Petite introduction au package purrr

class: center, middle, inverse, title-slide

# Petite introduction au package purrr
### Guillaume Devailly
### inra
### 2019/09/17

---

class: center

# Je veux en savoir plus sur Pikachu

![:img_scale 80%](img/PokéAPI.png)

---

```r
library(purrr)    # the package I'm going to present
library(jsonlite) # importing JSON data as R list
library(dplyr)    # imported for tible, pipe, etc.
```
---

```r
pikachu <- jsonlite::fromJSON(
    "https://pokeapi.co/api/v2/pokemon/pikachu"      
)
str(pikachu, max.level = 1)
## List of 17
##  $ abilities               :'data.frame':	2 obs. of  3 variables:
##  $ base_experience         : int 112
##  $ forms                   :'data.frame':	1 obs. of  2 variables:
##  $ game_indices            :'data.frame':	20 obs. of  2 variables:
##  $ height                  : int 4
##  $ held_items              :'data.frame':	2 obs. of  2 variables:
##  $ id                      : int 25
##  $ is_default              : logi TRUE
##  $ location_area_encounters: chr "https://pokeapi.co/api/v2/pokemon/25/encounters"
##  $ moves                   :'data.frame':	81 obs. of  2 variables:
##  $ name                    : chr "pikachu"
##  $ order                   : int 35
##  $ species                 :List of 2
##  $ sprites                 :List of 8
##  $ stats                   :'data.frame':	6 obs. of  3 variables:
##  $ types                   :'data.frame':	1 obs. of  2 variables:
##  $ weight                  : int 60
```

---

# Quel est l'Indice de Masse Corporelle de Pikachu ?
### (imc, ou bmi en anglais)

```r
with(pikachu, tibble(
    name = "Pikachu",
    weight = weight/10, # API in hectograms, we want kilograms
    height = height/10, # API in decimeters, we want meters
    BMI = weight / (height^2),
    type = paste(types$type$name, collapse = ", "), # Pokemons can have several types, collapsing
    sprite = sprites$front_default
))
## # A tibble: 1 x 6
##   name   weight height   BMI type    sprite                                
##   <chr>   <dbl>  <dbl> <dbl> <chr>   <chr>                                 
## 1 Pikac~      6    0.4  37.5 electr~ https://raw.githubusercontent.com/Pok~
```

---

# Mais si je m'intéresse aussi à Bulbizarre, Salamèche et Carapuce ?

---
Solution 1 : copier, coller, modifier

```r
bulbasaur <- jsonlite::fromJSON(
    "https://pokeapi.co/api/v2/pokemon/bulbasaur"    
)
with(bulbasaur, tibble(
    name = "bulbasaur",
    weight = weight/10, 
    height = height/10,
    BMI = weight / (height^2),
    type = paste(types$type$name, collapse = ", "),
    sprite = sprites$front_default
))
## # A tibble: 1 x 6
##   name    weight height   BMI type      sprite                             
##   <chr>    <dbl>  <dbl> <dbl> <chr>     <chr>                              
## 1 bulbas~    6.9    0.7  14.1 poison, ~ https://raw.githubusercontent.com/~

charmander <- jsonlite::fromJSON(
    "https://pokeapi.co/api/v2/pokemon/charmander"      
)
with(charmander, tibble(
    name = "charmander",
    weight = weight/10,
    height = height/10, 
    BMI = weight / (height^2),
    type = paste(types$type$name, collapse = ", "),
    sprite = sprites$front_default
))
## # A tibble: 1 x 6
##   name     weight height   BMI type  sprite                                
##   <chr>     <dbl>  <dbl> <dbl> <chr> <chr>                                 
## 1 charman~    8.5    0.6  23.6 fire  https://raw.githubusercontent.com/Pok~

squirtle <- jsonlite::fromJSON(
    "https://pokeapi.co/api/v2/pokemon/squirtle"      
)
with(squirtle, tibble(
    name = "squirtle",
    weight = weight/10,
    height = height/10, 
    BMI = weight / (height^2),
    type = paste(types$type$name, collapse = ", "),
    sprite = sprites$front_default
))
## # A tibble: 1 x 6
##   name    weight height   BMI type  sprite                                 
##   <chr>    <dbl>  <dbl> <dbl> <chr> <chr>                                  
## 1 squirt~      9    0.5    36 water https://raw.githubusercontent.com/Poke~
```

---
Solution 1 : copier, coller, **modifier**

```r
*bulbasaur <- jsonlite::fromJSON(
*   "https://pokeapi.co/api/v2/pokemon/bulbasaur"
)
*with(bulbasaur, tibble(
*   name = "bulbasaur",
    weight = weight/10, 
    height = height/10, 
    BMI = weight / (height^2),
    type = paste(types$type$name, collapse = ", "),
    sprite = sprites$front_default
))
## # A tibble: 1 x 6
##   name    weight height   BMI type      sprite                             
##   <chr>    <dbl>  <dbl> <dbl> <chr>     <chr>                              
## 1 bulbas~    6.9    0.7  14.1 poison, ~ https://raw.githubusercontent.com/~

*charmander <- jsonlite::fromJSON(
*   "https://pokeapi.co/api/v2/pokemon/charmander"
)
*with(charmander, tibble(
*   name = "charmander",
    weight = weight/10, 
    height = height/10, 
    BMI = weight / (height^2),
    type = paste(types$type$name, collapse = ", "),
    sprite = sprites$front_default
))
## # A tibble: 1 x 6
##   name     weight height   BMI type  sprite                                
##   <chr>     <dbl>  <dbl> <dbl> <chr> <chr>                                 
## 1 charman~    8.5    0.6  23.6 fire  https://raw.githubusercontent.com/Pok~

*squirtle <- jsonlite::fromJSON(
*   "https://pokeapi.co/api/v2/pokemon/squirtle"
)
*with(squirtle, tibble(
*   name = "squirtle",
    weight = weight/10,
    height = height/10, 
    BMI = weight / (height^2),
    type = paste(types$type$name, collapse = ", "),
    sprite = sprites$front_default
))
## # A tibble: 1 x 6
##   name    weight height   BMI type  sprite                                 
##   <chr>    <dbl>  <dbl> <dbl> <chr> <chr>                                  
## 1 squirt~      9    0.5    36 water https://raw.githubusercontent.com/Poke~
```

---

* laborieux
* source d'erreurs
* répercuter tout changement dans chaque copier - coller

Mais les IDE peuvent aider :
* Rechercher et Remplacer
* curseurs multiples (alt + click sour RStudio)

---
Solution 2: une boucle `for`

```r
my_pokemons <- c("pikachu", "bulbasaur", "charmander", "squirtle")

pokemon_bmi <- tibble( # pré-allocation
    name = character(0),
    weight = double(0),
    height = double(0),
    bmi = double(0),
    type = character(0),
    sprite = character(0)
)

for (i in seq_along(my_pokemons)) {
    poke_i <-  jsonlite::fromJSON(
        paste0("https://pokeapi.co/api/v2/pokemon/", my_pokemons[i])
    ) 
    pokemon_bmi[i, ] <- with(poke_i, tibble( 
        name =  my_pokemons[i],
        weight = weight/10,
        height = height/10, 
        BMI = weight / (height^2),
        type = paste(types$type$name, collapse = ", "),
        sprite = sprites$front_default
    ))
}
```

---

```r
pokemon_bmi
## # A tibble: 4 x 6
##   name     weight height   bmi type     sprite                             
## * <chr>     <dbl>  <dbl> <dbl> <chr>    <chr>                              
## 1 pikachu     6      0.4  37.5 electric https://raw.githubusercontent.com/~
## 2 bulbasa~    6.9    0.7  14.1 poison,~ https://raw.githubusercontent.com/~
## 3 charman~    8.5    0.6  23.6 fire     https://raw.githubusercontent.com/~
## 4 squirtle    9      0.5  36   water    https://raw.githubusercontent.com/~
```

* fonctionne
* rapide

Mais :
* verbeux
* crée des variables dans l'environnement global
* peut être lent si l'objet grandit : `pokemon_bmi <- rbind(pokemon_bmi, new_pokemon)`
* besoin de réfléchir pour ne pas tomber dans ce genre de piège
* non immédiatement parallélisable

---

La solution idiomatique : `lapply()`

Étape 1 : créer une fonction

```r
get_pokemon_bmi <- function(poke_name, sleep = 0) {
    my_pokemon <-  jsonlite::fromJSON(
        paste0("https://pokeapi.co/api/v2/pokemon/", poke_name)
    ) 
    Sys.sleep(sleep)
    return(
        with(my_pokemon, tibble( 
            name = name,
            weight = weight/10,
            height = height/10, 
            BMI = weight / (height^2),
            type = paste(types$type$name, collapse = ", "),
            sprite = sprites$front_default
        ))
    )
}
get_pokemon_bmi("pikachu")
## # A tibble: 1 x 6
##   name   weight height   BMI type    sprite                                
##   <chr>   <dbl>  <dbl> <dbl> <chr>   <chr>                                 
## 1 pikac~      6    0.4  37.5 electr~ https://raw.githubusercontent.com/Pok~
```

---

Étape 2 : appeler la fonction sur chaque élément d'un vecteur

```r
my_pokemons <- c("pikachu", "bulbasaur", "charmander", "squirtle")

pokemon_bmi <- lapply(
    my_pokemons, # for each element in my_pokemons
    get_pokemon_bmi # run function get_pokemon_bmi 
)

str(pokemon_bmi) # a list :-(
## List of 4
##  $ :Classes 'tbl_df', 'tbl' and 'data.frame':	1 obs. of  6 variables:
##   ..$ name  : chr "pikachu"
##   ..$ weight: num 6
##   ..$ height: num 0.4
##   ..$ BMI   : num 37.5
##   ..$ type  : chr "electric"
##   ..$ sprite: chr "https://raw.githubusercontent.com/PokeAPI/sprites/master/sprites/pokemon/25.png"
##  $ :Classes 'tbl_df', 'tbl' and 'data.frame':	1 obs. of  6 variables:
##   ..$ name  : chr "bulbasaur"
##   ..$ weight: num 6.9
##   ..$ height: num 0.7
##   ..$ BMI   : num 14.1
##   ..$ type  : chr "poison, grass"
##   ..$ sprite: chr "https://raw.githubusercontent.com/PokeAPI/sprites/master/sprites/pokemon/1.png"
##  $ :Classes 'tbl_df', 'tbl' and 'data.frame':	1 obs. of  6 variables:
##   ..$ name  : chr "charmander"
##   ..$ weight: num 8.5
##   ..$ height: num 0.6
##   ..$ BMI   : num 23.6
##   ..$ type  : chr "fire"
##   ..$ sprite: chr "https://raw.githubusercontent.com/PokeAPI/sprites/master/sprites/pokemon/4.png"
##  $ :Classes 'tbl_df', 'tbl' and 'data.frame':	1 obs. of  6 variables:
##   ..$ name  : chr "squirtle"
##   ..$ weight: num 9
##   ..$ height: num 0.5
##   ..$ BMI   : num 36
##   ..$ type  : chr "water"
##   ..$ sprite: chr "https://raw.githubusercontent.com/PokeAPI/sprites/master/sprites/pokemon/7.png"
```

---

Étape 3 : transformer la liste en tableau

```r
pokemon_bmi <- do.call(
    rbind,
    pokemon_bmi
)

pokemon_bmi
## # A tibble: 4 x 6
##   name     weight height   BMI type     sprite                             
##   <chr>     <dbl>  <dbl> <dbl> <chr>    <chr>                              
## 1 pikachu     6      0.4  37.5 electric https://raw.githubusercontent.com/~
## 2 bulbasa~    6.9    0.7  14.1 poison,~ https://raw.githubusercontent.com/~
## 3 charman~    8.5    0.6  23.6 fire     https://raw.githubusercontent.com/~
## 4 squirtle    9      0.5  36   water    https://raw.githubusercontent.com/~
```

---

* en théorie, moins besoin de mettre les mains dans le cambouis
* en pratique, `x <- do.call(rbind, x)` ...
* force a écrire des fonctions
* facilement parallélisable :

```r
pokemon_bmi <- parallel::mclapply(
    my_pokemons,
    get_pokemon_bmi,
    mc.cores = 8
)
## Error in parallel::mclapply(my_pokemons, get_pokemon_bmi, mc.cores = 8): 'mc.cores' > 1 is not supported on Windows
```

---

Solution 4: le tidyverse

Étape 1 : définir une fonction

Étape 2 : itérer

```r
my_pokemons <- c("pikachu", "bulbasaur", "charmander", "squirtle")

pokemon_bmi <- map_dfr( # le résultat est un _D_ata _F_rame aggloméré par _R_angs
    my_pokemons,
    get_pokemon_bmi
)

---

* moins besoin de mettre les mains dans le cambouis
* facilement parallélisable :

```r
library(furrr)
plan(multiprocess(workers = 8))
future_map_dfr(my_pokemons, get_pokemon_bmi)
## # A tibble: 4 x 6
##   name     weight height   BMI type     sprite                             
##   <chr>     <dbl>  <dbl> <dbl> <chr>    <chr>                              
## 1 pikachu     6      0.4  37.5 electric https://raw.githubusercontent.com/~
## 2 bulbasa~    6.9    0.7  14.1 poison,~ https://raw.githubusercontent.com/~
## 3 charman~    8.5    0.6  23.6 fire     https://raw.githubusercontent.com/~
## 4 squirtle    9      0.5  36   water    https://raw.githubusercontent.com/~
```

---

Une fonction `map_*` par type de résultat souhaité :

<table>
 <thead>
  <tr>
   <th style="text-align:left;"> function </th>
   <th style="text-align:left;"> result </th>
  </tr>
 </thead>
<tbody>
  <tr>
   <td style="text-align:left;"> map() </td>
   <td style="text-align:left;"> list </td>
  </tr>
  <tr>
   <td style="text-align:left;"> map_chr() </td>
   <td style="text-align:left;"> character vector </td>
  </tr>
  <tr>
   <td style="text-align:left;"> map_int() </td>
   <td style="text-align:left;"> integer vector </td>
  </tr>
  <tr>
   <td style="text-align:left;"> map_dbl() </td>
   <td style="text-align:left;"> double vector </td>
  </tr>
  <tr>
   <td style="text-align:left;"> map_lgl() </td>
   <td style="text-align:left;"> logical vector </td>
  </tr>
  <tr>
   <td style="text-align:left;"> map_dfr() </td>
   <td style="text-align:left;"> data frame, row bind </td>
  </tr>
  <tr>
   <td style="text-align:left;"> map_dfc() </td>
   <td style="text-align:left;"> data frame, col bind </td>
  </tr>
  <tr>
   <td style="text-align:left;"> walk() </td>
   <td style="text-align:left;"> no return (side effects) </td>
  </tr>
</tbody>
</table>

---

Fonctions anonymes :

```r
map_dbl(my_pokemons, function(x) {
    get_pokemon_bmi(x)[["height"]]
})
## [1] 0.4 0.7 0.6 0.5

map_dbl(my_pokemons, ~get_pokemon_bmi(.x)[["height"]])
## [1] 0.4 0.7 0.6 0.5
```

---

Préserve les noms :

```r
my_pokemons
## [1] "pikachu"    "bulbasaur"  "charmander" "squirtle"

set_names(my_pokemons)
##      pikachu    bulbasaur   charmander     squirtle 
##    "pikachu"  "bulbasaur" "charmander"   "squirtle"

my_pokemons %>%
    map_chr(~get_pokemon_bmi(.x)[["type"]])
## [1] "electric"      "poison, grass" "fire"          "water"

set_names(my_pokemons) %>%
    map_chr(~get_pokemon_bmi(.x)[["type"]])
##         pikachu       bulbasaur      charmander        squirtle 
##      "electric" "poison, grass"          "fire"         "water"
```

---
Attrapage d'erreur :

```r
my_pokemons <- c("pikachu", "bulbasaur", "charmander", "squirtle", "magicarpe")
map(my_pokemons, get_pokemon_bmi)
## Error in open.connection(con, "rb"): HTTP error 404.
```

**Tout** le calcul est perdu à cause d'une erreur sur **une seule** itération 
(T__T)

---

```r
safe_get_pokemon_bmi <- safely(get_pokemon_bmi)

safe_get_pokemon_bmi("pikachu")
## $result
## # A tibble: 1 x 6
##   name   weight height   BMI type    sprite                                
##   <chr>   <dbl>  <dbl> <dbl> <chr>   <chr>                                 
## 1 pikac~      6    0.4  37.5 electr~ https://raw.githubusercontent.com/Pok~
## 
## $error
## NULL

safe_get_pokemon_bmi("magicarpe")
## $result
## NULL
## 
## $error
## <simpleError in open.connection(con, "rb"): HTTP error 404.>
```

---

```r
prelim_results <- set_names(my_pokemons) %>%
    map(safe_get_pokemon_bmi)

success <- map_lgl(prelim_results, ~is.null(.x$error))
success
##    pikachu  bulbasaur charmander   squirtle  magicarpe 
##       TRUE       TRUE       TRUE       TRUE      FALSE

map_dfr(prelim_results[success], "result")
## # A tibble: 4 x 6
##   name     weight height   BMI type     sprite                             
##   <chr>     <dbl>  <dbl> <dbl> <chr>    <chr>                              
## 1 pikachu     6      0.4  37.5 electric https://raw.githubusercontent.com/~
## 2 bulbasa~    6.9    0.7  14.1 poison,~ https://raw.githubusercontent.com/~
## 3 charman~    8.5    0.6  23.6 fire     https://raw.githubusercontent.com/~
## 4 squirtle    9      0.5  36   water    https://raw.githubusercontent.com/~
```

---

Ou alors :

```r
possibly_get_pokemon_bmi <- possibly(get_pokemon_bmi, otherwise = NULL)

possibly_get_pokemon_bmi("pikachu")
## # A tibble: 1 x 6
##   name   weight height   BMI type    sprite                                
##   <chr>   <dbl>  <dbl> <dbl> <chr>   <chr>                                 
## 1 pikac~      6    0.4  37.5 electr~ https://raw.githubusercontent.com/Pok~

possibly_get_pokemon_bmi("magicarpe")
## NULL

my_pokemons
## [1] "pikachu"    "bulbasaur"  "charmander" "squirtle"   "magicarpe"
map_dfr(
    my_pokemons,
    ~possibly(get_pokemon_bmi, otherwise = NULL)(.x)
)
## # A tibble: 4 x 6
##   name     weight height   BMI type     sprite                             
##   <chr>     <dbl>  <dbl> <dbl> <chr>    <chr>                              
## 1 pikachu     6      0.4  37.5 electric https://raw.githubusercontent.com/~
## 2 bulbasa~    6.9    0.7  14.1 poison,~ https://raw.githubusercontent.com/~
## 3 charman~    8.5    0.6  23.6 fire     https://raw.githubusercontent.com/~
## 4 squirtle    9      0.5  36   water    https://raw.githubusercontent.com/~
```

---

Attention : les erreurs sont devenues silencieuses !

```r
get_pokemon_bmi("magicarpe")
## Error in open.connection(con, "rb"): HTTP error 404.

possibly(get_pokemon_bmi, otherwise = NULL)("magicarpe")
## NULL

possibly(get_pokemon_bmi, otherwise = NULL, quiet = FALSE)("magicarpe")
## Error: HTTP error 404.
## NULL
```

---
Et sur 151 pokémon ?

```r
pokemon_bmi <- map_dfr(1:151, possibly_get_pokemon_bmi, sleep = 1)

library(ggplot2)
library(ggimage)

ggplot(pokemon_bmi, aes(x = weight, y = height, text = name)) +
    stat_function(fun = function(w) sqrt(w/30), color = "darkred", linetype = "dotted") +
    stat_function(fun = function(w) sqrt(w/15), color = "orange", linetype = "dotted") +
    annotate("text", x = rep(350, 2), y = c(3.1, 4.5), label = c("BMI = 30", "BMI = 15"), color = c("darkred", "orange")) +
    geom_image(aes(image = sprite), size = 0.12) +
    theme_bw(base_size = 14)

```

---
class: center

![](intro_purrr_files/figure-html/unnamed-chunk-23-1.png)

---
# Bonus

Vectorisez vos fonctions, plus besoin de boucles !

```r
get_pokemon_bmi(c("pikachu", "bulbasaur", "charmander", "squirtle"))
## Error: lexical error: invalid char in json text.
##                                        https://pokeapi.co/api/v2/pokem
##                      (right here) ------^

get_pokemon_bmi_v <- function(poke_names, sleep = 0) {
    map_dfr(poke_names, get_pokemon_bmi, sleep = sleep)
}

get_pokemon_bmi_v(c("pikachu", "bulbasaur", "charmander", "squirtle"))
## # A tibble: 4 x 6
##   name     weight height   BMI type     sprite                             
##   <chr>     <dbl>  <dbl> <dbl> <chr>    <chr>                              
## 1 pikachu     6      0.4  37.5 electric https://raw.githubusercontent.com/~
## 2 bulbasa~    6.9    0.7  14.1 poison,~ https://raw.githubusercontent.com/~
## 3 charman~    8.5    0.6  23.6 fire     https://raw.githubusercontent.com/~
## 4 squirtle    9      0.5  36   water    https://raw.githubusercontent.com/~

get_pokemon_bmi_v("pikachu")
## # A tibble: 1 x 6
##   name   weight height   BMI type    sprite                                
##   <chr>   <dbl>  <dbl> <dbl> <chr>   <chr>                                 
## 1 pikac~      6    0.4  37.5 electr~ https://raw.githubusercontent.com/Pok~
```

---

# Pour aller plus loin

- passer deux séries d'arguments : `map2(x, y, fun)`
- passer n séries d'arguments : `pmap(list(x, y, z, ...), fun)`
- exécuter sur un sous ensemble : `map_at(x, at, fun)`, `map_if(x, cond, fun)`
- composer des fonctions : `compose(fun1, fun2)`
- et plus encore !

Liens :
- [Itération de fonctions avec purrr](http://perso.ens-lyon.fr/lise.vaudor/iterer-des-fonctions-avec-purrr/)
- [Lesser known purrr tricks](https://www.brodrigues.co/blog/2017-03-24-lesser_known_purrr/)