clasificacion.utf8

background-image: url(img/portada.png)
background-size: cover
class: animated slideInRight fadeOutLeft, middle

# Modelos de Clasificación

### 1º Congreso Latinoamericano de Mujeres en Bioinformática y Ciencia de Datos

---

## Vamos a modelar las especies 🐧

### Ingreso los datos

```r
library(tidymodels) 
library(palmerpenguins)

penguins <- palmerpenguins::penguins %>%
  drop_na() %>% #elimino valores perdidos
  select(-year,-sex, -island) #elimino columnas q no son numéricas
glimpse(penguins) #observo variables restantes
```

```
## Rows: 333
## Columns: 5
## $ species           <fct> Adelie, Adelie, Adelie, Adelie, Adelie, Adelie, Ade…
## $ bill_length_mm    <dbl> 39.1, 39.5, 40.3, 36.7, 39.3, 38.9, 39.2, 41.1, 38.…
## $ bill_depth_mm     <dbl> 18.7, 17.4, 18.0, 19.3, 20.6, 17.8, 19.6, 17.6, 21.…
## $ flipper_length_mm <int> 181, 186, 195, 193, 190, 181, 195, 182, 191, 198, 1…
## $ body_mass_g       <int> 3750, 3800, 3250, 3450, 3650, 3625, 4675, 3200, 380…
```

---

### Paso 1: Vamos a dividir el set de datos

```r
library(rsample)
set.seed(123) #setear la semilla
p_split <- penguins %>%
  initial_split(prop=0.75) # divido en 75%

p_train <- training(p_split)

p_split
```

```
## <Analysis/Assess/Total>
## <250/83/333>
```

```r
# para hacer validación cruzada estratificada
p_folds <- vfold_cv(p_train, strata = species) 
```

Estos son los datos de entrenamiento/prueba/total

* __Vamos a _entrenar_ con 250 muestras__
* __Vamos a _testear_ con 83 muestras__
* __Datos totales: 333__

---

### Paso 2: Preprocesamiento (receta)

```r
#creo la receta
recipe_dt <- p_train %>%
  recipe(species~.) %>%
  step_corr(all_predictors()) %>% #elimino las correlaciones
  step_center(all_predictors(), -all_outcomes()) %>% #centrado
  step_scale(all_predictors(), -all_outcomes()) %>% #escalado
  prep() 
recipe_dt #ver la receta
```

```
## Data Recipe
## 
## Inputs:
## 
##       role #variables
##    outcome          1
##  predictor          4
## 
## Training data contained 250 data points and no missing data.
## 
## Operations:
## 
## Correlation filter removed no terms [trained]
## Centering for bill_length_mm, ... [trained]
## Scaling for bill_length_mm, ... [trained]
```

---
background-image: url(img/dt-fondo.png)
background-size: cover

### Paso 3: Especificar el modelo  🌳

__Modelo de árboles de decisión__

__Vamos a utilizar el modelo por defecto__

```r
#especifico el modelo 
set.seed(123)
vanilla_tree_spec <- decision_tree() %>% #arboles de decisión
  set_engine("rpart") %>% #librería rpart
  set_mode("classification") #modo para clasificar
vanilla_tree_spec
```

```
## Decision Tree Model Specification (classification)
## 
## Computational engine: rpart
```

---
background-image: url(img/dt-fondo.png)
background-size: cover

### Paso 4: armo el workflow

```r
#armo el workflow
tree_wf <- workflow() %>%
  add_recipe(recipe_dt) %>% #agrego la receta
  add_model(vanilla_tree_spec) #agrego el modelo
tree_wf 
```

```
## ══ Workflow ═══════════════════════════════════════════════════════════════════════════════════════════
## Preprocessor: Recipe
## Model: decision_tree()
## 
## ── Preprocessor ───────────────────────────────────────────────────────────────────────────────────────
## 3 Recipe Steps
## 
## ● step_corr()
## ● step_center()
## ● step_scale()
## 
## ── Model ──────────────────────────────────────────────────────────────────────────────────────────────
## Decision Tree Model Specification (classification)
## 
## Computational engine: rpart
```

---

### Paso 5: Ajuste de la función

```r
#modelo vanilla sin tunning
set.seed(123) 
vanilla_tree_spec %>% 
  fit_resamples(species ~ ., 
                resamples = p_folds) %>% 
  collect_metrics() #desanidar las metricas
```

```
## # A tibble: 2 x 5
##   .metric  .estimator  mean     n std_err
##   <chr>    <chr>      <dbl> <int>   <dbl>
## 1 accuracy multiclass 0.953    10  0.0154
## 2 roc_auc  hand_till  0.946    10  0.0197
```

---

#### Paso 5.1: vamos a especificar 2 hiperparámatros

```r
set.seed(123) 
trees_spec <- decision_tree() %>% 
  set_engine("rpart") %>% 
  set_mode("classification") %>% 
  set_args(min_n = 20, cost_complexity = 0.1) #especifico hiperparámetros

trees_spec %>%
  fit_resamples(species ~ ., 
                resamples = p_folds) %>% 
  collect_metrics()
```

---

# Ejercicio

.bg-near-white.b--purple.ba.bw2.br3.shadow-5.ph4.mt5[
#### 1. ¿Por qué es el mismo valor obtenido en los dos casos?

#### 2. Dejando fijo el valor de min_n=20, pruebe C=1, C=0.5 y C=0.
#### 3. Dejando fijo el valor de C=0, pruebe min_n 1 y 5.
]

---

### Paso 6: Predicción del modelo  🔮

```r
#utilizamos la funcion last_fit junto al workflow y al split de datos
final_fit_dt <- last_fit(tree_wf,
  split = p_split
)
final_fit_dt %>%
  collect_metrics()
```

```
## # A tibble: 2 x 3
##   .metric  .estimator .estimate
##   <chr>    <chr>          <dbl>
## 1 accuracy multiclass     0.916
## 2 roc_auc  hand_till      0.929
```

---

### Paso 6.1: matriz de confusión

```r
final_fit_dt %>%
  collect_predictions() %>%
  conf_mat(species, .pred_class) #para ver la matriz de confusión
```

```
##            Truth
## Prediction  Adelie Chinstrap Gentoo
##   Adelie        33         2      0
##   Chinstrap      2        16      0
##   Gentoo         2         1     27
```

```r
final_fit_dt %>%
  collect_predictions() %>%
  sens(species, .pred_class) #sensibilidad global del modelo
```

```
## # A tibble: 1 x 3
##   .metric .estimator .estimate
##   <chr>   <chr>          <dbl>
## 1 sens    macro          0.911
```

---

# Ejercicio

.bg-near-white.b--purple.ba.bw2.br3.shadow-5.ph4.mt5[
#### Repetir estos pasos para el modelo de C=0 y min_n=5
]

---

## Resumiendo

* initial_split()

]

]

---

## Resumiendo

* step_*()

]

]

---

## Resumiendo

* set_engine()
* mode()

]

]

---

## Resumiendo

* workflow()
* add_recipe()
* add_model()

]

]

---

## Resumiendo

* last_fit()
]

]

---

## Resumiendo

* collect_metrics()
* collect_predictions() + conf_mat()

]

]

---

# Random Forest

---
background-image: url(img/rf-fondo.png)
background-size: cover
  
  
### Paso 2: Preprocesamos los datos

```r
p_recipe <- training(p_split) %>%
  recipe(species~.) %>%
  step_corr(all_predictors()) %>%
  step_center(all_predictors(), -all_outcomes()) %>%
  step_scale(all_predictors(), -all_outcomes()) %>%
  prep()
p_recipe
```

---

### Paso 3: Especificar el modelo

```r
rf_spec <- rand_forest() %>% 
  set_engine("ranger") %>% 
  set_mode("classification")
```

---

### Veamos como funciona el modelo sin tunning

```r
set.seed(123)

rf_spec %>% 
  fit_resamples(species ~ ., 
                resamples = p_folds) %>% 
  collect_metrics()
```

```
## # A tibble: 2 x 5
##   .metric  .estimator  mean     n std_err
##   <chr>    <chr>      <dbl> <int>   <dbl>
## 1 accuracy multiclass 0.972    10 0.0108 
## 2 roc_auc  hand_till  0.996    10 0.00192
```

---
background-image: url(img/rf-fondo.png)
background-size: cover

## Random Forest con mtry=2  🔧

```r
rf2_spec <- rf_spec %>% 
  set_args(mtry = 2)

set.seed(123)

rf2_spec %>% 
  fit_resamples(species ~ ., 
                resamples = p_folds) %>% 
  collect_metrics()
```

---
background-image: url(img/rf-fondo.png)
background-size: cover

## Random Forest con mtry=3  🔧

```r
rf3_spec <- rf_spec %>% 
  set_args(mtry = 3)

set.seed(123)

rf3_spec %>% 
  fit_resamples(species ~ ., 
                resamples = p_folds) %>% 
  collect_metrics()
```

```
## # A tibble: 2 x 5
##   .metric  .estimator  mean     n std_err
##   <chr>    <chr>      <dbl> <int>   <dbl>
## 1 accuracy multiclass 0.967    10 0.0104 
## 2 roc_auc  hand_till  0.997    10 0.00115
```

---

## Random Forest con mtry=4  🔧

```r
rf4_spec <- rf_spec %>% 
  set_args(mtry = 4)

set.seed(123)
rf4_spec %>% 
  fit_resamples(species ~ ., 
                resamples = p_folds) %>% 
  collect_metrics()
```

```
## # A tibble: 2 x 5
##   .metric  .estimator  mean     n std_err
##   <chr>    <chr>      <dbl> <int>   <dbl>
## 1 accuracy multiclass 0.964    10 0.00972
## 2 roc_auc  hand_till  0.996    10 0.00147
```

---
background-image: url(img/rf-fondo.png)
background-size: cover

### Tuneo de hiperparámetros automático 🔧

```r
tune_spec <- rand_forest(
  mtry = tune(),
  trees = 1000,
  min_n = tune()
) %>%
  set_mode("classification") %>%
  set_engine("ranger")
tune_spec
```

```
## Random Forest Model Specification (classification)
## 
## Main Arguments:
##   mtry = tune()
##   trees = 1000
##   min_n = tune()
## 
## Computational engine: ranger
```

---
background-image: url(img/rf-fondo.png)
background-size: cover

## Workflows

```r
tune_wf <- workflow() %>%
  add_recipe(p_recipe) %>%
  add_model(tune_spec)

set.seed(123)
cv_folds <- vfold_cv(p_train, strata = species)
tune_wf
```

```
## ══ Workflow ═══════════════════════════════════════════════════════════════════════════════════════════
## Preprocessor: Recipe
## Model: rand_forest()
## 
## ── Preprocessor ───────────────────────────────────────────────────────────────────────────────────────
## 3 Recipe Steps
## 
## ● step_corr()
## ● step_center()
## ● step_scale()
## 
## ── Model ──────────────────────────────────────────────────────────────────────────────────────────────
## Random Forest Model Specification (classification)
## 
## Main Arguments:
##   mtry = tune()
##   trees = 1000
##   min_n = tune()
## 
## Computational engine: ranger
```

---

### Paralelizamos los cálculos

```r
doParallel::registerDoParallel()

set.seed(123)
tune_res <- tune_grid(
  tune_wf,
  resamples = cv_folds,
  grid = 20
)

tune_res
```

```
## # Tuning results
## # 10-fold cross-validation using stratification 
## # A tibble: 10 x 4
##    splits           id     .metrics          .notes          
##    <list>           <chr>  <list>            <list>          
##  1 <split [224/26]> Fold01 <tibble [40 × 6]> <tibble [0 × 1]>
##  2 <split [224/26]> Fold02 <tibble [40 × 6]> <tibble [0 × 1]>
##  3 <split [224/26]> Fold03 <tibble [40 × 6]> <tibble [0 × 1]>
##  4 <split [224/26]> Fold04 <tibble [40 × 6]> <tibble [0 × 1]>
##  5 <split [224/26]> Fold05 <tibble [40 × 6]> <tibble [0 × 1]>
##  6 <split [225/25]> Fold06 <tibble [40 × 6]> <tibble [0 × 1]>
##  7 <split [225/25]> Fold07 <tibble [40 × 6]> <tibble [0 × 1]>
##  8 <split [226/24]> Fold08 <tibble [40 × 6]> <tibble [0 × 1]>
##  9 <split [227/23]> Fold09 <tibble [40 × 6]> <tibble [0 × 1]>
## 10 <split [227/23]> Fold10 <tibble [40 × 6]> <tibble [0 × 1]>
```

---

## Rango de valores para min_n y mtry

![](clasificacion_files/figure-html/unnamed-chunk-26-1.png)

---
background-image: url(img/rf-fondo.png)
background-size: cover

## Elijo el mejor modelo 🏆

* Con la función select_best

```r
best_auc <- select_best(tune_res, "roc_auc")

final_rf <- finalize_model(
  tune_spec,
  best_auc
)

final_rf
```

```
## Random Forest Model Specification (classification)
## 
## Main Arguments:
##   mtry = 4
##   trees = 1000
##   min_n = 4
## 
## Computational engine: ranger
```

---

## Valores finales

```r
set.seed(123)
final_wf <- workflow() %>%
  add_recipe(p_recipe) %>%
  add_model(final_rf)

final_res <- final_wf %>%
  last_fit(p_split)

final_res %>%
  collect_metrics()
```

```
## # A tibble: 2 x 3
##   .metric  .estimator .estimate
##   <chr>    <chr>          <dbl>
## 1 accuracy multiclass     0.952
## 2 roc_auc  hand_till      0.999
```

---

## Matriz de Confusión

```r
final_res %>%
  collect_predictions() %>%
  conf_mat(species, .pred_class)
```

```
##            Truth
## Prediction  Adelie Chinstrap Gentoo
##   Adelie        39         0      1
##   Chinstrap      2        11      1
##   Gentoo         0         0     29
```

---

## Ejercicio 3

.bg-near-white.b--purple.ba.bw2.br3.shadow-5.ph4.mt5[
#### Con el set de datos de iris realice una clasificación con random forest.

]

---

## Resumiendo

]

---

## Resumiendo

]
---

## Resumiendo

]

---

## Resumiendo

]

]

---

## Resumiendo

]

---

### Importancia de las variables

* libreria vip

```r
library(vip)
set.seed(123)
final_rf %>%
  set_engine("ranger", importance = "permutation") %>%
  fit(species ~ .,
      data = juice(p_recipe)) %>%
  vip(geom = "point")+
  theme_xaringan()
```

---

### Gráfico 
![](clasificacion_files/figure-html/unnamed-chunk-36-1.png)

---

### Ejemplos (un poco) más reales

#### [Tuning random forest hyperparameters with #TidyTuesday trees data](https://juliasilge.com/blog/sf-trees-random-tuning/)

<br><br><br><br>
#### [Hyperparameter tuning and #TidyTuesday food consumption](https://juliasilge.com/blog/food-hyperparameter-tune/)

---

# Bibliografía

#### Max Kuhn & Julia Silge - [Tidy Modeling (en desarrollo)](https://www.tmwr.org/)

#### Julia Silge's [Personal Blog](https://juliasilge.com/blog/)

#### Max Kuhn & Kjell Johnson - [Feature engineering and Selection: A Practical Approach for Predictive Models](http://www.feat.engineering/)

#### Max Kuhn & Kjell Johnson - Applied Predictive Modeling

#### Documentación de [`tidymodels`](https://www.tidymodels.org/)

#### Alison Hill [rstudio conf](https://conf20-intro-ml.netlify.app/materials/) y [curso virtual](https://alison.rbind.io/post/2020-06-02-tidymodels-virtually/) material

---

# Muchas gracias 🤖