index.qmd

---
title: "The untold story of palmerpenguins 🐧"
author:
  - Dr. Kristen Gorman, University of Alaska Fairbanks
  - Dr. Allison Horst, UC Santa Barbara
  - Dr. Alison Hill, Voltron Data
format: 
  revealjs:
    multiplex: true
    footer: "[apreshill.github.io/palmerpenguins-useR-2022/](https://apreshill.github.io/palmerpenguins-useR-2022/)"
    theme: ["pp.scss"]
    slide-number: c/t
    incremental: true
    title-slide-attributes:
      data-background-image: img/penguin-highway.jpg
      data-background-size: cover  
editor: visual
---

```{r setup}
#| include: false

library(palmerpenguins)

# for interactive plots
library(ggiraph)
library(patchwork)
library(shadowtext)

# for color
library(paletteer)

# for cran plots
library(cranlogs)
library(tidyverse)
library(scales)

# for fonts
library(showtext)
font_add_google("Jost", "Jost")
showtext_auto()
theme_set(theme_minimal(base_family = "Jost"))

library(recipes)
library(broom)
# for reporting cran downloads
start_date <- "2020-07-23"
end_date <- "2022-06-15"
```

```{r}
#| eval: false
#| include: false
# do this just once
dl <- cranlogs::cran_downloads("palmerpenguins",
                               from = start_date,
                               to = end_date) %>% 
  as_tibble() 
saveRDS(dl, file = "data/cran-downloads.RDS") 
```

## What is palmerpenguins?

::: columns
::: column
-   An R package featuring the `penguins` dataset

-   `r nrow(penguins)` penguins

-   `r n_distinct(penguins$species)` penguin species (Adélie, chinstrap, and gentoo)

-   `Inf` fun
:::

::: column
```{r}
#| fig.alt: "Scatterplot showing linear relationships between penguin body mass and flipper length, both across and within species (Adelie, Chinstrap and Gentoo)."
# make the base plot
penguin_linear_base <-
  penguins %>%
  mutate(species = as.character(species)) %>%
  mutate(species = case_when(
    species == "Adelie" ~ "Adélie",
    TRUE ~ species
  )) %>%
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  scale_color_paletteer_d("colorblindr::OkabeIto") +
  theme(legend.position = c(0.2,0.85)) +
  labs(x = "Flipper length (mm)",
       y = "Body mass (g)") +
  theme(plot.title.position = "plot",
        #panel.grid = element_blank(),
        panel.border = element_rect(fill = NA, color = "gray70"),
        legend.title = element_blank())

# add interactivity
penguins_tooltip <- c(str_c("Flipper length (mm) = ", penguins$flipper_length_mm,
                           "\n Body mass (g) = ", penguins$body_mass_g,
                           "\n Species = ", penguins$species))

penguin_linear_int <-
  penguin_linear_base +
  geom_point_interactive(aes(color = species,
                             shape = species,
                             tooltip = penguins_tooltip,
                             data_id = species),
                         size = 2,
                         alpha = .8
                         ) +
  geom_smooth_interactive(aes(group = species,
                              color = species,
                              data_id= species),
                          method = lm,
                          se = FALSE
                          )

girafe(ggobj = penguin_linear_int,
       options = list(
         opts_hover_inv(css = "opacity:0.2;"),
         opts_hover(css = "opacity:1; filter: brightness(90%);")
       ))
```
:::
:::

::: notes
Kristen presents
:::

## palmerpenguins popularity

```{r}
dl <- readRDS("data/cran-downloads.RDS")
```

::: columns
::: {.column width="80%"}
```{r cran-trend}
#| fig.alt: "Line graph showing palmerpenguins package downloads from CRAN, which increases quickly since published in 2020 and has leveled off near 1000 downloads per day."
#| fig.height: 4
# from https://github.com/hadley/cran-downloads/blob/master/server.R
show_trend <- function(df) {
  count_ts <- ts(df$count, frequency = 7)
  stl <- tryCatch(
    as.data.frame(stl(count_ts, "periodic")$time.series),
    error = function(e) data.frame(trend = df$count, remainder = 0)
  )
  df$trend <- stl$trend
  df$remainder <- stl$remainder
  df
}
trend <- dl %>% group_by(package) %>% do(show_trend(.))
ggplot(trend, aes(date)) +
  geom_linerange(aes(ymin = trend, ymax = trend + remainder), colour = "grey70", alpha = .5, lwd = 1) +
  geom_line(aes(y = trend), color = "#FB7967", lwd = 1) +
  labs(y = "Daily downloads (smoothed)", x = NULL) 
```

-   \> `r scales::number(sum(dl$count), big.mark = ",", accuracy = 100)` CRAN downloads since `r start_date`

-   Used globally in courses, workshops, blog posts, and other learning materials

-   `penguins` now in [Python](https://github.com/mcnakhaee/palmerpenguins), [Julia](https://juliahub.com/ui/Packages/PalmerPenguins/L7ZJc/0.1.4), and [TensorFlow](https://www.tensorflow.org/datasets/catalog/penguins)
:::

::: {.column width="20%"}
[![palmerpenguins hex sticker by Allison Horst](https://allisonhorst.github.io/palmerpenguins/reference/figures/palmerpenguins.png){fig-alt="palmerpenguins hex sticker with all three penguin species by Allison Horst"}](https://allisonhorst.github.io/palmerpenguins/index.html)
:::
:::

::: notes
Kristen presents
:::

##  {background-image="img/penguins.png"}

### How the penguins came to be {.r-fit-text}

## The research

::: columns
::: {.column width="50%"}
![Dr. Kristen Gorman in the field, surrounded by penguins, at islands near Palmer Archipelago, Antarctica](img/kristen.png){fig-alt="Dr. Kristen Gorman in the field, surrounded by penguins, at islands near Palmer Archipelago, Antarctica"}
:::

::: {.column width="50%"}
-   An integrative study of the breeding ecology and population structure of Pygoscelis penguins along the western Antarctic Peninsula as part of the Palmer LTER Program (US NSF)

-   The data were [originally published](https://doi.org/10.1371/journal.pone.0090081) in PLoS ONE in 2014[^1]

-   All data were made available through the [Environmental Data Initiative](https://environmentaldatainitiative.org)
:::
:::

[^1]: Gorman KB, Williams TD, Fraser WR (2014). Ecological sexual dimorphism and environmental variability within a community of Antarctic penguins (genus Pygoscelis). PLoS ONE 9(3):e90081. <https://doi.org/10.1371/journal.pone.0090081>

::: notes
Kristen presents
:::

##  {background-image="img/avian-island.jpeg"}

### Little did the penguin research team know... {.r-fit-text}

## Wanted: an alternative to iris

::: columns
::: column
[![Measurements of an iris flower (source: Science Direct)](img/iris.jpeg){fig-alt="Measurements of an iris flower"}](https://www.sciencedirect.com/topics/mathematics/virginica)
:::

::: column
-   Collected by botanist Edgar Anderson in 1935

-   Used *everywhere* in data science teaching & resources

-   `r nrow(iris)` size measurements for `r n_distinct(iris$Species)` species of iris

-   No missing values

-   Lacks metadata

-   Variables like `Sepal.Width`

-   Published in The Annals of Eugenics (RA Fisher, 1936)
:::
:::

::: notes
Allison presents
:::

## If something is a problem, offer a solution

1.  Keep using `iris` and use it as an opportunity to learn/teach about its problematic aspects.

2.  Find a better dataset to replace `iris`.

##  {background-image="img/penguin.jpeg"}

### The eternal search for (better) useful datasets for teaching {.r-fit-text}

## Serendipity

::: columns
::: {.column width="60%"}
-   Allison stumbles upon [Gorman et al.](https://dx.plos.org/10.1371/journal.pone.0090081) and shares it with Alison

-   Alison writes a [blog post](https://www.apreshill.com/blog/2020-02-take-a-sad-script-make-it-better-tidymodels-edition/) with penguins after Allison shares it with her

    [![Screenshot from Alison's first blog post using penguins from Allison](img/blog-post.png)](https://www.apreshill.com/blog/2020-02-take-a-sad-script-make-it-better-tidymodels-edition/)
:::

::: {.column width="40%"}
[![Art by Rohan Chakravarty](https://www.apreshill.com/blog/2020-02-take-a-sad-script-make-it-better-tidymodels-edition/penguins-of-the-world.JPG){fig-alt="21 of the described species with their respective IUCN status. Aquatic and flightless birds with remarkable swimming adaptations, penguins inhabit the Southern hemisphere (except the Galapagos Penguin). They range in size from the 45 kg Emperor Penguin to the 1.5 kg Little Penguin. Penguins are threatened by habitat loss and climate change, and the African, Galapagos and Yellow-eyed are some of the most endangered penguin species." width="80%"}](https://www.greenhumour.com/2018/04/penguins-of-world.html)
:::
:::

::: notes
Allison presents
:::

## Finding the one

::: columns
::: column
-   Meanwhile, Allison keeps playing with the penguins

-   And plotting with the penguins

-   And looking at more penguin pictures

-   Aligns eerily well with iris data
:::

::: column
```{r}
#| fig.alt: "Scatterplots highlighting linear relationships between penguin body mass and flipper length (left) and iris petal length and petal width (right), both across and within the three species for each dataset."
# Iris linear relationships (petal dimensions):
iris_linear_base <- ggplot(iris, aes(x = Petal.Length, y = Petal.Width)) +
  scale_color_manual(values = c("gray70","gray40","black")) +
  theme(legend.position = c(0.2, 0.85)) +
  labs(x = "Petal length (cm)",
       y = "Petal width (cm)") +
  theme(plot.title.position = "plot",
        #panel.grid = element_blank(),
        panel.border = element_rect(fill = NA, color = "gray70"),
        legend.title = element_blank())

iris_tooltip <- c(str_c("Petal length = ", iris$Petal.Length,
                        "\n Petal width = ", iris$Petal.Width,
                        "\n Species = ", iris$Species))

iris_linear_int <-
  iris_linear_base +
  geom_point_interactive(aes(color = Species,
                             shape = Species,
                             tooltip = iris_tooltip,
                             data_id = Species),
                         size = 2,
                         alpha = .8
                         ) +
  geom_smooth_interactive(aes(group = Species,
                              color = Species,
                              data_id= Species),
                          method = lm,
                          se = FALSE
  )

girafe(code = print(penguin_linear_int + iris_linear_int),
       width_svg = 8,
       height_svg = 4,
       options = list(
         opts_hover_inv(css = "opacity:0.2;"),
         opts_hover(css = "opacity:1; filter: brightness(90%);")
       ))
```
:::
:::

::: notes
Allison presents
:::

## June 2020 \~4 months into lockdown ☣️

::: columns
::: {.column width="60%"}
-   June 4: 🐦 In excitement, Allison Horst tweets about the penguins data

    -   📦 Alison asks "Do you wanna build a package?"

    -   ❓Allison deletes tweet to sort out license

-   1 day later: 🔏 Allison emails Kristen about use & license

-   3 days later (June 8): 🚀 palmerpenguins power team assembles for the first time!

-   Same day (June 8): 🏆 Allison Horst tweets (again) --- with clear license info

-   45 days later (July 23): 🎁 available on [CRAN](https://cran.r-project.org/web/packages/palmerpenguins/index.html)
:::

::: {.column width="40%"}
```{=html}
<blockquote class="twitter-tweet"><p lang="en" dir="ltr">🐧🐧🐧<br><br>This penguin data is a great alternative to iris &amp; available for use by CC0 🤩 Thank you Dr. Kristen Gorman w/ <a href="https://twitter.com/UAFcfos?ref_src=twsrc%5Etfw">@UAFcfos</a>, Marty Downs w/ <a href="https://twitter.com/USLTER?ref_src=twsrc%5Etfw">@USLTER</a>, &amp; <a href="https://twitter.com/PalmerLTER?ref_src=twsrc%5Etfw">@PalmerLTER</a> for help, info &amp; making it available for use 🎉<br><br>Data, examples, &amp; use info here: <a href="https://t.co/dSIqWNFlVw">https://t.co/dSIqWNFlVw</a> 🧵 1/6 <a href="https://t.co/2Eu4AxoeZl">pic.twitter.com/2Eu4AxoeZl</a></p>&mdash; Allison Horst (@allison_horst) <a href="https://twitter.com/allison_horst/status/1270046399418138625?ref_src=twsrc%5Etfw">June 8, 2020</a></blockquote> <script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>
```
:::
:::

::: notes
Allison presents
:::

##  {background-image="img/iceberg.jpeg"}

### The making of... {.r-fit-text}

## For educators, by educators

```{r echo=TRUE}
library(palmerpenguins)
path_to_file()
path_to_file("penguins_raw.csv")
head(read.csv(path_to_file("penguins.csv")))
```

## To tibble or...

```{r echo=FALSE}
palmerpenguins::penguins 
```

## Not to tibble, that is the question

> "I think this gives the best of both worlds --- if you're running in a low dependency environment, you'll get a data frame."
>
> --- Hadley Wickham[^2]

[^2]: Thanks to Hadley [for this PR!](https://github.com/allisonhorst/palmerpenguins/pull/25)

```{r echo=FALSE}
palmerpenguins::penguins %>% 
  as.data.frame()
```

::: notes
Alison HILL presents
:::

##  {background-iframe="https://allisonhorst.github.io/palmerpenguins/"}

## Penguins in the wild {background-image="img/penguins-wilder.png" background-size="contain"}

::: notes
Alison HILL presents
:::

## Alison: Why has it been so popular?

-   The R package format - so easy to just use

-   Reduces time to first pretty plot

-   [Allison's art](https://allisonhorst.github.io/palmerpenguins/articles/art.html)

    [![Allison Horst's illustration of the 3 penguin species](img/allison-penguins.png){width="800"}](https://allisonhorst.github.io/palmerpenguins/articles/art.html)

## Allison: Why has it been so popular?

::: columns
::: column
The data itself---distributions, linearity

```{r}
#| fig.alt: "Scatterplots of bill depth versus body mass for the three penguin species, showing a positive linear relationship within species that switches to a negative linear relationship if species is omitted, an example of Simpson’s paradox."
# Simpson's Paradox example (body mass, omitting species):
bmass_nospecies_base <- penguins %>%
  # doing this so ggiraph recognizes species across plots
  mutate(species = as.character(species)) %>%
  mutate(species = case_when(
    species == "Adelie" ~ "Adélie",
    TRUE ~ species)
  ) %>%
  ggplot(aes(x = bill_depth_mm, y = body_mass_g)) +
  theme(panel.border = element_rect(fill = NA, color = "gray70")) +
  labs(x = "Bill depth (mm)", y = "Body mass (g)")

# Body mass, including species:
bmass_wspecies_base <-
  penguins %>%
  mutate(species = as.character(species)) %>%
  mutate(species = case_when(
    species == "Adelie" ~ "Adélie",
    TRUE ~ species)
    ) %>%
  ggplot(aes(x = bill_depth_mm, y = body_mass_g, group = species)) +
  scale_color_paletteer_d("colorblindr::OkabeIto") +
  theme(panel.border = element_rect(fill = NA, color = "gray70")) +
  labs(x = "Bill depth (mm)", y = "Body mass (g)") +
  guides(color = guide_legend("Species"),
           shape = guide_legend("Species"))

nospecies_tooltip <- c(str_c("Bill depth (mm) = ", penguins$bill_depth_mm,
                            "\n Body mass (g) = ", penguins$body_mass_g,
                            "\n Species = ", penguins$species))

bmass_nospecies_int <-
  bmass_nospecies_base +
  geom_point_interactive(aes(tooltip = nospecies_tooltip,
                             data_id = species),
                         size = 2,
                         alpha = 0.8) +
  geom_smooth_interactive(method = lm,
                          se = FALSE,
                          color = "black"
  )


wspecies_tooltip <- c(str_c("Bill depth (mm) = ", penguins$bill_depth_mm,
                            "\n Body mass (g) = ", penguins$body_mass_g,
                            "\n Species = ", penguins$species))

bmass_wspecies_int <-
  bmass_wspecies_base +
  geom_point_interactive(aes(color = species,
                             shape = species,
                             tooltip = wspecies_tooltip,
                             data_id = species),
                         size = 2,
                         alpha = 0.8) +
  geom_smooth_interactive(aes(color= species,
                              data_id= species),
                          method = lm,
                          se = FALSE
  )


girafe(code = print(bmass_nospecies_int + bmass_wspecies_int),
       width_svg = 6,
       height_svg = 3,
       options = list(
         opts_hover_inv(css = "opacity:0.2;"),
         opts_hover(css = "opacity:1; filter: brightness(90%);")
       ))

```
:::

::: column
Simpson's paradox examples in the wild!

```{r}
#| fig.alt: "Scatterplot of bill length versus bill depth for the three penguin species, showing a positive linear relationship within species. If species is omitted as a variable, the relationship switches to a negative trend, another example of Simpson’s paradox in the data"
# Simpson's Paradox example (bill dimensions, omitting species):
simpson_nospecies_base <- penguins %>%
  # doing this so ggiraph recognizes species across plots
  mutate(species = as.character(species)) %>%
  mutate(species = case_when(
    species == "Adelie" ~ "Adélie",
    TRUE ~ species)
  ) %>%
  ggplot(aes(x = bill_length_mm, y = bill_depth_mm)) +
  theme(panel.border = element_rect(fill = NA, color = "gray70")) +
  labs(x = "Bill length (mm)", y = "Bill depth (mm)")

# Bill dimensions, including species:
simpson_wspecies_base <-
  penguins %>%
  mutate(species = as.character(species)) %>%
  mutate(species = case_when(
    species == "Adelie" ~ "Adélie",
    TRUE ~ species)
    ) %>%
  ggplot(aes(x = bill_length_mm, y = bill_depth_mm, group = species)) +

  scale_color_paletteer_d("colorblindr::OkabeIto") +
  theme(panel.border = element_rect(fill = NA, color = "gray70")) +
  labs(x = "Bill length (mm)", y = "Bill depth (mm)") +
  guides(color = guide_legend("Species"),
           shape = guide_legend("Species"))

nospecies_tooltip <- c(str_c("Bill length (mm) = ", penguins$bill_length_mm,
                            "\n Bill depth (g) = ", penguins$bill_depth_mm,
                            "\n Species = ", penguins$species))

simpson_nospecies_int <-
  simpson_nospecies_base +
  geom_point_interactive(aes(tooltip = nospecies_tooltip,
                             data_id = species),
                         size = 2,
                         alpha = 0.6) +
  geom_smooth_interactive(method = lm,
                          se = FALSE,
                          color = "black"
  )


wspecies_tooltip <- c(str_c("Bill length (mm) = ", penguins$bill_length_mm,
                             "\n Bill depth (g) = ", penguins$bill_depth_mm,
                             "\n Species = ", penguins$species))

simpson_wspecies_int <-
  simpson_wspecies_base +
  geom_point_interactive(aes(color = species,
                             shape = species,
                             tooltip = wspecies_tooltip,
                             data_id = species),
                         size = 2,
                         alpha = 0.6) +
  geom_smooth_interactive(aes(color= species,
                              data_id= species),
                          method = lm,
                          se = FALSE
  )

girafe(code = print(simpson_nospecies_int + simpson_wspecies_int),
       width_svg = 6,
       height_svg = 3,
       options = list(
         opts_hover_inv(css = "opacity:0.2;"),
         opts_hover(css = "opacity:1; filter: brightness(90%);")
       ))
```
:::
:::

## Allison: Why has it been so popular?

Good bones for multivariate analyses

::: columns
::: column
Principle Components Analysis

```{r}
#| fig.alt: "Principal component biplot comparing results for penguins and iris, highlighting strong similarities between the two"
## ---- pca ---------------------------------------------------------
# From Alison Hill's post for palmerpenguins pkgdown site

### PENGUINS PCA:

# Omit year
penguins_noyr <- penguins %>%
  select(-year) %>%
  mutate(species = as.character(species)) %>%
  mutate(species = case_when(
    species == "Adelie" ~ "Adélie",
    TRUE ~ species
  )) %>%
  mutate(species = as.factor(species))

penguin_recipe <-
  recipe(~., data = penguins_noyr) %>%
  update_role(species, island, sex, new_role = "id") %>%
  step_naomit(all_predictors()) %>%
  step_normalize(all_predictors()) %>%
  step_pca(all_predictors(), id = "pca") %>%
  prep()

penguin_pca <-
  penguin_recipe %>%
  tidy(id = "pca")

penguin_percvar <- penguin_recipe %>%
  tidy(id = "pca", type = "variance") %>%
  dplyr::filter(terms == "percent variance")

# Make the penguins PCA biplot:

# Get pca loadings into wider format
pca_wider <- penguin_pca %>%
  tidyr::pivot_wider(names_from = component, id_cols = terms)

# define arrow style:
arrow_style <- arrow(length = unit(.05, "inches"),
                     type = "closed")

penguins_juiced <- juice(penguin_recipe)

# Make the penguins PCA biplot:
pca_plot <-
  penguins_juiced %>%
  ggplot(aes(PC1, PC2)) +
  coord_cartesian(
    xlim = c(-3, 4),
    ylim = c(-2.5, 2))  +
  scale_color_paletteer_d("colorblindr::OkabeIto") +
  guides(color = guide_legend("Species"),
        shape = guide_legend("Species")) +
  theme(legend.position = "bottom",
        panel.border = element_rect(color = "gray70", fill = NA))
# For positioning (above):
# 1: bill_length
# 2: bill_depth
# 3: flipper length
# 4: body mass

penguins_biplot <- pca_plot +
  geom_segment(data = pca_wider,
               aes(xend = PC1, yend = PC2),
               x = 0,
               y = 0,
               arrow = arrow_style) +
  geom_point(aes(color = species, shape = species),
             alpha = 0.7,
             size = 2) +
  geom_shadowtext(data = pca_wider,
                  aes(x = PC1, y = PC2, label = terms),
                  nudge_x = c(0.7,0.7,1.7,1.2),
                  nudge_y = c(-0.1,-0.2,0.1,-0.1),
                  size = 4,
                  color = "black",
                  bg.color = "white")

### IRIS PCA:

iris_recipe <-
  recipe(~., data = iris) %>%
  update_role(Species, new_role = "id") %>%
  step_normalize(all_predictors()) %>%
  step_pca(all_predictors(), id = "pca") %>%
  prep()
iris_pca <-
  iris_recipe %>%
  tidy(id = "pca")

### Iris PCA biplot:

# Get pca loadings into wider format
iris_wider <- iris_pca %>%
  tidyr::pivot_wider(names_from = component, id_cols = terms)

# define arrow style:
arrow_style <- arrow(length = unit(.05, "inches"),
                     type = "closed")

iris_juiced <- juice(iris_recipe)

# Make the iris PCA biplot:
iris_pca_plot <-
  iris_juiced %>%
  ggplot(aes(PC1, PC2)) +
  coord_cartesian(
    xlim = c(-3, 3),
    ylim = c(-3, 3)) +
  scale_colour_manual(values = c("gray70","gray40","black")) +
  theme(panel.background = element_rect(fill = NA, color = "gray70"),
        legend.position = "bottom")

iris_biplot <- iris_pca_plot  +
  geom_point(aes(color = Species, shape = Species),
             alpha = 0.8,
             size = 2) +
  geom_segment(data = iris_wider,
               aes(xend = PC1, yend = PC2),
               x = 0,
               y = 0,
               arrow = arrow_style) +
  geom_shadowtext(data = iris_wider,
                  aes(x = PC1, y = PC2, label = terms),
                  nudge_x = c(0.5,0.3,1,1.2),
                  nudge_y = c(-0.1,-0.2,0.1,-0.1),
                  size = 4,
                  color = "black",
                  bg.color = "white")

penguins_biplot_tooltip <- str_c("Species = ", penguins_juiced$species,
                             "\n PC1 = ", round(penguins_juiced$PC1, 2),
                             "\n PC2 = ", round(penguins_juiced$PC2, 2))

penguins_biplot_int <- pca_plot +
  geom_segment_interactive(data = pca_wider,
               aes(xend = PC1, yend = PC2),
               x = 0,
               y = 0,
               arrow = arrow_style) +
  geom_point_interactive(aes(color = species, shape = species,
                             tooltip = penguins_biplot_tooltip,
                             data_id = species),
             alpha = 0.7,
             size = 2) +
  geom_shadowtext(data = pca_wider,
                  aes(x = PC1, y = PC2, label = terms),
                  nudge_x = c(0.7,0.7,1.7,1.2),
                  nudge_y = c(-0.1,-0.2,0.1,-0.1),
                  size = 4,
                  color = "black",
                  bg.color = "white")

# girafe(ggobj = penguins_biplot_int)

iris_biplot_tooltip <- str_c("Species = ", iris_juiced$Species,
                               "\n PC1 = ", round(iris_juiced$PC1, 2),
                               "\n PC2 = ", round(iris_juiced$PC2, 2))

iris_biplot_int <- iris_pca_plot +
  geom_segment_interactive(data = iris_wider,
                           aes(xend = PC1, yend = PC2),
                           x = 0,
                           y = 0,
                           arrow = arrow_style) +
  geom_point_interactive(aes(color = Species, shape = Species,
                             tooltip = iris_biplot_tooltip,
                             data_id = Species),
                         alpha = 0.7,
                         size = 2) +
  geom_shadowtext(data = iris_wider,
                  aes(x = PC1, y = PC2, label = terms),
                  nudge_x = c(0.5,0,1,1),
                  nudge_y = c(-0.1,-0.2,0.2,-0.1),
                  size = 4,
                  color = "black",
                  bg.color = "white")

girafe(code = print(penguins_biplot_int | iris_biplot_int),
       width_svg = 6,
       height_svg = 3,
       options = list(
         opts_hover_inv(css = "opacity:0.2;"),
         opts_hover(css = "opacity:1; filter: brightness(90%);")
       ))

```
:::

::: column
K-means

```{r}
#| fig.alt: "Compared k-means clustering scatterplots for penguins and iris."
## ---- kmeans ---------------------------------------------------------

# TWO VARIABLE k-means comparison
# Penguins: Bill length vs. bill depth
pb_species <- penguins %>%
  select(species, starts_with("bill")) %>%
  drop_na() %>%
  mutate(species = as.character(species)) %>%
  mutate(species = case_when(
    species == "Adelie" ~ "Adélie",
    TRUE ~ species
  )) %>%
  mutate(species = as.factor(species))

# Prep penguins for k-means:
pb_nospecies <- pb_species %>%
  select(-species) %>%
  recipe() %>%
  step_normalize(all_numeric()) %>%
  prep() %>%
  juice()

# Perform k-means on penguin bill dimensions (k = 3, w/20 centroid starts)

# Save augmented data
set.seed(100)
pb_clust <-
  pb_nospecies %>%
  kmeans(centers = 3, nstart = 20) %>%
  broom::augment(pb_species)

# Get counts in each cluster by species
pb_clust_n <- pb_clust %>%
  count(species, .cluster) %>%
  pivot_wider(names_from = species, values_from = n, names_sort = TRUE) %>%
  arrange(.cluster) %>%
  replace_na(list(`Adelie` = 0))

### Iris k-means
# Iris: petal length vs. petal width

# Select only petal dimensions:
ip_species <- iris %>%
  select(Species, starts_with("Petal"))

# Remove species factor & scale petal dimensions:
ip_nospecies <- ip_species %>%
  select(-Species) %>%
  recipe() %>%
  step_normalize(all_numeric()) %>%
  prep() %>%
  juice()

# Perform k-means on iris petal dimensions (k = 3, w/20 centroid starts):
set.seed(100)
ip_clust <-
  ip_nospecies %>%
  kmeans(centers = 3, nstart = 20) %>%
  broom::augment(ip_species)

# Get iris counts in each cluster by species:
ip_clust_n <- ip_clust %>%
  count(Species, .cluster) %>%
  pivot_wider(names_from = Species, values_from = n, names_sort = TRUE) %>%
  arrange(.cluster) %>%
  replace_na(list(`setosa` = 0, `versicolor` = 0, `virginica` = 0))


# Plot penguin k-means clusters:
# make a base plot b/c https://github.com/plotly/plotly.R/issues/1942
pb_kmeans_base <-
  pb_clust %>%
  ggplot(aes(x = bill_length_mm, y = bill_depth_mm)) +
  scale_color_paletteer_d("colorblindr::OkabeIto") +
  scale_fill_paletteer_d("colorblindr::OkabeIto") +
  scale_x_continuous(limits = c(30, 60),
                     breaks = c(30, 40, 50, 60)) +
  theme(legend.position = "bottom",
        panel.border = element_rect(fill = NA, color = "gray70")) +
  labs(x = "Bill length (mm)",
       y = "Bill depth (mm)",
       color = "Species")

# Plot iris k-means clusters:
# make a base plot b/c https://github.com/plotly/plotly.R/issues/1942
ip_kmeans_base <-
  ip_clust %>%
  ggplot(aes(x = Petal.Length, y = Petal.Width)) +
  theme(legend.position = "bottom",
        panel.border = element_rect(fill = NA, color = "gray70")) +
  scale_color_manual(values = c("gray70","gray50","black"))  +
  labs(x = "Petal length (cm)",
       y = "Petal width (cm)")

pb_tooltip <- str_c("Cluster: ", pb_clust$.cluster,
                      "\n Species: ", pb_clust$species,
                      "\n Bill length (mm): ", pb_clust$bill_length_mm,
                      "\n Bill depth (mm) : ", pb_clust$bill_depth_mm)

pb_kmeans_int <- pb_kmeans_base +
  geom_text_interactive(aes(
    label = .cluster,
    color = species,
    tooltip = pb_tooltip,
    data_id = species),
    size = 3,
    check_overlap = TRUE)

ip_tooltip <- str_c("Cluster: ", ip_clust$.cluster,
                      "\n Species: ", ip_clust$Species,
                      "\n Petal Length (cm): ", ip_clust$Petal.Length,
                      "\n Petal Width (cm): ", ip_clust$Petal.Width)

ip_kmeans_int <- ip_kmeans_base +
  geom_text_interactive(aes(
    label = .cluster,
    color = Species,
    tooltip = ip_tooltip,
    data_id = Species),
    size = 3,
    check_overlap = TRUE)

girafe(code = print(pb_kmeans_int + ip_kmeans_int),
       width_svg = 6,
       height_svg = 3,
       options = list(
         opts_hover_inv(css = "opacity:0.1;"),
         opts_hover(css = "opacity:1; filter: brightness(90%);")
       ))
```
:::
:::

## Kristen: Why has it been so popular?

::: columns
::: column
-   Penguins! (rare to see IRL, but so easy to see & say)

-   Understandable structural measurements

-   [Open license](https://allisonhorst.github.io/palmerpenguins/LICENSE.html) underlying the data

-   Ecology of species is relevant to global climate change
:::

::: column
![Penguin bills art by Allison Horst](https://allisonhorst.github.io/palmerpenguins/reference/figures/culmen_depth.png){alt="Penguin bills art by Allison Horst" fig-alt="Penguin bill length is measured from start to tip of bill, whereas bill depth is measured from top of bill to bottom" fig-align="right"}
:::
:::

::: notes
Kristen presents
:::

## From polar research project to global teaching product

::: columns
::: column
-   Antarctic data should be available to the world - no geopolitical boundaries

-   When you collect data as a scientist, you are collecting knowledge that you may not even imagine how it will be used
:::

::: column
![Map of Antarctica by Kristen Gorman](img/antarctic-modis.jpg){fig-alt="Map of Antarctica by Kristen Gorman" fig-align="center"} <!--[![Map of Antarctica by R. M. Chapin](img/map.jpg){fig-alt="Map of Antarctica by R. M. Chapin" fig-align="center"}](https://www.flickr.com/photos/crossettlibrary/6941620363)-->
:::
:::

## Thank you!

What questions do you have?

We wish to thank:

-   Marty Downs at LTER

-   Hadley Wickham (RStudio)

-   Laura Navarro for [this plot](https://lauranavarroviz.wordpress.com/2020/08/01/palmer-penguins/)

-   Cédric Scherer for [this plot](https://www.behance.net/gallery/101517403/Bill-Dimensions-of-Penguins)

-   Silvia Canelón for her [great tutorial](https://www.silviacanelon.com/talk/2020-tour-of-the-tidyverse/)

-   The #rstats community for blogging, tweeting, visualizing, analyzing, teaching, and sharing about `palmerpenguins`