---
title: "Who Let the Dogs Out?"
author: Jonah Mergler
output: 
  flexdashboard::flex_dashboard:
    theme:
      version: 4
      bootswatch: minty5
    orientation: columns
    vertical_layout: fill
    source_code: embed
---

```{r Setup, include=FALSE}
library(flexdashboard)
pacman::p_load(tidyverse, knitr, DT, plotly, treemap, scales)
df <- read.csv("breeds.csv", na.strings=c("NA","NaN", " ", "?", "null", "NULL"))
df1 <- df[!df$weight == "null", ]
options(scipen = 10)
```

<style>
.chart-title { /* chart_title */
  font-size: 20px;
  }
body{ /* Normal */
  font-size: 18px;
  }
</style>


Introduction 
===

Column {.tabset data-width=600}
---

### Introduction to the Study

``` {r BehindTheScenes}
df$average <- apply(df[ ,c(3:33)], 1, function(x) mean(x, na.rm=T))
df$intelligence_average <- apply(df[ ,c(22:28)], 1, function(x) mean(x, na.rm=T))
df$adaptability_average <- apply(df[ ,c(3:9)], 1, function(x) mean(x, na.rm=T))
df$behavior_average <- apply(df[ ,c(10:14)], 1, function(x) mean(x, na.rm=T))
df$health_average <- apply(df[ ,c(15:21)], 1, function(x) mean(x, na.rm=T))
df$energy_average <- apply(df[ ,c(29:33)], 1, function(x) mean(x, na.rm=T))
```

**Data Overview**

For this final presentation, I have chosen to look at a data set containing information on various dog breeds. The data set looks at 348 different dog breeds! Majority of the dataset is ordinal variables with the scale being 1-5. If a dog is bad at a category, their score will be a 1. Conversely, if a dog excels at a category, their score will be a 5. The variables are grouped into 5 different categories; a, b , c, d, and e. Category a focuses on the dogs adaptability to its environment, category b focuses on variables related to the dogs behavior, category c focuses on variables related to the dogs health and well-being, category d focuses on how easy a dog is to train and it's primal urges, lastly, category e focuses on the exercise needs of the dog. 

My final project attempts to answer some of the questions below:

- What dog class/type scores the highest?

- What breed of dog has the best average score?

- What kind of trends, if any, can I find when conducting unsupervised learning? 

### Table of Data

```{r Table}
DT::datatable(df)
```

Column {.tabset data-width=400}
---

### Variable Explanation

Variables

- **breed**: Name of dog breed.
- **url**: URL takes you to a site with information about the dog.
- **a_adaptability**: General adaptability.
- **a1_adapts_well_to_apartment_living**: Adaptability to apartment living.
- **a2_good_for_novice_owners**: Generally how easy the dog is to take care of for new owners.
- **a3_sensitivity_level**: How sensitive the dog is.
- **a4_tolerates_being_alone**: How well the dog can handle being alone.
- **a5_tolerates_cold_weather**: How well the dog can handle colder weather.
- **a6_tolerates_hot_weather**: How well the dog can handle warmer weather.
- **b_all_around_friendliness**: General friendliness of the dog.
- **b1_affectionate_with_family**: How affectionate the dog is to its owners.
- **b2_incredibly_kid_friendly_dogs**: How kind/friendly the dog is.
- **b3_dog_friendly**: How friendly the dog is to other dogs.
- **b4_friendly_toward_strangers**: How friendly the dog is to unfamiliar people.
- **c_health_grooming**: The length of the movie in minutes.
- **c1_amount_of_shedding**: The amount of shedding the dog produces.
- **c2_drooling_potential**: Amount of drool the dog produces.
- **c3_easy_to_groom**: How easy it is to groom the dog.
- **c4_general_health**: Generally how healthy the dog is.
- **c5_potential_for_weight_gain**: How likely the dog is to gain weight.
- **c6_size**: The size of the dog.
- **d_trainability**: How well the dog can be trained.
- **d1_easy_to_train**: How easy it is to train the dog.
- **d2_intelligence**: How intelligent the dog is.
- **d3_potential_for_mouthiness**: 
- **d4_prey_drive**: How much hunting drive the dog have.
- **d5_tendency_to_bark_or_howl**: How much the dog barks/howls.
- **d6_wanderlust_potential**: How adventurous and exploratory the dog is.
- **e_exercise_needs**: How much exercise the dog needs.
- **e1_energy_level**: How much energy the dog has.
- **e2_intensity**: Intensity level of the dog.
- **e3_exercise_needs**: Duplicate variable.
- **e4_potential_for_playfulness**: How playful the dog is.
- **breed_group**: Type/Class of dog (ie. hound dogs, herding dogs, guard dogs, etc).
- **height**: Height range of the dog (in inches and feet).
- **weight**: Weight range of the dog (in pounds).
- **life_span**: Lifespan range of the dog (in years).

Tree Map
===

Column {.tabset data-width=600}
---

### Tree

``` {r HeatMap}
treemap(df, 
       index = "breed_group",
       vSize = "health_average",
       vColor = "energy_average",
       title = "Dog Breeds Tree Map",
       palette = "Blues")
```

Column {data-width=400}
---

### Analysis

For this tree, the size of the box represents how healthy the breed is on average,
where the color represents the amount of energy the breed has on average. The darker the blue
the more energy the category of dog has. Similarly, the larger the box, the healthier the type of
breed.


Best Breeds/Size
===

Column {.tabset data-width=600}
---

### Best Dog by Size

``` {r Boxplot}
df <- mutate(df, general_size = case_when(
  c6_size == 1 | c6_size == 2 ~ "Small",
  c6_size == 3 ~ "Medium",
  c6_size == 4 | c6_size == 5 ~ "Large"))

size_order <- c("Small", "Medium", "Large")
df$general_size <- factor(df$general_size, size_order)

ggplot(df,aes(x = general_size, y = average)) +
  geom_boxplot(col = "black", fill = "lightblue") +
  labs(x = "Size of Breed",
       y = "Overall Average",
       title = "Overall Average vs Size") -> average_vs_size_plot

ggplotly(average_vs_size_plot)
```


### Best Dog by Group

``` {r Boxplot2}
ggplot(df,aes(x = breed_group, y = average)) +
  geom_boxplot(col = "black", fill = "lightblue") +
  labs(x = "Group of Dog",
       y = "Overall Average",
       title = "Overall Average vs Dog Group") -> average_vs_size_plot1

ggplotly(average_vs_size_plot1)
```

### Best Dog by Breed

``` {r Barchart}
df_best <- df %>% 
  filter(average > mean(average, na.rm = T))

df_top10 <- df_best %>% 
  group_by(breed) %>%
  summarise(n = n()) %>%
  arrange(desc(n)) %>%
  .[1:10,] 

df_best10 <- df_best %>% 
  semi_join(df_top10, by = "breed")

ggplot(df_best10, aes(x = breed, y = average)) +
  geom_col(aes(text = paste0(breed, " \n",
                             "Average Score: ", round(average, 2), " \n",
                             "Intelligence: ", round(intelligence_average,2), " \n",
                             "Adaptability: ", round(adaptability_average,2), " \n",
                             "Behavior: ", round(behavior_average,2), " \n",
                             "Health: ", round(health_average,2), " \n",
                             "Energy: ", round(energy_average,2), " \n",
                             "Size: ", general_size)),
           fill = "lightblue", col = "black") +
  ylim(c(0,5)) +
  theme(axis.text.x = element_text(angle=20, hjust=.8),
        axis.text = element_text(size = 10)) +
  labs(title = "Overall Best Breeds",
       x = "Breed",
       y = "Overall Average") -> top_10_breeds_chart

ggplotly(top_10_breeds_chart, tooltip = "text")
```


Column {data-width=400}
---

### Boxplot Analysis

For the boxplot, I made a new variable called "general size". I used the 
size variable to put them into a new categorical variable. Dog breeds that scored
a 1 or 2 were classified as small, breeds that scored a 3 were deemed medium, and 
breeds that scored a 4 or 5 were deemed large. I mainly did it this way because when
I grouped them like that each category had around 120. It would be harder to get
accurate data if the sample sizes were drastically different,

### Barplot Analysis

For the barplot, I subset the data to the top 10 dog breeds that had the 
highest average across all their scores. I also added a function that shows their
general rating for the 5 main categories I touched on earlier in the presentation.



Hierarchical Clustering
===

Column {.tabset data-width=600}
---

### Dendrogram 

``` {r Dendrogram}
df_features <- df[,39:43]

d <- dist(df_features)

hc <- hclust(d, method = "complete")

plot(hc, labels = FALSE, hang = -1, main = "Hierarchical Clustering Dendrogram")
```

### Hierarchical Clustering Table

``` {r HierarchicalClusteringTable}
Cluster <- cutree(hc, 5)
Cluster <- as.factor(Cluster)

table(Cluster, df$breed_group)
```

### Hierarchical Clustering Plot
``` {r HierarchicalClusteringPlot}
ggplot(df, aes(x = intelligence_average, y = behavior_average, color = Cluster)) +
  geom_point(position = position_jitter(width = 0.1, height = 0.1),
             alpha = 0.5, size = 3.5, 
             aes(text = paste0 (breed, " \n",
                                "Dog Type: ", breed_group, " \n",
                                "Behavior: ", round(behavior_average,2), " \n", 
                                "Intelligence: ", round(intelligence_average,2)))) +
  scale_color_manual(values = c("#ef476f", "#ffd166", "#06d6a0", "#118ab2", "#073b4c"))+
  ylim(c(1.5,5)) +
  xlim(c(1.5,5)) +
  labs(x = "Intelligence",
       y = "Behavior",
       title = "Unsupervised Learning Between Intelligence and Adaptability")-> hc_plot

ggplotly(hc_plot, tooltip = "text")
```

Column {data-width=400}
---

### Dendrogram

This dendrogram has a very unique shape. It is common for the result to show three or four clusters, however here it would be better to do four or five.

### Hierarchical Clustering

As you can see, the five clusters are not in a spherical pattern and some of the clusters overlap a little.
This is no particular/extremely noticeable groups which was unfortunate. However it showed that
each dog breed is relatively different and can be hard to classify them into specific groups due to
their differences.

K-Means
===

Column {.tabset data-width=600}
---

### K-Means with Two Clusters

``` {r TwoCLusters}
iris_features <- df[,39:43]

set.seed(100)
kmeans_result <- kmeans(iris_features,
                        centers = 2)

plot(iris_features[,1], iris_features[,2], col = kmeans_result$cluster,
     xlab = 'Intelligence', ylab = 'Adaptability', main = 'K-Means Cluster')
points(kmeans_result$centers[,1], kmeans_result$centers[,2],
       col = 1:4, pch = 8, cex = 2)
```

### K-Means with Three Clusters

``` {r ThreeClusters}
iris_features <- df[,39:43]

set.seed(100)
kmeans_result <- kmeans(iris_features,
                        centers = 3)

plot(iris_features[,1], iris_features[,2], col = kmeans_result$cluster,
     xlab = 'Intelligence', ylab = 'Adaptability', main = 'K-Means Cluster')
points(kmeans_result$centers[,1], kmeans_result$centers[,2],
       col = 1:4, pch = 8, cex = 2)
```

### K-Means with Four Clusters

``` {r FourClusters}
iris_features <- df[,39:43]

set.seed(100)
kmeans_result <- kmeans(iris_features,
                        centers = 4)

plot(iris_features[,1], iris_features[,2], col = kmeans_result$cluster,
     xlab = 'Intelligence', ylab = 'Adaptability', main = 'K-Means Cluster')
points(kmeans_result$centers[,1], kmeans_result$centers[,2],
       col = 1:4, pch = 8, cex = 2)
```

Column {data-width=400}
---

### Analysis and Conclusion

For my K-means clustering I choose to do three clusters each with different K values.
I personally think two to three clusters were the best. That being said, two clusters
were very broad and not narrow enough, and three clusters had no real obvious groups. Similarly with 
four clusters, most of the points overlap into other clusters which really doesn't 
show any clear groups.

About the Author
===

Column {.tabset data-width=500}
---

### Who am I?

My name is Jonah Mergler and I am an undergraduate student at the University of Dayton studying Applied Mathematical Economics major with a minor in Data Analytics. I am on track to graduate in May 2025.

The summer after my sophomore year and into my 5th semester at UD, I was a Financial Analysts inter with an insurance broker firm named McGohan Brabender. During my time at McGohan Brabender, I worked with the Stop Loss model for the 100+ segment and created a lot of medical renewals for a large number of unique organizations.

My goal is to acquire an internship outside of the insurance industry to see what else the math workforce has else to offer. I am actively seeking employment on both Handshake and LinkedIn. If you are interested in reaching out, my linked in profile can be found [here](https://www.linkedin.com/in/jonah-mergler-dayton). 


### References and Limitations

**Limitations**

I think my biggest limitation was the categorical variables that were ranges of values.
It would have been okay to have a handful of ranges to choose from, however almost each range for 
height or weight, for example, were completely unique. Since I had so many data entries, it
would have taken a long time to adjust the data into similar categories or numerical data values. This
limited what I was able to do with those values when I really wished I could have 
explored them more.

**References**

[My data](https://www.kaggle.com/datasets/mexwell/dog-breeds-dogtime-dataset?resource=download)


Column {data-width=500}
---

### Picture of Me

``` {r Picture, fig.width=6, echo=FALSE, fig.cap="My friend Sam(left) and I(right) in the streets of San Deigo, California."}
knitr::include_graphics("Sam&MeCali.jpg")
```