The Art of Design

.info-box.w-50.bg-white.f3[
These slides are viewed best by Chrome or Firefox and occasionally need to be refreshed if elements did not load properly. See <a href=emi-tanaka-talk.pdf>here for the PDF </a>.
]

---

count: false
class: fullscreen title-slide
background-image: url("images/art-01.png")
background-size: cover

.animate-text[
.monash-green2[.s[S].s[o].s[f].s[t].s[w].s[a].s[r].s[e]] & &emsp; .monash-ruby2[.s[E].s[x].s[p].s[e].s[r].s[i].s[m].s[e].s[n].s[t].s[a].s[l]] Design
]

]

.bottom_abs.w-100.bg-shadow.h-45.slideInUp.animated.pl3.delay-1s[

Presenter: *Emi Tanaka*

.f3[
Department of Econometrics and Business Statistics 
<img src="images/accessories/monash-stacked-reversed-white.png" height="100px"> 
 emi.tanaka@monash.edu &nbsp;&nbsp; @statsgen

Tue 26th Oct 2021 | R-Ladies Melbourne

]

```r
library(ggplot2)
library(mathart)
df1 <- harmonograph(A1 = 1, A2 = 1.5, A3 = 1, A4 = 1,
 d1 = 0.004, d2 = 0.0065, d3 = 0.008, d4 = 0.019,
 f1 = 3.001, f2 = 2, f3 = 3, f4 = 2,
 p1 = 0, p2 = pi/4, p3 = pi/2, p4 = 3*pi/2)
ggplot(df1, aes(x, y)) +
 geom_path(alpha = 0.4, size = 0.4, color = "#181818") +
 coord_equal() +
 xlim(min(df1$x) - 5, max(df1$x) + 1) +
 annotate("text", x = max(df1$x) + 0.2, y = max(df1$y) - 0.2, 
 label = "Harmonograph", color = "gray",
 family = "Atma", size = 6) + 
 theme_void()
```

]

---

.flex.h-100.bg-rladies-dark[
.w-75.white.pa3.f1[

.header.f1[ Now Showing]

.f2.pa3.lh-copy[

.box[Act I] **Software design** 
.f3[ Drawing faces under different .yellow[programming paradigms]] 
.box[Act II] **Experimental design** 
.f3[ Comparative experiments with `edibble`]

.border-box.bg-rladies.white.f3[
.center[
 These slides made using R powered by HTML/CSS/JS can be found at <a style="color:yellow!important" href="https://emitanaka.org/slides-RLadiesMelb2021">emitanaka.org/slides-RLadiesMelb2021</a> 
 All code to reproduce this slide can be found at <a style="color:yellow!important" href="https://github.com/emitanaka/slides-RLadiesMelb2021">github.com/emitanaka/slides-RLadiesMelb2021</a>

]]

]]
.w-25[
 
<img src="images/art-02.png" width="100%">
]]

```r
library(ggplot2)
library(mathart)
df2 <- fractal_fern(n = 250000,
 a = c(0, 0.85, 0.2, -0.15),
 b = c(0, 0.04, -0.26, 0.28),
 c = c(0, -0.04, 0.23, 0.26),
 d = c(0.16, 0.85, 0.22, 0.24),
 e = c(0, 0, 0, 0),
 f = c(0, 1.6, 1.6, 0.44),
 p = c(0.01, 0.85, 0.07, 0.07))
ggplot(df2, aes(x, y)) +
 geom_point(size = 0.03, alpha = 0.06, color = "white") +
 coord_equal() +
 labs(caption = "Fractal ferns") +
 theme_blankcanvas(margin_cm = 1) +
 theme(plot.caption = element_text(family = "Atma", size = 20, color = "gray"),
 plot.background = element_rect(fill = "transparent"),
 panel.background = element_rect(fill = "transparent"))
```

]

</div>

---

# .gray[Act I]

# Software Design

---

##

# Drawing faces under different .yellow[programming paradigms]

##

---

# Drawing a .yellow[happy] face .circle[1]

???

* To illustrate the different programming paradigms, I'm going to show different ways to design the software to draw faces

---

# Drawing a .yellow[happy] face .circle[1]

```r
library(grid)
# face shape
grid.circle(x = 0.5, y = 0.5, r = 0.5)
```
]
]
.w-50[
.f4[

<img src="images/happy-1-1.png" width="432" style="display: block; margin: auto;" />
]

]

---

# Drawing a .yellow[happy] face .circle[1]

```r
library(grid)
# face shape
grid.circle(x = 0.5, y = 0.5, r = 0.5)

# eyes
grid.circle(x = c(0.35, 0.65),
            y = c(0.6, 0.6),
            r = 0.05,
            gp = gpar(fill = "black"))
```
]
]
.w-50[
.f4[

<img src="images/happy-2-1.png" width="432" style="display: block; margin: auto;" />
]

]

---

# Drawing a .yellow[happy] face .circle[1]

```r
library(grid)
# face shape
grid.circle(x = 0.5, y = 0.5, r = 0.5)

# eyes
grid.circle(x = c(0.35, 0.65),
            y = c(0.6, 0.6),
            r = 0.05,
            gp = gpar(fill = "black"))

# mouth
grid.curve(x1 = 0.4, y1 = 0.4,
           x2 = 0.6, y2 = 0.4,
           square = FALSE)
```
]
]
.w-50[
.f4[

<img src="images/happy-3-1.png" width="432" style="display: block; margin: auto;" />
]

]

---

# Drawing a .rladies-gray[sad] face .circle[1]

???

* Switch to drawing a sad face

---

# Drawing a .rladies-gray[sad] face .circle[1]

```r
library(grid)
# face shape
grid.circle(x = 0.5, y = 0.5, r = 0.5)

# eyes
grid.circle(x = c(0.35, 0.65),
            y = c(0.6, 0.6),
            r = 0.05,
            gp = gpar(fill = "black"))

# mouth
grid.curve(x1 = 0.4, y1 = 0.4,
           x2 = 0.6, y2 = 0.4,
           square = FALSE,
*          curvature = -1)
```
]
]
.w-50[
.f4[

.animated.delay-1s.zoomIn[
<img src="images/sad-1-1.png" width="432" style="display: block; margin: auto;" />
]
]
]

]

---

# Drawing faces .circle[1] .yellow[*Imperative programming*] style

```r
library(grid)
grid.circle(x = 0.5, y = 0.5, r = 0.5)
grid.circle(x = c(0.35, 0.65),
 y = c(0.6, 0.6),
 r = 0.05,
 gp = gpar(fill = "black"))
grid.curve(x1 = 0.4, y1 = 0.4,
 x2 = 0.6, y2 = 0.4,
 square = FALSE)
```
.animated.flash[
<img src="images/happy-1.png" width="151.2" style="display: block; margin: auto;" />
]]
]
.w-50[
.f4[

```r
grid.circle(x = 0.5, y = 0.5, r = 0.5)
grid.circle(x = c(0.35, 0.66),
 y = c(0.6, 0.6),
 r = 0.05,
 gp = gpar(fill = "black"))
grid.curve(x1 = 0.4, y1 = 0.4,
 x2 = 0.6, y2 = 0.4,
 square = FALSE,
 curvature = -1)
```
.animated.flash[
<img src="images/sad-1.png" width="151.2" style="display: block; margin: auto;" />
]]

]

.absolute.left-3.bottom-0[
.info-box.w-70[
**Imperative programming** is a programming paradigm that uses statements to change a program's state.
.fr[&mdash;Wikipedia]

Think of this as _**"instructions meant for the computer"**_

]]

---

# Drawing faces .circle[2] .yellow[*Functional programming*] style

```r
face1 <- function() {
 grid::grid.circle(x = 0.5, y = 0.5, r = 0.5)
 grid::grid.circle(x = c(0.35, 0.65),
 y = c(0.6, 0.6),
 r = 0.05,
 gp = gpar(fill = "black"))
 grid::grid.curve(x1 = 0.4, y1 = 0.4,
 x2 = 0.6, y2 = 0.4,
 square = FALSE)
}

face2 <- function() {
 grid::grid.circle(x = 0.5, y = 0.5, r = 0.5)
 grid::grid.circle(x = c(0.35, 0.65),
 y = c(0.6, 0.6),
 r = 0.05,
 gp = gpar(fill = "black"))
 grid::grid.curve(x1 = 0.4, y1 = 0.4,
 x2 = 0.6, y2 = 0.4,
 square = FALSE,
 curvature = -1)
}
```

]
.w-40[
.info-box[
**Functional programming** is a declarative programming paradigm where programs are constructed by applying and composing functions.
.fr[&mdash;Wikipedia]
]

]
]

???

* Wrap the previous imperative commands as a function

---

# Drawing faces .circle[2] .yellow[*Functional programming*] style

]
.w-40[
.info-box[
**Functional programming** is a declarative programming paradigm where programs are constructed by applying and composing functions.
.fr[&mdash;Wikipedia]
]
.flex[
.w-50.pr2[

```r
face1()
```

]
.w-50[

]

]
]
]

---

# Drawing faces .circle[2] .yellow[*Functional programming*] style

]
.w-40[
.info-box[
**Functional programming** is a declarative programming paradigm where programs are constructed by applying and composing functions.
.fr[&mdash;Wikipedia]
]
.flex[
.w-50.pr2[

```r
face1()
```

]
.w-50[

```r
face2()
```

]
]
]
]

---

# Drawing faces .circle[2] .yellow[*Functional programming*] style

]
.w-40[
.info-box[
**Functional programming** is a declarative programming paradigm where programs are constructed by applying and composing functions.
.fr[&mdash;Wikipedia]
]
.flex[
.w-50.pr2[

```r
face1()
```

```r
face1()
```

]
.w-50[

```r
face2()
```

<img src="images/sad-fn-1.png" width="68.4" style="display: block; margin: auto;" />
]
]
]
]

???

* I can repeatedly call the function
* Drawing happy or sad faces are easier now

---

# Drawing faces .circle[2] .yellow[*Functional programming*] style

]
.w-40[
.info-box[
**Functional programming** is a declarative programming paradigm where programs are constructed by applying and composing functions.
.fr[&mdash;Wikipedia]
]
.flex[
.w-50.pr2[

```r
face1()
```

```r
face1()
```

]
.w-50[

```r
face2()
```

```r
face1()
```

<img src="images/happy-fn-1.png" width="68.4" style="display: block; margin: auto;" />
]

]

]
]

---

# Drawing faces .circle[3] Human-centered design

.info-box.w-90[
**Syntactic sugar** means using function name or syntax in a programming language that is designed to make things **_easier to read or to express for humans_**.
]
]
]
.w-50[

]

---

# Drawing faces .circle[3] Human-centered design

.info-box.w-90[
**Syntactic sugar** means using function name or syntax in a programming language that is designed to make things **_easier to read or to express for humans_**.
]
]
]
.w-50[
.flex[
.w-33.pr2[

```r
face1()
```

<img src="images/happy-fn-1.png" width="144" style="display: block; margin: auto;" />
]
.w-33.pr2[

```r
face2()
```

<img src="images/sad-fn-1.png" width="144" style="display: block; margin: auto;" />
]
.w-33.pr2[
```r
face3()
```
.center.f-headline[
?

]
]
]
]

]

---

# Drawing faces .circle[3] Human-centered design

```r
face1()
```

<img src="images/happy-fn-1.png" width="144" style="display: block; margin: auto;" />
]
.w-33.pr2[

```r
face2()
```

<img src="images/sad-fn-1.png" width="144" style="display: block; margin: auto;" />
]
.w-33.pr2[
```r
face3()
```
.center.f-headline[
?
]
]
]

Alternative function names:

]
.w-33.pr2[
```r
face_angry()
```

]
]
]

]

---

.f1[**_You can't_** without writing a whole new function that describes the entire face from scratch.]

---

`remotes::install_github("emitanaka/portrait")`
.f-headline[
`library(portrait)`
]

Note: the above package was made for demonstration purpose only for this talk.

---

# Drawing faces .circle[4] Rethinking function arguments as facial parts
--

* Let's **_reframe_** how we think
* A face is made up of:
  * eyes
  * mouth
  * shape

]
.w-50[
{{content}}

]

* We can consider a design like below instead:

```r
library(portrait)
face(eyes = "googly",
     mouth = "smile",
     shape = "round")
```

<img src="images/face-googly-1.png" width="216" style="display: block; margin: auto;" />
{{content}}

* But what about hair and nose?

---

# Drawing faces .circle[4] Rethinking function arguments as facial parts

* Let's **_reframe_** how we think
* A face is made up of:
 * eyes
 * mouth
 * shape
 * **hair** 
 * **nose**

]
.w-50[
* Adding more arguments:

```r
library(portrait)
face(eyes = "googly",
     mouth = "smile",
     shape = "round",
*    hair = "mohawk",
*    nose = "simple")
```

<img src="images/face-googly-2-1.png" width="216" style="display: block; margin: auto;" />
{{content}}

]

---

---

# Drawing faces .circle[5] .yellow[*Object-oriented programming*] style

.info-box.w-70[
**Object-oriented programming** (OOP) is a programming paradigm based on objects that have certain attributes and behaviours
]

* Rethink everything as an **_object_**

---

.flex.h-100[
.w-30.pr3[

```r
library(portrait)
*f <- face()

str(f)
```

```
## Portrait contains:
## - shape
## - eyes
## - mouth
```

]
.w-30.pr3[
.output-box.w-100[

```r
print(f)
```

<img src="images/face1-1.png" width="144" style="display: block; margin: auto;" />
]]
.w-40[
.summary-box.w-100[
* Function creates an object that contains a "standard" smiley face
* Only draws the face upon print
]
]
]

---

.flex.h-100[
.w-30.pr3[

```r
library(portrait)
f <- face() +
* cat_shape()

str(f)
```

```
## Portrait contains:
## - shape
## - eyes
## - mouth
```

]
.w-30.pr3[
.output-box.w-100[

```r
print(f)
```

<img src="images/face2-1.png" width="144" style="display: block; margin: auto;" />
]]
.w-40[
.summary-box.w-100[
* .black[`cat_shape`] function modifies the shape information within the object
]
]
]

---

.flex.h-100[
.w-30.pr3[

```r
library(portrait)
f <- face() +
 cat_shape() +
* cat_eyes()

str(f)
```

```
## Portrait contains:
## - shape
## - eyes
## - mouth
```

]
.w-30.pr3[
.output-box.w-100[

```r
print(f)
```

<img src="images/face3-1.png" width="144" style="display: block; margin: auto;" />
]]
.w-40[
.summary-box.w-100[
* .black[`cat_eyes`] function modifies the eye information within the object
]
]
]

---

.flex.h-100[
.w-30.pr3[

```r
library(portrait)
f <- face() +
 cat_shape() +
 cat_eyes() +
* cat_nose()

str(f)
```

```
## Portrait contains:
## - shape
## - eyes
## - mouth
## - nose
```

]
.w-30.pr3[
.output-box.w-100[

```r
print(f)
```

<img src="images/face4-1.png" width="144" style="display: block; margin: auto;" />
]]
.w-40[
.summary-box.w-100[
* .black[`cat_nose`] function adds nose information
]
]]

---

.flex.h-100[
.w-30.pr3[

```r
library(portrait)
f <- face() +
 cat_shape() +
 cat_eyes() +
 cat_nose() +
* cat_whiskers()

str(f)
```

```
## Portrait contains:
## - shape
## - eyes
## - mouth
## - nose
## - whiskers
```

]
.w-30.pr3[
.output-box.w-100[

```r
print(f)
```

<img src="images/face5-1.png" width="144" style="display: block; margin: auto;" />
]
]
.w-40[
.summary-box.w-100[
* .black[`cat_whiskers`] function adds whiskers information
]

]
]

---

.flex.h-100[
.w-30.pr3[

```r
library(portrait)
f <- face() +
* dog_shape() +
* cat_eyes(fill = "red") +
 cat_nose() +
 cat_whiskers()

str(f)
```

```
## Portrait contains:
## - shape
## - eyes
## - mouth
## - nose
## - whiskers
```

]
.w-30.pr3[
.output-box.w-100[

```r
print(f)
```

<img src="images/face6-1.png" width="144" style="display: block; margin: auto;" />
]
]
.w-40[
.summary-box.w-100[
* Mix-and-match functions to make other faces
* The functions are modular so can be replaced by user's own functions
]

]
]

---

--

.animated.slideInLeft[
# .f1[Finite number of functions to draw _infinite_ possible _incomplete_ and complete faces]

]

---

# .gray[Act II]

# Experimental Design

---

# .f1[ Classical "named" experimental designs]

---

background-color: #e3e3e3

# Classical split-plot design

.context-box[
Study of **two irrigation methods** and **two fertilizer brands** on the yields of a crop.
]

???

To begin with let's have a look at this classical split-plot design.
The context of this experiment is that it's a study on the yields of a crop with two irrigation methods, either irrigated or rain-fed, and two fertilizer brands.

So in order to conduct this study, the experimental resources that is available to us is this land where there are four fields, which I'm going to referring to these as the wholeplot.

---

background-color: #e3e3e3
count: false

# Classical split-plot design

.context-box[
Study of **two irrigation methods** and **two fertilizer brands** on the yields of a crop.
]

???

Now for the irrigation method, we have a restriction such that only one irrigation method can be applied per wholeplot. So what you see here, where we got the irrigated wholeplots on the edges and the rainfed wholeplots in the middle, is just one possible randomisation of the irrigation method on the whole plots.

---

background-color: #e3e3e3
count: false

# Classical split-plot design

.context-box[
Study of **two irrigation methods** and **two fertilizer brands** on the yields of a crop.
]

???

Then what we do next is to split the wholeplot into two subplots, which means that we have 8 subplots in total.

---

# Classical split-plot design

.context-box[
Study of **two irrigation methods** and **two fertilizer brands** on the yields of a crop.
]

???

Now the fertilizer can be applied to each subplots independently. We don't want to confound the fertilizer brands to the wholeplot, so for each wholeplot, we randomly apply one fertilizer brand to one subplot and the the remaining subplot receives the other fertilizer brand. What you see here is one possible restricted randomisation of fertilizer onto subplot.

---

background-color: #e3e3e3
count: false

# Classical split-plot design

.context-box[
Study of **two irrigation methods** and **two fertilizer brands** on the yields of a crop.
]

???

Putting together the two randomisation results, here we have our classical split plot design.

---

.lh-solid[
# .f-subheadline[CRAN Task View of Design of Experiments ] .f1[& Analysis of Experimental Data]
]

contains

# 📦 .yellow[109 R-packages ]

---

# `agricolae::design.split()`

**Split-plot design** for `$t = 2 \times 4$` treatments with `$2$` replication for each treatment

.overflow-scroll.h-60[

```
## $parameters
## $parameters$design
## [1] "split"
## 
## $parameters[[2]]
## [1] TRUE
## 
## $parameters$trt1
## [1] "I" "R"
## 
## $parameters$applied
## [1] "crd"
## 
## $parameters$r
## [1] 2 2
## 
## $parameters$serie
## [1] 2
## 
## $parameters$seed
## [1] -1888132024
## 
## $parameters$kinds
## [1] "Super-Duper"
## 
## 
## $book
##   plots splots r trt1 trt2
## 1   101      1 1    R    B
## 2   101      2 1    R    A
## 3   102      1 1    I    B
## 4   102      2 1    I    A
## 5   103      1 2    I    A
## 6   103      2 2    I    B
## 7   104      1 2    R    A
## 8   104      2 2    R    B
```
]

</div>

---

.f1[Common .rladies-purple-dark[**software design**] for computational systems for constructing experimental design:]

* Functions construct the experimental design from **_zero to its entirety_**
  * In another words, you can't have an intermediate construct of an experimental design
{{content}}

* Experimental units, observational units, blocks and allocation of treatment to units are often **_implicitly_ defined**
{{content}}

* The intention of what responses to record are **_not explicitly_ specified**

---

# Let's _reframe_ how we think

but first some terminology so we're on the same page

---

# A basic comparative experiment

.w-75[
* There are three components that are *necessary* to run a comparative experiment:
 * a set of `$\color{gray}{n}$` **experimental units** `$(\Omega)$`
 * a set of `$\color{gray}{t}$` **treatments** `$(\mathcal{T})$`
 * **allocation** of treatments to experimental units `$(D: \Omega \rightarrow \mathcal{T})$` or design matrix `$(\mathbf{X}_{n\times t})$`
* And decide on **observational units** `$(\Omega_o)$` which may or may not be the same as `$\Omega$`
{{content}}
]

* For the analysis of experimental, you additionally require:
  * the **record of responses** `$(\boldsymbol{Y})$` on observational units

---

# Terminology .f3.gray[modified versions of Bailey (2008)]

.flex[
.w-60.pr3[
.info-box.f3[
**Experimental unit** `$(\Omega)$` is the smallest unit that the treatment can be independently applied to.
]
.info-box.f3[
**Observational unit** `$(\Omega_o)$` is the smallest unit in which the response will be measured on. Not to be confused with observations  `$\boldsymbol{Y}$`.
]
.info-box.f3[
**Block**, also called **cluster**, is the unit that group some other units (e.g. experimental units) such that the units within the same block (cluster) are more alike (homogeneous).
]
]
.w-40[
.info-box.f3[
A **treatment** `$(\mathcal{T})$` is the entire description of what can be applied to an experimental unit.
]

.info-box.f3[
A **design** `$(D: \Omega \rightarrow \mathcal{T})$` is the allocation of treatments to units.
]

.info-box.f3[
A **plan** or **layout** is the design translated into actual units; randomisation is usually involved in the process.
]
]]

.footnote[
Bailey, R. (2008). Design of Comparative Experiments (Cambridge Series in Statistical and Probabilistic Mathematics). Cambridge: Cambridge University Press. doi:10.1017/CBO9780511611483
]

---

# Experimental Structures .f3.gray[as defined by Bailey (2008)]

.flex[
.w-50[
.info-box.f3[
**Unit structure** means meaningful ways of dividing up `$\Omega$` and `$\Omega_o$`.

For example:
* **Unstructured**
* **Blocking**
]
]
.w-50.pl3[
.info-box.f3[
 **Treatment Structure** means meaningful ways of dividing up `$\mathcal{T}$`.

For example:
* **Unstructured**: no grouping within `$\mathcal{T}$`
* **Factorial**: all combinations of at least two factors
* **Factorial + control**
]

]]

.footnote[
Bailey, R. (2008). Design of Comparative Experiments (Cambridge Series in Statistical and Probabilistic Mathematics). Cambridge: Cambridge University Press. doi:10.1017/CBO9780511611483
]

---

background-image: url("images/art-03.png")
background-size: 100% 100%
class: middle center

.animated.slideInLeft.f-headline[
**The Grammar of**]  
.animated.slideInRight.f-headline[
.rladies-purple-dark[**Experimental Designs**]]

```r
library(tidyverse)
library(mathart)
set.seed(2021)
df3 <- map_dfr(1:100, ~lissajous(a = runif(1, 0, 10), A = runif(1, 0, 1)), .id = "id")
ggplot(df3, aes(x, y)) +
 geom_path(color = "#027EB6", size = 0.25, lineend = "round", alpha = 0.07) +
 facet_wrap(~id, nrow = 8) +
 coord_equal() +
 labs(caption = "Lissajous curves") +
 theme_blankcanvas(margin_cm = 1) +
 theme(plot.caption = element_text(family = "Atma", size = 20, color = "gray"))
```

]

---

# Context is key in experimental design

.flex[
.w-33[
<img src="images/domain-expert-1.png" width="144" style="display: block; margin: auto;" />
Domain expert
]
.w-33[
<img src="images/statistican-1.png" width="144" style="display: block; margin: auto;" />
Statistician
]
.w-33[
<img src="images/technician-1.png" width="144" style="display: block; margin: auto;" />
Technician
]
]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

```r
library(edibble)
start_design("Wood water resistance")
```

]
.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6[

```
## Wood water resistance
```

]
<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Domain expert
]

<blockquote>
"I want to run an experiment to study the water resistance property of wood"
</blockquote>

]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

```r
library(edibble)
start_design("Wood water resistance") %>%
  set_units(board = 10,
            panel = nested_in(board, 4))
```

]

.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6.w-100[

```
## Wood water resistance
## └─board (10 levels)
##   └─panel (40 levels)
```

]
<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Domain expert
]

<blockquote>
"I have 10 boards that I can divide each into 4 small wood panels"
</blockquote>

]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

```r
library(edibble)
start_design("Wood water resistance") %>%
  set_units(board = 10,
            panel = nested_in(board, 4)) %>%
  set_trts(pretreatment = c("copper", "linseed"),
           stain = 4)
```

]
.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6.w-100[

```
## Wood water resistance
## ├─board (10 levels)
## │ └─panel (40 levels)
## ├─pretreatment (2 levels)
## └─stain (4 levels)
```

]
<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Domain expert
]

<blockquote>
"I want to test linseed oil and copper azole preservative as wood pretreatments and four types of stain."
</blockquote>

]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

]
.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6.w-100[

```
## Wood water resistance
## ├─board (10 levels)
## │ └─panel (40 levels)
## ├─pretreatment (2 levels)
## └─stain (4 levels)
```

]
<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Domain expert
]

<blockquote>
"The combination of the the treatment factors should be applied on each wood panel."
</blockquote>

]
]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

]
.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6.w-100[

```
## Wood water resistance
## ├─board (10 levels)
## │ └─panel (40 levels)
## ├─pretreatment (2 levels)
## └─stain (4 levels)
```

]

<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Domain expert
]

<blockquote>
"Oh wait, it's hard to apply pretreatment to small wood panels. We can only apply it to the board."
</blockquote>

]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

]
.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6.w-100[

```
## Wood water resistance
## ├─board (10 levels)
## │ └─panel (40 levels)
## ├─pretreatment (2 levels)
## └─stain (4 levels)
```

]
<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Domain expert
]

<blockquote>
"Oh wait, it's hard to apply pretreatment to small wood panels. We can only apply it to the board. The stain can be applied independently to wood panels."
</blockquote>

]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

```r
library(edibble)
set.seed(2021)
start_design("Wood water resistance") %>%
  set_units(board = 10,
            panel = nested_in(board, 4)) %>%
  set_trts(pretreatment = c("copper", "linseed"),
           stain = 4) %>%
  allocate_trts(pretreatment ~ board,
                       stain ~ panel) %>%
  randomise_trts() %>%
  serve_table()
```

]
.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6.w-100[

```
## # An edibble: 40 x 4
## board panel pretreatment stain
## <unit(10)> <unit(40)> <trt(2)> <trt(4)>
## 1 board1 panel1 copper stain4
## 2 board1 panel2 copper stain3
## 3 board1 panel3 copper stain1
## 4 board1 panel4 copper stain2
## 5 board2 panel5 linseed stain2
## 6 board2 panel6 linseed stain3
## 7 board2 panel7 linseed stain1
## 8 board2 panel8 linseed stain4
## 9 board3 panel9 copper stain1
## 10 board3 panel10 copper stain3
## # … with 30 more rows
```

]
<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Domain expert
]

<blockquote>
"Okay, looks good. I'll let the technician run the experiment then."
</blockquote>

]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

]
.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6.w-100[

]
<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Technician
]

<blockquote>
"What am I supposed to measure?"
</blockquote>

]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

]
.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6.w-100[

```
## # An edibble: 40 x 6
## board panel pretreatment stain swelling supplier
## <unit(10)> <unit(40)> <trt(2)> <trt(4)> <rcrd> <rcrd>
## 1 board1 panel1 copper stain4 ■ ■
## 2 board1 panel2 copper stain3 ■ x
## 3 board1 panel3 copper stain1 ■ x
## 4 board1 panel4 copper stain2 ■ x
## 5 board2 panel5 linseed stain2 ■ ■
## 6 board2 panel6 linseed stain3 ■ x
## 7 board2 panel7 linseed stain1 ■ x
## 8 board2 panel8 linseed stain4 ■ x
## 9 board3 panel9 copper stain1 ■ ■
## 10 board3 panel10 copper stain3 ■ x
## # … with 30 more rows
```

]
<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Domain expert
]

<blockquote class="f5">
"Oh follow <a href="https://link.springer.com/article/10.1007/s00107-014-0868-7">Kubojima & Yoshida (2015)</a> and measure diameter swelling. Remember to record wood supplier too."
</blockquote>

]

---

# Constructing experimental designs with `edibble`

.flex.h-100[
.w-50[
<img src="images/statistician2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Statistician
]

```r
library(edibble)
set.seed(2021)
des <- start_design("Wood water resistance") %>%
 set_units(board = 10,
 panel = nested_in(board, 4)) %>%
 set_trts(pretreatment = c("copper", "linseed"),
 stain = 4) %>%
 allocate_trts(pretreatment ~ board,
 stain ~ panel) %>%
 randomise_trts() %>%
 set_rcrds(panel = swelling,
 board = supplier) %>%
 expect_rcrds(swelling = to_be_numeric(with_value(between = c(0, 5)))) %>%
 serve_table()
```

]
.w-50.pl3.pr3.monash-bg-gray10[
.output-box.f6.w-100[

```r
export_design(des, "design.xlsx", overwrite = TRUE)
```

```
## Loading required package: openxlsx
```

```
## ✓ Wood water resistance has been written to 'design.xlsx'
```

]

<img src="images/domain-expert2-1.png" width="144" style="display: block; margin: auto;" />
.center[
Technician
]

<blockquote>
"Okay, I'll record the swelling. The values should be between 0 and 5 for that."
</blockquote>

]

---

# Wood water resistance experiment

* This experiment is based on the description by Kowalski and Potcner (2003) 
* The resulting design is a split-plot design... but we didn't need to know that to construct the experiment!

]]

.footnote[
Kowalski and Potcner (2003) How to recognise a split-plot experiment. *Quality Progress* 60-66
]

---

```r
library(tidyverse)
library(mathart)
set.seed(2021)
colors <- c(blue = "#027EB6", purple = "#746FB2", fuchsia = "#9651A0", 
 ruby = "#C8008F", pink = "#ee64a4", red = "#EE0220", orange = "#D93F00", 
 umber = "#795549", olive = "#6F7C4D", green = "#008A25")
df4 <- map_dfr(1:16, function(n) {
 map_dfr(1:7, function(d) {
 rose_curve(n, d) %>% 
 mutate(id = paste(n, d), color = sample(colors, 1))
 })
}) %>% 
 mutate(id = factor(id, levels = sample(unique(id))))

ggplot(df4, aes(x, y, color = I(color))) +
  geom_path(size = 0.35, lineend = "round", alpha = 0.07) +
  coord_equal() +
  facet_wrap(~id, nrow = 7) + 
  labs(caption = "Rose curves") +
  theme_blankcanvas(margin_cm = 1) +
  theme(plot.caption = element_text(family = "Atma", size = 20, color = "gray"))
```

]

.flex[
.w-60.pr3[
.summary-box.w-100[
* The *grammar of experimental design* is a (programming language agnostic) framework that functionally maps the fundamental components of the experiment to an object oriented programming system to build and modify an experimental design, i.e. reframes how we construct experimental design using software
{{content}}
]
]
.w-40[

.border-box.bg-rladies.white[
.center[
 These slides made using R powered by HTML/CSS/JS can be found at <a style="color:yellow!important" href="https://emitanaka.org/slides-RLadiesMelb2021">emitanaka.org/slides-RLadiesMelb2021</a> 
 All code to reproduce this slide can be found at <a style="color:yellow!important" href="https://github.com/emitanaka/slides-RLadiesMelb2021">github.com/emitanaka/slides-RLadiesMelb2021</a>

]]

* The *edibble* R-package is an implementation of the grammar of experimental design in the R language
<center>
<a href="https://github.com/emitanaka/edibble"> https://github.com/emitanaka/edibble</a>
</center>
{{content}}
--

* The approach is designed to be *human-friendly* and capture natural order of thinking for specifying experimental structure and encourage steps to be explicitly specified
{{content}}
--

* The approach (IMO) also *promotes higher order thinking about experimental design*
{{content}}
--

* Finally, the grammar makes each step modular... you can *easily extend* or *mix-and-match methods*

---

**Statistical Society of Australia Victoria Branch** presents the

]

* Calling for submission for your statistical software product!
* Open only to students (or recent graduates) of Victorian or Tasmanian institutes
* Submissions close at **Fri 26th Nov 2021**
* Winner, announced in March 2022, will win $1,000
* Find more information at

]
.w-60.yellow.pa3[
https://statsocaus.github.io/dicook-award/
]
]