Modeling: Universal Framework

class: center, middle, inverse, title-slide

.title[
# <img src="figures/bat-cartoon.png" /> Modeling: Universal Framework
]
.author[
### <font size="5"> Son Nguyen </font>
]

---

<style>

.remark-slide-content {
  background-color: #FFFFFF;
  border-top: 80px solid #F9C389;
  font-size: 17px;
  font-weight: 300;
  line-height: 1.5;
  padding: 1em 2em 1em 2em
}

.inverse {
  background-color: #696767;
  border-top: 80px solid #696767;
  text-shadow: none;
  background-image: url(https://github.com/goodekat/presentations/blob/master/2019-isugg-gganimate-spooky/figures/spider.png?raw=true);
	background-position: 50% 75%;
  background-size: 150px;
}

.your-turn{
  background-color: #8C7E95;
  border-top: 80px solid #F9C389;
  text-shadow: none;
  background-image: url(https://github.com/goodekat/presentations/blob/master/2019-isugg-gganimate-spooky/figures/spider.png?raw=true);
	background-position: 95% 90%;
  background-size: 75px;
}

.title-slide {
  background-color: #F9C389;
  border-top: 80px solid #F9C389;
  background-image: none;
}

.title-slide > h1  {
  color: #111111;
  font-size: 40px;
  text-shadow: none;
  font-weight: 400;
  text-align: left;
  margin-left: 15px;
  padding-top: 80px;
}
.title-slide > h2  {
  margin-top: -25px;
  padding-bottom: -20px;
  color: #111111;
  text-shadow: none;
  font-weight: 300;
  font-size: 35px;
  text-align: left;
  margin-left: 15px;
}
.title-slide > h3  {
  color: #111111;
  text-shadow: none;
  font-weight: 300;
  font-size: 25px;
  text-align: left;
  margin-left: 15px;
  margin-bottom: -30px;
}

</style>

# Other packages for Trees and Forests

- Decision Tree can be implemented by `C50`, `RWeka`, `party`... packages

- Random Forest can be implemented by `ranger`, `foreach`,`e1071`... packages

---
# Consistency Issue in R

- Modeling Packages are made by different people

---
# Consistency Issue in R

- Modeling Packages are made by different people

- They have slightly different interfaces

---
# Consistency Issue in R

- Modeling Packages are made by different people

- They have slightly different interfaces

- Trying to keep everything in line can be frustrating

---
# Unified Interface

- A package to unify all the model interfaces is needed

- This is called a `wrapper`

- There are several wrappers for machine learning in R:

- caret
  
  - mlr3
  
  - tidymodels

---
# CARET

- Short for Classification And REgression Training

- Attempt to streamline the process for creating predictive models.

- Created by Max Kuhn

---
# Data Preparation

```r
library(tidyverse)
library(caret)
library(tidyverse)
df = read_csv("https://bryantstats.github.io/math421/data/titanic.csv")

# Remove some columns
df <- df %>% select(-PassengerId, -Ticket, -Name, -Cabin)

# Set the target variable
df <- df %>% rename(target=Survived)

# Correct variables' types
df <- df %>% 
  mutate(target = as.factor(target),
         Pclass = as.factor(Pclass),
         )

# Handle missing values
df$Age[is.na(df$Age)] = mean(df$Age, na.rm = TRUE)

df = drop_na(df)

splitIndex <- createDataPartition(df$target, p = .70, 
                                  list = FALSE)
df_train <- df[ splitIndex,]
df_test <- df[-splitIndex,]
```

---
# Create a Decision Tree with Caret

```r
model1 <- train(target~., data=df_train, 
*               method = "rpart2",
                maxdepth=3)

pred <- predict(model1, df_test)

cm <- confusionMatrix(data = pred, reference = df_test$target, positive = "1")

cm$overall[1]
```

```
##  Accuracy 
## 0.8157895
```

---
# Create a Random Forest with Caret

```r
model2 <- train(target~., data=df_train, 
*               method = "rf",
                ntree = 1000)

pred <- predict(model2, df_test)

cm <- confusionMatrix(data = pred, reference = df_test$target, positive = "1")

cm$overall[1]
```

```
##  Accuracy 
## 0.8383459
```

---
# Variable Importance

```r
# Tree
varImp(model1)
```

```
## rpart2 variable importance
## 
##            Overall
## Sexmale   100.0000
## Fare       80.6302
## Pclass3    74.9213
## Age        46.8959
## Parch      17.6597
## SibSp      15.4525
## EmbarkedS   9.4607
## Pclass2     0.9702
## EmbarkedQ   0.0000
```

---
# Variable Importance

```r
# Forest
varImp(model2)
```

```
## rf variable importance
## 
##           Overall
## Sexmale   100.000
## Fare       67.184
## Age        48.652
## Pclass3    29.739
## SibSp      13.615
## Parch      10.680
## Pclass2     4.881
## EmbarkedS   4.324
## EmbarkedQ   0.000
```

---
# Plot Variable Importance

```r
# Tree
*plot(varImp(model1))
```

---
# Plot Variable Importance

```r
# Forest
*plot(varImp(model2))
```

---
# Other Models

Find all the available models at

[Available Models](https://topepo.github.io/caret/available-models.html)