Reproducible workflows with {drake}

class: center, middle, inverse

# Reproducible workflows with {drake}

Bioinformatics London Meetup, 2020-01-30

<i class="fab  fa-twitter "></i> [mattdray](https://twitter.com/mattdray)
<i class="fab  fa-github "></i> [matt-dray](https://github.com/matt-dray)
<i class="fas  fa-globe "></i> [rostrum.blog](https://www.rostrum.blog/)

???
* Thank you for being here instead of the RStudio conference
* I have an entomology MSc and a PhD on ecosystem processes under environmental change (watching leaves rot, if you want to get technical)
* Over five years as a government analyst, much of it spent encouraging a switch to R from Excel-led processes
* (I'm talking today as me, not on behalf of my employer)

---
class: middle, center

???
* An apology: I'm sorry if you came here expecting a different Dr Dray

---
class: middle, center

???
* Another apology: I'm not talking about Canada's best rapper

---
class: middle, center

[{drake}](https://docs.ropensci.org/drake/) by [Will Landau](https://wlandau.github.io/)

???
* {drake}: the R package built by Will Landau for managing workflows and making your analysis pipeline more reproducible
* It's symbols are a brain and a hammer; I think because {drake} stops you from hammering out your own brain

---
class: middle

🙋 Do you: 
--

* use R?
--

* use a workflow manager?
--

* use {drake}?
--

* have FOMO?

---
class: middle, inverse

# Reproducevangelism

???
* I don't want to condescend or dictate, I want to build from the general principle of 'reproducible is good'
* So let's define 'reproducibility'
* Very important in scientific fields given the reproducibility crisis

---
class: middle, center

![](img/turing-way-reproducibility.jpg)

From [The Turing Way](https://the-turing-way.netlify.com/introduction/introduction) by The Alan Turing Institute

???
* We could get into a heated debate about the definition
* The book is a great free resource for reproduciblity in general
* I think we care about both (i) recreating ouputs and (ii) updating with fresh data
* I"m going to refer to this top row as reproducibility throughout

---
class: middle

Can I recreate what you did:
--

* from scratch?
--

* on a different machine?
--

* in the future?
--

* without you present?

???
* Do I have all the code, files and documentation?
* Can I do it on another OS with dependencies missing?
* Will the code still work next year, or will dependency chenages break it?
* Did you document it well enough? Is everything stored somewhere remotely?

---
class: middle, center

[Reproducible Analytical Pipelines](https://ukgovdatascience.github.io/rap-website/) (RAP)

[Can {drake} RAP?](https://www.rostrum.blog/2019/07/23/can-drake-rap/)

???
* In government we recognise the benefits of reproducible analysis and publications
* Literally this is a badge of pride in government
* We want to make sure the whole end-to-end process from data to publication is repeatable
* We want to build trust, but also make it easier to update publications as new data emerges
* I wrote a thing about how {drake} might be useful for RAP in government

---
class: middle

Can I trust these outputs?

Can _you_ trust these outputs?

???
* Working in a reproducible way benefits you, with the excellent by-product of also helping others

---
class: middle

Today's focus:

1. Make workflows reproducible
1. Use {drake}

???
* We could talk about this all day, but I'm focusing on one particular thing today
* If these both mean nothing to you, then great
* If you know about workflow reproducibility, maybe you haven't considered {drake}
* If you do oth these things, then you should be here talking instead of me

---
class: heading-slide, middle, inverse

# 1. Make workflows reproducible

???
* I'm referring to analysis 'workflows', but you might say 'pipelines'
* As in you read, wrangle, model, plot and report data in some fashion
* The process of turning inputs to outputs
* This talk assumes that you do this in R, but ideas are transferable

---
class: middle

R has [many reproducibility tools](https://annakrystalli.me/talks/r-in-repro-research-dc.html#1), like:

* [RStudio Projects](https://swcarpentry.github.io/r-novice-gapminder/02-project-intro/) to keep everything together
* [R Markdown](https://rmarkdown.rstudio.com/) for reproducible docs
* [packages](https://www.hvitfeldt.me/blog/usethis-workflow-for-package-development/) for reusable functions
* [{here}](https://github.com/jennybc/here_here) for relative filepaths
* [{renv}](https://rstudio.github.io/renv/articles/renv.html) for dependency management

???
* Some general tools, but also packages for very specific things like filepath management
* Of course, there's agnostic tools like Git and Docker too
* There are similar tools in other languages

---
class: middle

What about your analytical workflow itself?

How do you keep track of function, file and object relationships?

???
* Have you considered this before?
* You might be asking why you might even do this -- is it really a big deal?

---
class: middle

What if:
--

* you haven't recorded the steps?
--

* the interdependencies become complex?
--

* some steps are computationally intensive?
--

* something changes?

???
* You might be thinking 'no, what's the point?'
* What impact does a change to file 'A' have?
* What if it's time intensive?
* Do you re-run everything from scratch if there's a change?
* Maybe this isn't a big deal for a small analysis, but things can get out of hand quickly

---
class: middle, center

You can't remember this

???
* Here's a real analysis
* All of the files, functions and objects that result in one final output on the right
* The spaghetti is all the relationships
* How could you possibly remember the exact things to re-run if something changes?

---
class: middle

Maybe:

Maybe  `01-read.R`, `02-wrangle.R`, etc?

???
* This doesn't scale well -- might be okay for smaller projects
* Implies the analysis is perfectly sequential -- but some objects from file `01` might be applied in `03` rather than `02`
* You'll still have to re-run from scratch sometimes

---

class: middle, center

Let a workflow manager handle it

<i class="fab  fa-github "></i> [pditommaso/awesome-pipeline](https://github.com/pditommaso/awesome-pipeline)

???
* There are many workflow managers -- some general, some language-specific
* The one that most people might be familiar with is Make (capital 'M')
* It's language agnostic, but has some limitations -- it operates on files, for example and requires you to learn how to use Make
* This is a powerful thought technology

---
class: heading-slide, middle, inverse

# 2. Use {drake}

???
* If there's so many managers then why {drake}?

---
class: middle

{drake} is compelling because it's:
--

* R-specific
--

* free
--

* part of [rOpenSci](https://ropensci.org/)
--
 
* under active development
--

* got great documentation and examples

???
* For {drake}, targets are any arbitrary values in memory (they must be files in Make)
* Proprietary software might be okay when working alone or with specific collaborators who also have the tool, but what about the wider world?
* rOpenSci is 'carefully vetted, staff- and community-contributed R software tools that lower barriers to working with scientific data sources and data that support research applications on the web' -- requires peer review
* Will is active at responding to issues and developing new things
* Will has provided a lot of materials to explain the use of {drake}, as well as learning materials like a Shiny app and installable example projects (I've listed these on a slide at the end)

---
class: middle

At its simplest:

1. `drake_plan()`
1. `make()`
1. Change stuff
1. Go to 2

???
* The smallest amount of things you need to do is run two functions
* You create a 'plan' object that' contains the functions, objects and data in your's a bit like a recipe for generating your analysis
* Then you run `make()` on that object to run the steps and 'bake your cake'
* Let's go through a simple, arbitrary, trivial example
* We're not going to think too hard about folder structure in this example
* In fact, this workflow will be running from within this {xaringan} (RMarkdown slides)

---
class: middle, inverse

#Small demo

???
* Using the built-in `beaver1` and `beaver2` datasets
* You can see something similar in Chapter 3 of the {drake} user manual
* This example is run entirely within this {xaringan} presentation, giving you a flavour of its versatility

---
class: middle

```r
# Get from CRAN with install.packages()

library(drake)
library(dplyr)
library(ggplot2)
library(rphylopic)
```

???
* The first couple of steps look like a 'normal' analysis
* First we load the packages we want to use
* {dplyr} and {ggplot2} from the tidyverse suite
* {rphylopic} is a package by Scott Chamberlain and David Miller that taps into the Phylopic API so we can grab a royalty-free image to adorn our plot

---
class: middle

```r
b_plot <- function(data, image) {
  ggplot(data, aes(id, temp)) +
    geom_boxplot() +
    labs(title = "Weasel temperature") +
    add_phylopic(image)
}

b_table <- function(data) {
    group_by(data, id) %>% 
    summarise(
      mean = mean(temp), sd = sd(temp),
      min = min(temp), max = max(temp)
    ) %>%
    ungroup()
}
```

???
* And then perhaps we've developed a couple of functions
* Here we have one that plots to particular specifications
* And one that produces a very simple table of descriptive statistics
* The beady-eyed among you may have spotted an error in my functions; keep that in mind

---
class: middle

```r
plan <- drake_plan(

b1 = mutate(beaver1, id = "A"),
  b2 = mutate(beaver2, id = "B"),
  beavers = bind_rows(b1, b2),

uid = "be8670c2-a5bd-4b44-88e8-92f8b0c7f4c6",
  png = image_data(uid, size = "512")[[1]],

plot = b_plot(beavers, png),
  table = b_table(beavers),
  report = rmarkdown::render(
*   knitr_in("beavers-report.Rmd"), quiet = TRUE,
*   output_file = file_out("beavers-report.html"),
  )
  
)
```

???
* We wrap the steps of the analysis in `drake_plan()` -- wrangling, fetching a phylopic image, building the outputs (a plot, table and RMarkdown report)
* This doesn't look too dissimilar to a regular script file
* `knitr_in()` and `file_out()` are special -- they ensure that reports are refreshed when a dependency changes

---
class: middle

```r
plan  # it's a dataframe
```

```
## # A tibble: 8 x 2
##   target  command                                                                         
##   <chr>   <expr>                                                                          
## 1 b1      mutate(beaver1, id = "A")                                                      …
## 2 b2      mutate(beaver2, id = "B")                                                      …
## 3 beavers bind_rows(b1, b2)                                                              …
## 4 uid     "be8670c2-a5bd-4b44-88e8-92f8b0c7f4c6"                                         …
## 5 png     image_data(uid, size = "512")[[1]]                                             …
## 6 plot    b_plot(beavers, png)                                                           …
## 7 table   b_table(beavers)                                                               …
## 8 report  rmarkdown::render(knitr_in("beavers-report.Rmd"), quiet = TRUE,      output_fil…
```

???
* What does the plan object look like?
* 'Dataframes in R for Make' == {drake}
* Targets are the things being acted upon and commands are the tasks
* This plan is the key to how {drake} 'remembers' your workflow: it knows the relationships between all the steps and the order they're in
* So far we haven't built anything, we've just created the receipe

---
class: middle

```r
drake::make(plan)  # build the targets
```
```
## target uid
## target b1
## target b2
## target png
## target beavers
## target plot
## target table
## target report
```

???
* To run the workflow, you use `make()`

---
class: middle

The [rendered output]("https://matt-dray.github.io/drake-bioinformatics/weasel-report.html")

???
* Here's what the RMarkdown output looks like after rendering to HTML
* The data was wrangled, a plot and table were produced and this was recorded in an R Markdown file
* You'll need an internet connection to see this

---
class: middle

```r
config <- drake_config(plan)  # a list
config$graph  # e.g. an igraph object
```

```
## IGRAPH 23d1bec DN-- 13 13 -- 
## + attr: name (v/c), imported (v/l)
## + edges from 23d1bec (vertex names):
##  [1] n-OJWWC4TLMRXXO3R2HJZGK3TEMVZA ->report                           
##  [2] plot                           ->report                           
##  [3] table                          ->report                           
##  [4] p-MJSWC5TFOJZS24TFOBXXE5BOKJWWI->report                           
##  [5] report                         ->p-MJSWC5TFOJZS24TFOBXXE5BONB2G23A
##  [6] beavers                        ->plot                             
##  [7] beavers                        ->table                            
##  [8] b_table                        ->table                            
## + ... omitted several edges
```

---
class: middle,

```r
vis_drake_graph(config, main = "", width = 800, height = 450,
                navigationButtons = FALSE, font_size = 15)
```

<div id="htmlwidget-2f9c1e647e63442f78c3" style="width:800px;height:450px;" class="visNetwork html-widget"></div>
<script type="application/json" data-for="htmlwidget-2f9c1e647e63442f78c3">{"x":{"nodes":{"id":["n-OJWWC4TLMRXXO3R2HJZGK3TEMVZA","p-MJSWC5TFOJZS24TFOBXXE5BOKJWWI","b_table","b_plot","uid","b1","b2","png","beavers","plot","table","report","p-MJSWC5TFOJZS24TFOBXXE5BONB2G23A"],"imported":[true,true,true,true,false,false,false,false,false,false,false,false,false],"label":["rmarkdown::render","file beavers-report.Rmd","b_table","b_plot","uid\n0.006s","b1\n0.01s","b2\n0.005s","png\n1.647s","beavers\n0.01s","plot\n0.423s","table\n0.04s","report\n1.572s","file beavers-report.html"],"status":["imported","imported","imported","imported","up to date","up to date","up to date","up to date","up to date","up to date","up to date","up to date","up to date"],"type":["function","file","function","function","object","object","object","object","object","object","object","object","file"],"font.size":[15,15,15,15,15,15,15,15,15,15,15,15,15],"color":["#1874CD","#1874CD","#1874CD","#1874CD","#228B22","#228B22","#228B22","#228B22","#228B22","#228B22","#228B22","#228B22","#228B22"],"shape":["triangle","square","triangle","triangle","dot","dot","dot","dot","dot","dot","dot","dot","square"],"level":[1,1,1,1,1,1,1,2,2,3,3,4,5],"x":[-1,-1,-1,-1,-1,-1,-1,-0.5,-0.5,0,0,0.5,1],"y":[-1,-0.666666666666667,-0.333333333333333,0,0.333333333333333,0.666666666666667,1,-0.444444444444445,0.444444444444444,-0.444444444444445,0.444444444444444,0,0]},"edges":{"from":["n-OJWWC4TLMRXXO3R2HJZGK3TEMVZA","plot","table","p-MJSWC5TFOJZS24TFOBXXE5BOKJWWI","report","beavers","beavers","b_table","b_plot","png","uid","b1","b2"],"to":["report","report","report","report","p-MJSWC5TFOJZS24TFOBXXE5BONB2G23A","plot","table","table","plot","plot","png","beavers","beavers"],"arrows":["to","to","to","to","to","to","to","to","to","to","to","to","to"]},"nodesToDataframe":true,"edgesToDataframe":true,"options":{"width":"100%","height":"100%","nodes":{"shape":"dot","physics":false},"manipulation":{"enabled":false},"edges":{"smooth":false},"physics":{"stabilization":false}},"groups":null,"width":800,"height":450,"idselection":{"enabled":false,"style":"width: 150px; height: 26px","useLabels":true,"main":"Select by id"},"byselection":{"enabled":false,"style":"width: 150px; height: 26px","multiple":false,"hideColor":"rgba(200,200,200,0.5)","highlight":false},"main":{"text":"","style":"font-family:Georgia, Times New Roman, Times, serif;font-weight:bold;font-size:20px;text-align:center;"},"submain":null,"footer":null,"background":"rgba(0, 0, 0, 0)","highlight":{"enabled":false,"hoverNearest":false,"degree":1,"algorithm":"all","hideColor":"rgba(200,200,200,0.5)","labelOnly":true},"collapse":{"enabled":true,"fit":false,"resetHighlight":true,"clusterOptions":null,"keepCoord":true,"labelSuffix":"(cluster)"},"legend":{"width":0.2,"useGroups":false,"position":"left","ncol":1,"stepX":100,"stepY":100,"zoom":true,"nodes":{"label":["Up to date","Imported","Object","Function","File"],"color":["#228B22","#1874CD","#888888","#888888","#888888"],"shape":["dot","dot","dot","triangle","square"],"font.color":["black","black","black","black","black"],"font.size":[15,15,15,15,15],"id":[1,5,7,9,10]},"nodesToDataframe":true},"igraphlayout":{"type":"square"}},"evals":[],"jsHooks":[]}</script>

???
* {drake} has a function that takes advantage of {visNetwork} to display the dependency graph
* You pass it the `config` object
* Other arguments here are aesthetic
* Targets: circle is object, square is fie, triangle is function
* The arrows trace the dependencies

---
class: middle

```r
cached()  # targets in hidden .drake/ folder
```

```
## [1] "b1"      "b2"      "beavers" "plot"    "png"     "report"  "table"   "uid"
```

???
* You might be wondering where the targets 'live'
* They get cached -- stored in a folder
* That folder is hidden in the project directory as `.drake/`

---
class: middle

```r
readd(table)  # read from the cache
```

```
## # A tibble: 2 x 5
##   id     mean    sd   min   max
##   <chr> <dbl> <dbl> <dbl> <dbl>
## 1 A      36.9 0.193  36.3  37.5
## 2 B      37.6 0.447  36.6  38.4
```

???
* How can you access objects in the cache?
* `readd()` and `loadd()` do this
* In fact, this is how the R Markdown file in this example fetches the plot and table

---
class: middle

```r
b_plot <- function(data, image) {
  ggplot(data, aes(id, temp)) +
    geom_boxplot() +
*   labs(title = "Beaver temperature") +  # fix!
    add_phylopic(image)
}
```

```r
outdated(config)  # check what's impacted
```

```
## [1] "plot"   "report"
```

???
* Remember I said there was a problem with one of my functions?
* Yes, the plot function used the word 'weasel' instead of 'beaver' in the title
* Let's go back and update the function
* This update means that the targets that depend on this function need to be updated
* These can be checked with a call to `outdated()`

---
class: middle

```r
vis_drake_graph(config, main = "", width = 800, height = 450,
                navigationButtons = FALSE, font_size = 15)
```

<div id="htmlwidget-dcafbfe41505535bb38c" style="width:800px;height:450px;" class="visNetwork html-widget"></div>
<script type="application/json" data-for="htmlwidget-dcafbfe41505535bb38c">{"x":{"nodes":{"id":["n-OJWWC4TLMRXXO3R2HJZGK3TEMVZA","p-MJSWC5TFOJZS24TFOBXXE5BOKJWWI","b_table","b_plot","uid","b1","b2","png","beavers","plot","table","report","p-MJSWC5TFOJZS24TFOBXXE5BONB2G23A"],"imported":[true,true,true,true,false,false,false,false,false,false,false,false,false],"label":["rmarkdown::render","file beavers-report.Rmd","b_table","b_plot","uid\n0.006s","b1\n0.01s","b2\n0.005s","png\n1.647s","beavers\n0.01s","plot\n0.423s","table\n0.04s","report\n1.572s","file beavers-report.html"],"status":["imported","imported","imported","imported","up to date","up to date","up to date","up to date","up to date","outdated","up to date","outdated","outdated"],"type":["function","file","function","function","object","object","object","object","object","object","object","object","file"],"font.size":[15,15,15,15,15,15,15,15,15,15,15,15,15],"color":["#1874CD","#1874CD","#1874CD","#1874CD","#228B22","#228B22","#228B22","#228B22","#228B22","#000000","#228B22","#000000","#000000"],"shape":["triangle","square","triangle","triangle","dot","dot","dot","dot","dot","dot","dot","dot","square"],"level":[1,1,1,1,1,1,1,2,2,3,3,4,5],"x":[-1,-1,-1,-1,-1,-1,-1,-0.5,-0.5,0,0,0.5,1],"y":[-1,-0.666666666666667,-0.333333333333333,0,0.333333333333333,0.666666666666667,1,-0.444444444444445,0.444444444444444,-0.444444444444445,0.444444444444444,0,0]},"edges":{"from":["n-OJWWC4TLMRXXO3R2HJZGK3TEMVZA","plot","table","p-MJSWC5TFOJZS24TFOBXXE5BOKJWWI","report","beavers","beavers","b_table","b_plot","png","uid","b1","b2"],"to":["report","report","report","report","p-MJSWC5TFOJZS24TFOBXXE5BONB2G23A","plot","table","table","plot","plot","png","beavers","beavers"],"arrows":["to","to","to","to","to","to","to","to","to","to","to","to","to"]},"nodesToDataframe":true,"edgesToDataframe":true,"options":{"width":"100%","height":"100%","nodes":{"shape":"dot","physics":false},"manipulation":{"enabled":false},"edges":{"smooth":false},"physics":{"stabilization":false}},"groups":null,"width":800,"height":450,"idselection":{"enabled":false,"style":"width: 150px; height: 26px","useLabels":true,"main":"Select by id"},"byselection":{"enabled":false,"style":"width: 150px; height: 26px","multiple":false,"hideColor":"rgba(200,200,200,0.5)","highlight":false},"main":{"text":"","style":"font-family:Georgia, Times New Roman, Times, serif;font-weight:bold;font-size:20px;text-align:center;"},"submain":null,"footer":null,"background":"rgba(0, 0, 0, 0)","highlight":{"enabled":false,"hoverNearest":false,"degree":1,"algorithm":"all","hideColor":"rgba(200,200,200,0.5)","labelOnly":true},"collapse":{"enabled":true,"fit":false,"resetHighlight":true,"clusterOptions":null,"keepCoord":true,"labelSuffix":"(cluster)"},"legend":{"width":0.2,"useGroups":false,"position":"left","ncol":1,"stepX":100,"stepY":100,"zoom":true,"nodes":{"label":["Up to date","Outdated","Imported","Object","Function","File"],"color":["#228B22","#000000","#1874CD","#888888","#888888","#888888"],"shape":["dot","dot","dot","dot","triangle","square"],"font.color":["black","black","black","black","black","black"],"font.size":[15,15,15,15,15,15],"id":[1,2,5,7,9,10]},"nodesToDataframe":true},"igraphlayout":{"type":"square"}},"evals":[],"jsHooks":[]}</script>

???
* We can revisit the network graph to get a visual representation of the impact of our change
* You can now see that the colour of some targets has turned black
* These are the targets that are now out of date
* {drake} remembered this to you didn't have to

---
class: middle

```r
drake::make(plan)  # rebuild outdated targets
```
```
## target plot
## target report
```

???
* Of course, we can now re-`make()` the plan
* {drake} will only update the things that have been impacted by our change
* This is a trivial example -- consider how helpful this would be if you had even more targets and more complicated relationships, especially if you have steps that are computationally-intensive

---
class: middle

The [rendered output]("https://matt-dray.github.io/drake-bioinformatics/beavers-report.html"), updated

???
* And we can peak at the rendered R Markdown once more to check that our change to the plot function had its desired effect
* You'll need an internet connection to see this

---
class: middle

To recap:
--

1. `drake_plan()`
--

1. `make()`
--

1. Change stuff
--

1. Go to 2

???
* Let's recap those steps again
* Create a plan with targets and commands -- it's a dataframe
* Make that plan to run the analysis
* Something changes -- input data, function specifications, etc
* (You can check what's outdated and check the network graph)
* Now re-`make()`
* And you can do this over and over
* What gets done stays done

---
class: inverse, middle

# What now?

???
* We talked about reproduciblity, about workflow managers, about {drake}
* What do I think you should do now?

---
class: middle

Try it

```
install.packages("drake")
library(drake)
drake_example("main")
```
???
* You can install an example structure from the package itself
* Otherwise there are many documents you can find on the web

---
class: middle

Check out official {drake} materials:

* rOpenSci [{drake} website](https://docs.ropensci.org/drake/)
* the [user manual](https://books.ropensci.org/drake/)
* an [rOpenSci community call](https://ropensci.org/commcalls/2019-09-24/)
* [learndrake](https://github.com/wlandau/learndrake) in the cloud
* [drakeplanner](https://wlandau.shinyapps.io/drakeplanner/_w_7935044f/) Shiny app
* launch [drake examples](https://github.com/wlandau/drake-examples) in the cloud
* source [on GitHub](https://github.com/ropensci/drake)
* the [CRAN listing](https://cloud.r-project.org/web/packages/drake/index.html)

---
class: middle

We didn't cover everything, like:

* high-performance computing
* target-creation history and recovery
* static and dynamic branching
* time prediction
* memory strategies
* trigger customisation

???
* Clearly there isn't enough time to go through all the great things about {drake}
* Luckily {drake} has rich documentation

---
class: middle

Revisit today's materials:

* [matt-dray.github.io/drake-bioinformatics/](https://matt-dray.github.io/drake-bioinformatics/#1)
* [github.com/matt-dray/drake-bioinformatics](https://github.com/matt-dray/drake-bioinformatics)
* [R script](https://github.com/matt-dray/drake-bioinformatics/blob/master/drake-beavers-workflow.R) of the code used

Plus:

* [Can {drake} RAP?](https://www.rostrum.blog/2019/07/23/can-drake-rap/) blog post
* run this example in Binder [![Launch Rstudio Binder](http://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/matt-dray/drake-egg-rap/master?urlpath=rstudio)

???
* Slides and source on GitHub, along with a single script file of the code used in the presentation
* Binder is a service that lets you launch a Notebook or RStudio instance in the cloud with data and code that you provide

---
class: inverse, middle

# Reproducible workflows with {drake}

1. Make workflows reproducible
1. Use {drake}

What gets done stays done!

???
* I hope I've convinced you to consider a workflow manager
* And hopefully to take a look at {drake} as your tool of choice
* At worst, you might have learnt about {rphylopic}!
* I'm gong to finish here, but we have time for some words from our special guest, Will Landau

---
class: middle

Sources

* Dr Not-Dray [via Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Snoop_Dogg_and_Dr._Dre.jpg)
* Canadian rapper [via Wikimedia Commons](https://commons.wikimedia.org/wiki/Category:Drake)
* {drake} [hex logo](https://camo.githubusercontent.com/44749362ca36c9e3295f2bcf18975d811564c121/68747470733a2f2f646f63732e726f70656e7363692e6f72672f6472616b652f7265666572656e63652f666967757265732f6c6f676f2e737667)
* {drake} hairball by [Frederik Aust](https://twitter.com/FrederikAust/status/1205103780938833921?s=20)
* _Castor canadensis_ from [PhyloPic](http://phylopic.org/image/b3dd721e-6084-4413-8300-44e10d8fd3ca/)