{ggdims}

class: inverse, bottom
background-image: url(https://plus.unsplash.com/premium_photo-1686050878751-89499d28d153?ixlib=rb-4.1.0&ixid=M3wxMjA3fDB8MHxzZWFyY2h8MXx8bmlnaHRza3l8ZW58MHx8MHx8fDA%3D&auto=format&fit=crop&q=60&w=500)
background-size: cover

# .Large[{ggdims}]
## .small[Toward a viz- and code-first hitchhiker’s guide to dimensionality reduction]
## .small[github.com/EvaMaeRey/ggdims]
#### .tiny[Gina (Evangeline) Reynolds | 2025-11-14 | ASA, COWY, Image credit: Ales Krivec, Upsplash]

/* Change font size for a specific class (e.g., for larger text) */
.large-text {
  font-size: 1.5em;
}

/* Change font size for code blocks */
.remark-code {
  font-size: 0.9em;
}

/* Change font size for inline code */
.remark-inline-code {
  font-size: 0.9em;
}

</style>

---

### cliffnotes for talk

---

This pattern for ggplot2 extension ...

``` r
library(ggplot2)

ggplot(data = my_high_dimensional_data) + 
  aes(dims = dims(var1:var200, var205)) +      
  geom_dim_red_technique()  # geom_pca(), geom_tsne            
```

... isn't a pattern* ...

And maybe it should be.

\* or is it and I don't know that?

---

`aes(dims = dims(var1:var200, var205))`

![](images/clipboard-3430965112.png)

---
class: inverse, middle, center

{ggdims} explores viz- and code- first dimensionality reduction

- ~~A 3-week-old github repo/package~~

Rest of talk: how we got here (a longer time coming)

---
class: inverse, middle, center

# ggplot2 (2007) lets you intuitively translate variables to visual representation.

---

`gplot(my_data) +`

`aes(x = age) + `

`aes(color = sex) +`

`aes(size = employment status) +`

`geom_line()`

---
class: inverse

# But... ggplot2 allows individual-variable- 
--
to- individual-visual-channel mapping

---
class: inverse

# A problem with 1-to-1:

## Doesn't lend itself easily to visualizations in the world of dimension reduction (e.g. PCA, t-SNE, UMAP)

???

## However, in ggplot2 these specifications are individual-variable-to-individual-visual-channel which does not lend itself easily to visualizations in the world of dimension reduction (e.g. PCA, T-sne, UMAP).

---

``` r
library(ggplot2)

ggplot(data = my_high_dimensional_data) + 
  aes(dims = dims(var1:var200, var205)) +   # <<   
  geom_dimred_technique()  # e.g. geom_pca(), geom_tsne(), geom_umap()             
```

---

> "You really should 'algorithm' first, and then use the results in a data-viz."

(algorithm used as a verb here)

---

# 2016
https://distill.pub/2016/misread-tsne/

![](images/clipboard-667229542.png)

---

![](images/clipboard-2214898865.png)

---

Martin Wattenberg 'How to use t-SNE effectively' 2016:

> Paraphrase: 'it's worth taking the time to get a feel for what the algorithm does with some familiar, well understood distributions'

> Further paraphrase: Jump right in. Do some dimensionality reduction. Visualize.

---

![](images/clipboard-2170207951.png)

---

> "But maybe Wattenberg got it wrong..."

---

Leland McInnes "Bluffer's Guide"

> "I'm gonna try and give as much of a whirlwind tour of the 
entire field of dimention reduction as I can in the time available."

![](images/clipboard-3696481417.png)

---

![](images/clipboard-93834723.png)

> 'The similarities are greater than the differences' ?...

---

> Shouldn't we be able to visually fly (hitchhike) through trying out some of these techniques?

> Should we have to learn a different paradigm/syntax for every technique?

> What if the syntax were something already familiar (2D viz - ggplot2 scatterplot)?

---

``` r
*library(ggplot2)
```
]
 
.panel2-scatter-auto[

]

---
count: false

``` r
library(ggplot2)

# data
*ggplot(mtcars)
```
]
 
.panel2-scatter-auto[
![](asa-cowy-fall-2025_files/figure-html/scatter_auto_02_output-1.png)
]

---
count: false

``` r
library(ggplot2)

# data
ggplot(mtcars) +
# variable mapping
* aes(x = disp,
*     y = mpg)
```
]
 
.panel2-scatter-auto[
![](asa-cowy-fall-2025_files/figure-html/scatter_auto_03_output-1.png)
]

---
count: false

``` r
library(ggplot2)

# data
ggplot(mtcars) +
# variable mapping
  aes(x = disp,
      y = mpg) +
# layer
* geom_point()
```
]
 
.panel2-scatter-auto[
![](asa-cowy-fall-2025_files/figure-html/scatter_auto_04_output-1.png)
]

---

``` r
ggplot(mtcars) + # data
  aes(????)    # var mapping
  geom_pca()   # layer - options inside parentheses

ggplot(mtcars) + # data
  aes(????)    # var mapping
  geom_tsne()  # layer

ggplot(mtcars) + # data
  aes(????)    # var mapping
  geom_umap()  # layer
```

---

# Posit Conf speaker Orientation, July 18, 2025: "Give a Ted Talk like Sathiya"

Sathiya Manivannan, Center for Cardiovascular and Pulminary Research

![](images/clipboard-964179201.png)

---

Posit::conf(2025) Speaker Orientation

---

![](images/clipboard-2703820773.png)

---

> "It's easier to type thing which are smaller, but when things are getting bigger and bigger you prefer to click on buttons and work [with a GUI]." (made a shiny wrapper for {seurat})

But maybe there isn't a good ggplot2 extension to handle this?

---

![](images/clipboard-3469228869.png)

[2025-07-18 t-SNE 'labnotebook' experiment](https://evamaerey.github.io/mytidytuesday/2025-07-18-seurat_tsne_plot/seurat_tsne_plot.html)

---

## Nashville JSM Talk, "Developments in Statistical Graphics" panel, Tuesday, August 5

![](images/clipboard-2692987354.png)

---

![](legislative_dim_red.jpg)

[umap labnotebook experiment, 2025-08-19](https://evamaerey.github.io/mytidytuesday/2025-08-19-umap/umap.html)

---

``` r
ggplot(voting_data) + # data
  aes(????)    # var mapping
  geom_pca()   # layer - options inside

ggplot(voting_data) + # data
  aes(????)    # var mapping
  geom_tsne()  # layer

ggplot(voting_data) + # data
  aes(????)    # var mapping
  geom_umap()  # layer
```

---

### June-Sept 2024... `geom_pca_point()`

![](images/clipboard-1445857279.png)

---

``` r
palmerpenguins::penguins %>% 
  ggplot() + 
  aes(outcome = species, 
      pred1 = bill_length_mm, 
      pred2 = bill_depth_mm, 
      pred3 = flipper_length_mm) +
  geom_pca_point()  
```

---

![](images/clipboard-2464072259.png)

---

``` r
palmerpenguins::penguins %>% 
  ggplot() + 
  aes(outcome = species, 
      dims = vars(bill_length_mm,  
                  bill_depth_mm, 
                  flipper_length_mm)) +
  geom_pca_point()
```

---

tsne, umap - big data (lots of variables)

> "It's easier to type thing which are smaller, but when things are getting bigger and bigger you prefer to click on buttons and work [with a GUI]."

`aes(dims = vars(var1, var2, var3, var4, var5, var6)`

`aes(dims = dims(var1:var6))`

---

![](images/clipboard-3430965112.png)

---

``` r
library(ggplot2)

ggplot(data = my_high_dimensional_data) + 
  aes(dims = dims(var1:var200, var205)) + # or similar
  geom_reduction_technique() 
```

---

### utilities

``` r
ggdims::dims_listed()
ggdims::data_vars_unpack()
ggdims::dims_expand()  #<< expand var1:var20
```

---

### user-facing functions

``` r
ggdims::geom_pca()
ggdims::geom_tsne()
ggdims::geom_umap()
```

---

# Does it work?

---

# concise, consistent, 'graphical poems'

``` r
library(ggdims)

ggplot(unga_rcid_wide) + 
* aes(dims = dims(rc3:rc9147)) +
* geom_pca() +
  aes(fill = continent) +
  labs(title = "PCA") ->
unga_pca
```

---

# concise, consistent, 'graphical poems'

``` r
library(ggdims)

ggplot(unga_rcid_wide) + 
* aes(dims = dims(rc3:rc9147)) +
* geom_tsne() +
  aes(fill = continent) +
  labs(title = "t-SNE") ->
unga_tsne
```

---

# concise, consistent, 'graphical poems'

``` r
library(ggdims)

ggplot(unga_rcid_wide) + 
* aes(dims = dims(rc3:rc9147)) +
* geom_umap() +
  aes(fill = continent) +
  labs(title = "UMAP") ->
unga_umap
```

---

``` r
library(patchwork)

unga_pca + unga_tsne + unga_umap + 
  plot_layout(guides = "collect") + 
  plot_annotation(title = "UN General Assembly Role Call Voting")
```

![](asa-cowy-fall-2025_files/figure-html/unnamed-chunk-16-1.png)

---

![](legislative_dim_red.jpg)

---

How does it work? (for extenders meetup) Under the hood:

``` r
ggplot(my_data) + 
  aes(dims = dims(var1:var3, var7)) + 
  dims_expand() + 
  geom_umap0() 
```

# 1. `dims_expand`

- a. look at plot data and mapping

- b. extract implicitly referenced variables in mapping from plot data

var1:var3, var7 .... to var1, var2, var3, var7

---

# 1. `dims_expand`

- c. overwrite dims mapping with expression that lists all implicitly reference vars (future implementation my be new list col var that contains all referenced vars).

`dims = dims(var1:var3, var7)`

`dims = dims_listed(var1, var2, var3, var7)`

---
  
# 2. helper `geom_umap0` uses overwritten dim mapping for dimension reduction, taking input vars and computing x and y based on type dimred

---

# 3. We can write user-facing `geom_umap()`, bundles `dims_expand()` and helper `geom_umap0()`

``` r
geom_umap <- function(...){
  
  list(dims_expand(),
       geom_umap0(...))
  
}
```

---

### Getting:

``` r
ggplot(my_data) + 
  aes(dims = 
        dims(var1:var10, var15)) +
  geom_umap()
```

---

## Across-methods fly-over (pca, tsne, umap)

## Within-method fly-over? (tsne specifications, data)

---

![](images/clipboard-667229542.png)

---

![](images/clipboard-3992794559.png)

---

![](images/clipboard-2639177458.png)

---

![](asa-cowy-fall-2025_files/figure-html/unnamed-chunk-20-1.png)

---

``` r
*two_clusters
```
]
 
.panel2-pp2-auto[

```
          dim1         dim2 type
1   -0.5357287 -0.158254930    A
2   -0.4603359 -0.090857442    A
3   -0.3965400 -0.181554289    A
4   -0.4974581  0.106132971    A
5   -0.5729073 -0.007593519    A
6   -0.4308280 -0.045092558    A
7   -0.4197852 -0.012650899    A
8   -0.6454726 -0.008424184    A
9   -0.4513991 -0.214707887    A
10  -0.3131011 -0.020732053    A
11  -0.5635106  0.200063307    A
12  -0.6802282  0.113228751    A
13  -0.4671671 -0.027005700    A
14  -0.4593054 -0.068642035    A
15  -0.6159025  0.139260603    A
16  -0.5274706 -0.048721368    A
17  -0.4913888  0.091440641    A
18  -0.4746525  0.105385605    A
19  -0.6631743  0.134228505    A
20  -0.4540370  0.049525989    A
21  -0.5340863 -0.022069238    A
22  -0.4591989  0.068009894    A
23  -0.3724366 -0.024270429    A
24  -0.5338696  0.060912500    A
25  -0.4881712  0.066703762    A
26  -0.4479319  0.145153683    A
27  -0.4100356  0.036337216    A
28  -0.5366970 -0.029124000    A
29  -0.4113923 -0.023560498    A
30  -0.4349242 -0.025011625    A
31  -0.5059801  0.026554423    A
32  -0.3454215 -0.104671746    A
33  -0.6856915 -0.036302949    A
34  -0.4129278 -0.008105794    A
35  -0.2584186 -0.095262008    A
36  -0.5828491  0.020498686    A
37  -0.7487111  0.093079867    A
38  -0.3775321 -0.072232954    A
39  -0.5980546 -0.144940645    A
40  -0.4076318 -0.083561926    A
41  -0.3919189 -0.041298155    A
42  -0.6375004 -0.244553739    A
43  -0.6168814 -0.190197900    A
44  -0.5315187 -0.062031195    A
45  -0.5145287 -0.046462443    A
46  -0.4160438  0.197317704    A
47  -0.3931976 -0.041479515    A
48  -0.6571367 -0.046417266    A
49  -0.5226180 -0.120945937    A
50  -0.3934646  0.168484206    A
51  -0.4833463 -0.014768899    A
52  -0.5692205  0.044248820    A
53  -0.4926059  0.033500087    A
54  -0.3977887 -0.083249030    A
55  -0.2947652 -0.140350732    A
56  -0.6452856  0.094217235    A
57  -0.5158384 -0.189577055    A
58  -0.4925182 -0.087128492    A
59  -0.6553099 -0.017168386    A
60  -0.4006728 -0.222864041    A
61  -0.4146802  0.129645996    A
62  -0.4074571 -0.003922020    A
63  -0.5670353 -0.096400143    A
64  -0.4133183 -0.029236775    A
65  -0.5634832  0.127548206    A
66  -0.5671065 -0.129206785    A
67  -0.4521314  0.058483023    A
68  -0.3951574 -0.132199846    A
69  -0.2594366  0.063479376    A
70  -0.3845447  0.048875937    A
71  -0.4612730  0.029444030    A
72  -0.3541722  0.065608719    A
73  -0.3481411 -0.225318419    A
74  -0.4471973  0.108038841    A
75  -0.4235374  0.042559782    A
76  -0.4922954  0.081371883    A
77  -0.4545600 -0.053815694    A
78  -0.5857024 -0.037144865    A
79  -0.5018638 -0.038791169    A
80  -0.4938678 -0.059709975    A
81  -0.3750402  0.010759252    A
82  -0.4374003 -0.055493015    A
83  -0.5234888  0.103608133    A
84  -0.5619091  0.062872984    A
85  -0.4495126 -0.030400778    A
86  -0.3325314  0.012240477    A
87  -0.3489993 -0.092286488    A
88  -0.2913570 -0.086854253    A
89  -0.5642657 -0.017283381    A
90  -0.6731230 -0.072955045    A
91  -0.5319685 -0.099023766    A
92  -0.6504516  0.288148443    A
93  -0.6074628  0.046699775    A
94  -0.5617616 -0.007106916    A
95  -0.5260821 -0.098312730    A
96  -0.5227355  0.004991032    A
97  -0.5499129  0.163495761    A
98  -0.3925901 -0.104594754    A
99  -0.4356371  0.044225388    A
100 -0.4654224  0.070153340    A
101 -0.4187834 -0.186017354    A
102  0.5606903  0.135705951    B
103  0.4797486 -0.055312440    B
104  0.4183161 -0.081727621    B
105  0.6608417 -0.057504167    B
106  0.4696202  0.038560367    B
107  0.5108098 -0.055498693    B
108  0.4883599 -0.026284564    B
109  0.4410329 -0.001418396    B
110  0.4625084  0.079891023    B
111  0.3538451  0.028688468    B
112  0.3674274  0.035336611    B
113  0.4891274 -0.159566779    B
114  0.5566714  0.109369754    B
115  0.4966641  0.004084352    B
116  0.6231104 -0.041932167    B
117  0.5384671 -0.013715868    B
118  0.7544933  0.067691117    B
119  0.5334281  0.106882091    B
120  0.5845310  0.020439962    B
121  0.5000153  0.101844066    B
122  0.4748349  0.032864464    B
123  0.3542561 -0.067419903    B
124  0.4568676  0.172067517    B
125  0.2951672  0.181870248    B
126  0.5274469  0.034762617    B
127  0.3738903 -0.089173900    B
128  0.6319767  0.013811068    B
129  0.4758679 -0.112147293    B
130  0.2392581  0.019593448    B
131  0.5963978  0.058669812    B
132  0.5795330  0.059367882    B
133  0.4463065  0.045962269    B
134  0.4762261 -0.187948212    B
135  0.3813057 -0.082105191    B
136  0.5990435 -0.032782405    B
137  0.4256180 -0.156611906    B
138  0.6521080 -0.119362165    B
139  0.4919072 -0.058501426    B
140  0.5210185  0.140633282    B
141  0.5962070 -0.064638501    B
142  0.7162381 -0.322461265    B
143  0.5825375  0.102024799    B
144  0.4554914 -0.254290157    B
145  0.6458607  0.041606003    B
146  0.4521540 -0.015526346    B
147  0.3281885  0.076548531    B
148  0.5946617 -0.064311903    B
149  0.6148766 -0.143126072    B
150  0.5905803  0.099651771    B
151  0.4898780  0.057742666    B
152  0.3730359  0.013030242    B
153  0.3798805  0.155704161    B
154  0.4859656 -0.151298359    B
155  0.6868888  0.090086722    B
156  0.3912322 -0.043332386    B
157  0.4152529 -0.108574521    B
158  0.5099164  0.051162112    B
159  0.4358920  0.047650274    B
160  0.5475221  0.015635954    B
161  0.4009169  0.071916454    B
162  0.6612316  0.191754664    B
163  0.5222793  0.151908888    B
164  0.4604178 -0.084217285    B
165  0.6336940  0.008370056    B
166  0.5569161 -0.095242013    B
167  0.3630500 -0.046647038    B
168  0.5259380  0.011356280    B
169  0.6010525 -0.011316699    B
170  0.4744200  0.051207119    B
171  0.5730595 -0.137777461    B
172  0.6183336  0.141974631    B
173  0.5196704 -0.005082690    B
174  0.4003640 -0.068865048    B
175  0.4272597  0.005734501    B
176  0.4941395  0.148361504    B
177  0.5752577 -0.013311538    B
178  0.4736307 -0.054618941    B
179  0.3304741 -0.100005928    B
180  0.5609077  0.041335106    B
181  0.5010521 -0.023998367    B
182  0.5115131 -0.013954765    B
183  0.2969717  0.146663327    B
184  0.4235824  0.010328176    B
185  0.4767219  0.083918737    B
186  0.4321685 -0.034251659    B
187  0.5426292  0.007279260    B
188  0.5737919 -0.005723749    B
189  0.5357621  0.163366253    B
190  0.4604271 -0.030695260    B
191  0.7259624  0.046748228    B
192  0.3803555  0.044444520    B
193  0.5401534 -0.013522739    B
194  0.4126256  0.105441312    B
195  0.5047353  0.037931225    B
196  0.4851634 -0.140709581    B
197  0.4321986 -0.017397553    B
198  0.3771970 -0.037598042    B
199  0.4394506  0.049680431    B
200  0.4491798  0.061752594    B
201  0.5035890 -0.140302578    B
202  0.5967145  0.027716646    B
```
]

---
count: false

``` r
two_clusters |>
* ggplot()
```
]
 
.panel2-pp2-auto[
![](asa-cowy-fall-2025_files/figure-html/pp2_auto_02_output-1.png)
]

---
count: false

``` r
two_clusters |>
  ggplot() +
* aes(dims = dims(dim1:dim2))
```
]
 
.panel2-pp2-auto[
![](asa-cowy-fall-2025_files/figure-html/pp2_auto_03_output-1.png)
]

---
count: false

``` r
two_clusters |>
  ggplot() +
  aes(dims = dims(dim1:dim2)) +
* geom_tsne(perplexity = 2)
```
]
 
.panel2-pp2-auto[
![](asa-cowy-fall-2025_files/figure-html/pp2_auto_04_output-1.png)
]

---
count: false

``` r
two_clusters |>
  ggplot() +
  aes(dims = dims(dim1:dim2)) +
  geom_tsne(perplexity = 2) +
* labs(title = "perplexity = 2")
```
]
 
.panel2-pp2-auto[
![](asa-cowy-fall-2025_files/figure-html/pp2_auto_05_output-1.png)
]

---
count: false

``` r
two_clusters |>
  ggplot() +
  aes(dims = dims(dim1:dim2)) +
  geom_tsne(perplexity = 2) +
  labs(title = "perplexity = 2") +
* aes(fill = type)
```
]
 
.panel2-pp2-auto[
![](asa-cowy-fall-2025_files/figure-html/pp2_auto_06_output-1.png)
]

---
count: false

``` r
two_clusters |>
  ggplot() +
  aes(dims = dims(dim1:dim2)) +
  geom_tsne(perplexity = 2) +
  labs(title = "perplexity = 2") +
  aes(fill = type) ->
*pp2; pp2
```
]
 
.panel2-pp2-auto[
![](asa-cowy-fall-2025_files/figure-html/pp2_auto_07_output-1.png)
]

---

``` r
two_clusters |>
  ggplot() +
  aes(dims = dims(dim1:dim2)) +
  geom_tsne(perplexity = 5) +
  labs(title = "perplexity = 5") +
  aes(fill = type) ->
pp5; pp5
```
]
 
.panel2-pp5-2[
![](asa-cowy-fall-2025_files/figure-html/pp5_2_01_output-1.png)
]

---
count: false

---

``` r
two_clusters |>
  ggplot() +
  aes(dims = dims(dim1:dim2)) +
  geom_tsne(perplexity = 30) +
  labs(title = "perplexity = 30") +
  aes(fill = type) ->
pp30; pp30
```
]
 
.panel2-pp30-2[
![](asa-cowy-fall-2025_files/figure-html/pp30_2_01_output-1.png)
]

---
count: false

---

``` r
two_clusters |>
  ggplot() +
  aes(dims = dims(dim1:dim2)) +
  geom_tsne(perplexity = 50) +
  labs(title = "perplexity = 50") +
  aes(fill = type) ->
pp50; pp50
```
]
 
.panel2-pp50-2[
![](asa-cowy-fall-2025_files/figure-html/pp50_2_01_output-1.png)
]

---
count: false

---

``` r
two_clusters |>
  ggplot() +
  aes(dims = dims(dim1:dim2)) +
  geom_tsne(perplexity = 100) +
  labs(title = "perplexity = 100") +
  aes(fill = type) ->
pp100; pp100
```
]
 
.panel2-pp100-2[
![](asa-cowy-fall-2025_files/figure-html/pp100_2_01_output-1.png)
]

---
count: false

---

``` r
library(patchwork)

original + pp2 + pp5 + pp30 + pp50 + pp100
```

![](asa-cowy-fall-2025_files/figure-html/unnamed-chunk-21-1.png)

---

``` r
library(patchwork)

original + pp2 + pp5 + pp30 + pp50 + pp100 &
  guides(fill = "none") 
```

![](asa-cowy-fall-2025_files/figure-html/unnamed-chunk-22-1.png)

---

``` r
library(patchwork)

original + pp2 + pp5 + pp30 + pp50 + pp100 &
  guides(fill = "none") &
  theme_ggdims() 
```

![](asa-cowy-fall-2025_files/figure-html/unnamed-chunk-23-1.png)

---

``` r
library(patchwork)

original + pp2 + pp5 + pp30 + pp50 + pp100 & 
  guides(fill = "none") &
  theme_ggdims()  ->
two_clusters_5_perplexities_tsne
```

---

![](images/clipboard-2639177458.png)

---

``` r
two_clusters_5_perplexities_tsne & # contains all six of the 'hello world of tsne' specifications
  two_close_and_one_far #<< # change data
```

![](asa-cowy-fall-2025_files/figure-html/unnamed-chunk-25-1.png)

---

![](images/clipboard-1971882774.png)

---

``` r
two_clusters_5_perplexities_tsne & #contains all six of the 'hello world of tsne' specifications
  big_and_small_cluster # change data
```

![](asa-cowy-fall-2025_files/figure-html/unnamed-chunk-26-1.png)

---

![](images/clipboard-553036957.png)

---

``` r
two_clusters_5_perplexities_tsne  & # contains all six of the 'hello world of tsne' specifications
  random_noise & # change data
  aes(fill = I("navy"))
```

![](asa-cowy-fall-2025_files/figure-html/unnamed-chunk-27-1.png)

---

# Is this the right implementation?

Maybe not

# Are there some known fragilities

Yes

# Are you the right person to be writing things like `geom_umap`

Maybe not..

---
class: inverse, middle, center

Thank you!

---

### To Do

#### gg side: `aes(dims = dims())` and `unpack()` implementation

- ggpcp's approach (Heike)?
- ggswim approach (Teun and Rich, via June C.)?

- fundamentally, can `aes(dims = dims())` be totally compatible w/ ggplot2?

- ggplot2 allows new vars to be computed on-the-fly in aes, but classically from vectors (be they referenced from variables data or otherwise).

---

### To do

#### on the dimred side

- layer_pca/layer_tsne/layer_umap requirements (e.g. Cory, Joyce for PCA)
- borrow internals with tidymodels for consistency?

---

![](../mytidytuesday/2025-12-10-joss-umap-v-tidyverse/joss-gg-umap-2021.png)
---

@media print {
  .has-continuation {
    display: block;
  }
}

code.r.hljs.remark-code{
  position: relative;
  overflow-x: hidden;
}

code.r.hljs.remark-code:hover{
  overflow-x:visible;
  width: 500px;
  border-style: solid;
}

</style>