---
title: "PART 2"
author: "Eric Womack"
date: "5/3/2020"
output: word_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

############# PART 2 ##############

######Creating a Boosted Regression Tree Model###################

```{r}
# 1. Import the Data
library(dismo)

data <- read.csv("data.csv")
```

```{r}
#2. Create the Model
fish.step<- gbm.step(data=data, gbm.x = 10:16, gbm.y = 9, family = "poisson", tree.complexity = 5, learning.rate = 0.003, bag.fraction = 0.6)

```

```{r}
#5. Examine the Residuals Plot
plot(fish.step$fitted, fish.step$residuals)
abline(0, 0) 
```


```{r}
#6. Display relative influence
summary(fish.step)
```

```{r}
#7. Plotting fitted functions
 gbm.plot(fish.step,n.plots=4,common.scale = TRUE, write.title = FALSE, plot.layout = c(2,4))
```


############### Creating a Multiple Regression Tree Model ######################

```{r}
#1. Load the Required Packages
library(relaimpo)
library(car)
```

```{r}
#2. Create the multiple regression model
fish.glm<- glm(fish_rich ~ Reef.complexity + No.tall.corals + No.corals + Sponge.max.height + Octocoral.max.height + Slope.angle + Country, data = data)
```

```{r}
#3. Examine the diagnostic plots
par(mfrow=c(2,2))
plot(fish.glm)
```

```{r}
#4. Check for multicollinearity
vif(fish.glm)
```


```{r}
#5. Calculate the deviance explained
summary(fish.glm)
```




```{r}
#6. Calculate relative influence
calc.relimp(fish.glm, type = "lmg", rela = TRUE)
```

################### Comparing the Predictive Power of our Models ##########################

```{r}
#2. Root mean squared error
RMSE <- function(error) { sqrt(mean(error^2)) }
RMSE(fish.glm$residuals)
RMSE(fish.step$residuals)

```