2022_01_25_board_games.Rmd

---
title: "Board games"
date: 2022-01-25
output: html_output
---

# TidyTuesday

Join the R4DS Online Learning Community in the weekly #TidyTuesday event!
Every week we post a raw dataset, a chart or article related to that dataset, and ask you to explore the data.
While the dataset will be “tamed”, it will not always be tidy! As such you might need to apply various R for Data Science techniques to wrangle the data into a true tidy format.
The goal of TidyTuesday is to apply your R skills, get feedback, explore other’s work, and connect with the greater #RStats community!
As such we encourage everyone of all skills to participate!

```{r setup, include=FALSE}

knitr::opts_chunk$set(echo = TRUE)

library(tidyverse)
library(tidytuesdayR)

library(scales)
theme_set(theme_light())
```

# Load the weekly Data

Dowload the weekly data and make available in the `tt` object.

```{r Load}

tt <- tt_load("2022-01-25")

```


# Readme

Take a look at the readme for the weekly data to get insight on the dataset.
This includes a data dictionary, source, and a link to an article on the data.

```{r Readme, eval = interactive()}

tt

```


# Glimpse Data

Take an initial look at the format of the data available.

```{r Glimpse}

tt %>% 
  map(glimpse)

```

# Wrangle

Explore the data and process it into a nice format for plotting! Access each dataset by name by using a dollarsign after the `tt` object and then the name of the data set.

```{r Wrangle}

ratings <- tt$ratings
details <- tt$details

```


# Visualize

Using your processed dataset, create your unique visualization.

Is the target "average" or "bayes_average"?

```{r Visualize}

ratings %>% 
  ggplot(aes(average)) +
  geom_histogram()

ratings %>% 
  ggplot(aes(bayes_average)) +
  geom_histogram()

ratings %>% 
  ggplot(aes(users_rated)) +
  geom_histogram() +
  scale_x_log10()

```

Game rank is according to "bayes_average". So the target is likely to be "bayes_average", which tends to increase as "users_rated" increases.

```{r}

ratings %>% 
  count(rank, sort = TRUE)

ratings %>% 
  ggplot(aes(rank, average)) +
  geom_point(alpha = 0.2)

ratings %>% 
  ggplot(aes(rank, bayes_average)) +
  geom_point(alpha = 0.2)

ratings %>% 
  ggplot(aes(users_rated, bayes_average)) +
  geom_point(alpha = 0.2) +
  scale_x_log10()
```

Game year doesn't mean much. "yearpublished" doesn't either.

```{r}

ratings %>% 
  ggplot(aes(year, bayes_average)) +
  geom_point(alpha = 0.2)

rat_det <- ratings %>% 
  left_join(details, by = "id")

rat_det %>% 
  ggplot(aes(yearpublished, bayes_average)) +
  geom_point(alpha = 0.2)

```

```{r}

rat_det %>% 
  ggplot(aes(minplayers, bayes_average, group = minplayers)) +
  geom_boxplot()

rat_det %>% 
  ggplot(aes(maxplayers, bayes_average, group = maxplayers)) +
  geom_boxplot() +
  scale_x_log10()

```

```{r}

rat_det %>% 
  ggplot(aes(playingtime, bayes_average)) +
  geom_point(alpha = 0.2) +
  scale_x_log10()

rat_det %>% 
  ggplot(aes(minplaytime, bayes_average)) +
  geom_point(alpha = 0.2) +
  scale_x_log10()

rat_det %>% 
  ggplot(aes(maxplaytime, bayes_average)) +
  geom_point(alpha = 0.2) +
  scale_x_log10()

```

```{r}
rat_det %>% 
  count(boardgamecategory, sort = TRUE)

plot_cat <- function(df, var) {
  df %>% 
    filter(!is.na({{ var }})) %>% 
  mutate({{ var }} := str_remove_all({{ var }}, "\\[|\\]|'|\\\"")) %>% 
  separate_rows({{ var }}, sep = ", ") %>% 
  mutate({{ var }} := fct_lump_n({{ var }}, n = 10)) %>% 
  add_count({{ var }}) %>% 
  mutate(
    {{ var }} := paste0(as.character({{ var }}), " (", n, ")"),
    {{ var }} := fct_reorder({{ var }}, bayes_average)
  ) %>% 
  ggplot(aes(bayes_average, {{ var }})) +
  geom_boxplot()
}

rat_det %>% 
  plot_cat(boardgamecategory)

```

```{r}
rat_det %>% 
  count(boardgamemechanic, sort = TRUE)

rat_det %>% 
  plot_cat(boardgamemechanic)

```

```{r}
rat_det %>% 
  count(boardgamefamily, sort = TRUE)

```

```{r}
rat_det %>% 
  count(boardgameexpansion, sort = TRUE)
```

```{r}
rat_det %>% 
  count(boardgameimplementation, sort = TRUE)
```

```{r}
rat_det %>% 
  count(boardgamedesigner, sort = TRUE)

rat_det %>% 
  plot_cat(boardgamedesigner)

```

```{r}
rat_det %>% 
  count(boardgameartist, sort = TRUE)

rat_det %>% 
  plot_cat(boardgameartist)
```

```{r}
rat_det %>% 
  count(boardgamepublisher, sort = TRUE)

rat_det %>% 
  plot_cat(boardgamepublisher)
```


```{r}
rat_det %>% 
  ggplot(aes(owned, bayes_average)) +
  geom_point(alpha = 0.2) +
  scale_x_log10()

rat_det %>% 
  ggplot(aes(trading, bayes_average)) +
  geom_point(alpha = 0.2) +
  scale_x_log10()

rat_det %>% 
  ggplot(aes(wanting, bayes_average)) +
  geom_point(alpha = 0.2) +
  scale_x_log10()

rat_det %>% 
  ggplot(aes(wishing, bayes_average)) +
  geom_point(alpha = 0.2) +
  scale_x_log10()

```


```{r}
library(tidytext)

tidy_desc <- details %>% 
  unnest_tokens(word, description) %>% 
  anti_join(get_stopwords(), by = "word") %>% 
  filter(str_detect(word, "^[:alpha:]+$"))

tidy_desc %>% 
  count(word, sort = TRUE)

p <- tidy_desc %>% 
  left_join(ratings, by = "id") %>% 
  group_by(word) %>% 
  summarize(
    n = n(),
    rating = mean(bayes_average)
  ) %>% 
  ggplot(aes(n, rating)) +
  geom_hline(yintercept = mean(ratings$bayes_average), color = "blue") +
  geom_text(aes(label = word), check_overlap = TRUE,
            vjust = "top", hjust = "left") +
  scale_x_log10(limits = c(3, 1e+02)) +
  scale_y_continuous(limits = c(3, 9)) +
  labs(x = "# of appearances in description", y = "Bayes average of ratings",
       title = "'disguise', 'flocks' and 'bumped' tend to lower ratings,\nwhile common words coverge to the average rating",
       subtitle = "Top-left of each word indicates rating\nBlue horizontal line is the average rating",
       caption = "Source: Kaggle")

p
```

# Save Image

Save your image for sharing. Be sure to use the `#TidyTuesday` hashtag in your post on twitter! 

```{r}

# This will save your most recent plot
ggsave("image/board_games.png", p, width = 6, height = 6)

```

# Imitate Julia Silge

```{r}
ratings_joined <- ratings %>% 
  left_join(details, by = "id")

ratings_joined %>% 
  ggplot(aes(average)) +
  geom_histogram()
```

```{r}
ratings_joined %>% 
  filter(!is.na(minage)) %>% 
  mutate(minage = cut_number(minage, 4)) %>% 
  ggplot(aes(minage, average, fill = minage)) +
  geom_boxplot(show.legend = FALSE, alpha = 0.2)
```

## Build models

Let's consider how to [spend our data budget](https://www.tmwr.org/splitting.html):

- create training and testing sets
- create resampling folds from the *training* set

```{r}
library(tidymodels)

set.seed(123)
game_split <- ratings_joined %>% 
  select(name, average, starts_with(c("min", "max")), boardgamecategory) %>% 
  na.omit() %>% 
  initial_split(strata = average)
game_train <- training(game_split)
game_test <- testing(game_split)

set.seed(234)
game_folds <- vfold_cv(game_train, strata = average)
game_folds
```

Let's create a recipe.

```{r}
library(textrecipes)

split_category <- function(x) {
  x %>% 
    str_remove_all("[:punct:]") %>% 
    str_trim() %>% 
    str_squish() %>% 
    str_to_lower() %>% 
    str_replace_all(" ", "_") %>% 
    str_split(", ")
}

game_rec <- recipe(average ~ ., data = game_train) %>% 
  update_role(name, new_role = "id") %>% 
  step_tokenize(boardgamecategory, custom_token = split_category) %>% 
  step_tokenfilter(boardgamecategory, max_tokens = 30) %>% 
  step_tf(boardgamecategory)

game_rec %>% 
  prep() %>% 
  bake(new_data = NULL) %>% names()
```

Let's create a [**model specification**](https://www.tmwr.org/models.html) for each model we want to try:

```{r}
xgb_spec <-
  boost_tree(
    trees = tune(),
    mtry = tune(),
    min_n = tune(),
    learn_rate = 0.01
  ) %>% 
  set_engine("xgboost") %>% 
  set_mode("regression")

xgb_wf <- workflow(game_rec, xgb_spec)
xgb_wf
```

## Evaluate models

These models have no tuning parameters so we can evaluate them as they are. [Learn about tuning hyperparameters here.](https://www.tidymodels.org/start/tuning/)

```{r}
library(finetune)
doParallel::registerDoParallel(cores = 5)

set.seed(234)
xgb_game_rs <- tune_race_anova(
  xgb_wf,
  resamples = game_folds,
  grid = 20,
  control = control_race(verbose_elim = TRUE)
)

xgb_game_rs
```

```{r}
plot_race(xgb_game_rs)
```

```{r}
show_best(xgb_game_rs)
```

```{r}
select_best(xgb_game_rs)
```

These models perform very similarly, so perhaps we would choose the simpler, linear model. The function `last_fit()` *fits* one final time on the training data and *evaluates* on the testing data. This is the first time we have used the testing data.

```{r}
xgb_last <- xgb_wf %>% 
  finalize_workflow(select_best(xgb_game_rs)) %>% 
  last_fit(game_split)

xgb_last
```

```{r}
library(vip)

xgb_fit <- xgb_last %>% 
  extract_fit_parsnip()

vip(xgb_fit, geom = "point", num_features = 12)
```

Give up SHAP.