID, test_early_data))
```

```{r, message=FALSE}
# Change names for simplicity
names(agg_test_id_grade_data)[names(agg_test_id_grade_data) == "SubjectID"] <- "id"
names(agg_test_id_grade_data)[names(agg_test_id_grade_data) == "AssignmentID"] <- "assignment_id"
names(agg_test_id_grade_data)[names(agg_test_id_grade_data) == "ProblemID"] <- "problem_id"
```

<p>Linear Models</p>
```{r, message=FALSE}

# Avg_attempts vs. final grade
model1 <- lm(avg_attempts~grade, data=agg_train_id_grade_data)

#Percent eventually correct vs. final grade
model2 <- lm(per_CorrectEventually~grade, data = agg_train_id_grade_data)

# Percent eventually correct AND average attempts vs. final grade
model3 <- lm(grade~avg_attempts+per_CorrectEventually, data=agg_train_id_grade_data)

attempts_grade_plot <- agg_train_id_grade_data %>% 
  ggplot(aes(x = avg_attempts, y = grade)) +
  geom_point() + geom_smooth(method="lm")
  labs(x = "Average attempts", y = "Final grade")
  
percent_grade_plot <- agg_train_id_grade_data %>% 
  ggplot(aes(x = per_CorrectEventually, y = grade)) +
  geom_point() + geom_smooth(method="lm")
  labs(x = "Percent correct eventually", y = "Final grade")
  
s1 <- summary(model1)
s2 <- summary(model2)
s3 <- summary(model3)
s1$r.squared # Correlation between average attempts and final grade
s2$r.squared # Correlation between percent eventually correct and final grade
s3$r.squared # Correlation between both factors and final grade

```


```{r, message=FALSE}
# these scatter plots are colored to differentiate problem/assignment number
assignment_plot1 <- agg_train_id_grade_data %>% ggvis(~avg_attempts, ~grade, fill = ~assignment_id) %>% layer_points()
assignment_plot1

assignment_plot2 <- agg_train_id_grade_data %>% ggvis(~per_CorrectEventually, ~grade, fill = ~assignment_id) %>% layer_points()
assignment_plot2


problem_plot1 <- agg_train_id_grade_data %>% ggvis(~avg_attempts, ~grade, fill = ~problem_id) %>% layer_points()
problem_plot1

problem_plot2 <- agg_train_id_grade_data %>% ggvis(~per_CorrectEventually, ~grade, fill = ~problem_id) %>% layer_points()
problem_plot2
```

<p>Random Forest Regression Models</p>

```{r, message=FALSE} 
# Create a random forest with the training data
rf_train_data <- agg_train_id_grade_data
rf_train_data$assignment_id = as.factor(rf_train_data$assignment_id) # Create assignment ID as a factor
rf_train_data$problem_id = as.factor(rf_train_data$problem_id) # Create problem ID as a factor
# summary(rf_train_data)
```

```{r, message=FALSE}

# Create a random forest with the testing data
rf_test_data <- agg_test_id_grade_data
rf_test_data$assignment_id = as.factor(rf_test_data$assignment_id)
rf_test_data$problem_id = as.factor(rf_test_data$problem_id)
# summary(rf_test_data)
```

```{r, message=FALSE}
# creating a random forest
head(rf_train_data)
sapply(rf_train_data, class)
summary(rf_train_data)
```
```{r, message=FALSE}
# Perform random forest regression on training data
rf_regression <- randomForest(grade ~ ., data = rf_train_data, ntree=500, mtry = 8, importance = TRUE, na.action = na.omit)

print(rf_regression)
# plot(rf_regression)

```

```{r, message=FALSE}
pred_rf <- predict(rf_regression, rf_test_data)
summary(pred_rf)
```
<p>Writing predictions to a file</p>
```{r, message=FALSE}
# This creates a new file called prediction.csv and writes the predicted grades to it
write.csv(pred_rf, "prediction.csv", row.names=FALSE)
# currently the column is titled 'x' and I can't figure out how to change that

```

```{r, message = FALSE}
end_data <- rf_test_data[,c('id')]
end_data['grade'] = pred_rf

end_data$Mean <- with(end_data, ave(end_data$grade, id, FUN=mean))
end_data <- subset(end_data, select = -c(grade))

unique_values <- end_data[!duplicated(end_data$id), ]
names(unique_values)[names(unique_values) == "id"] <- "SubjectID"
names(unique_values)[names(unique_values) == "Mean"] <- "X-Grade"
```
}
```{r}
# merge predictions with student id number and write to prediction.csv

write.csv(unique_values, 'predictions.csv', row.names = FALSE)
```


<h3>Troubleshooting</h3>
<p>
<b>Q:</b> I am getting an error that "assignment-id is not located in agg_train_id_grade_data"<br>
<b>A:</b> Confirm that the downloaded folder is named properly, "cs496-csedm-main".
</p>