tidymodels

# tidymodels
### Dr. D’Agostino McGowan

---

<div class="my-footer">
<span>
Dr. Lucy D'Agostino McGowan <i> adapted from Alison Hill's Introduction to ML with the Tidyverse</i>
</span>
</div>

---

## tidymodels

- tidymodels is an opinionated collection of R packages designed for modeling and statistical analysis.
- All packages share an underlying philosophy and a common grammar.
]

---

## Step 1: Specify the model

* Pick the **model**
--

* Set the **engine**

---

## Specify the model

```r
linear_reg() %>%
  set_engine("lm")
```

---

## Specify the model

```r
linear_reg() %>%
  set_engine("glmnet")
```

---

## Specify the model

```r
linear_reg() %>%
  set_engine("spark")
```

---

## Specify the model

```r
decision_tree() %>%
  set_engine("ranger")
```

---

## Specify the model

* All available models:

https://www.tidymodels.org

---

## <svg style="height:0.8em;top:.04em;position:relative;" viewBox="0 0 640 512"><path d="M512 64v256H128V64h384m16-64H112C85.5 0 64 21.5 64 48v288c0 26.5 21.5 48 48 48h416c26.5 0 48-21.5 48-48V48c0-26.5-21.5-48-48-48zm100 416H389.5c-3 0-5.5 2.1-5.9 5.1C381.2 436.3 368 448 352 448h-64c-16 0-29.2-11.7-31.6-26.9-.5-2.9-3-5.1-5.9-5.1H12c-6.6 0-12 5.4-12 12v36c0 26.5 21.5 48 48 48h544c26.5 0 48-21.5 48-48v-36c0-6.6-5.4-12-12-12z"/></svg> `Specify Model`

Write a pipe that creates a model that uses `lm()` to fit a linear regression using tidymodels. Save it as `lm_spec` and look at the object. What does it return?

_Hint: you'll need  https://www.tidymodels.org_

---

```r
lm_spec <- 
  linear_reg() %>% # Pick linear regression
  set_engine(engine = "lm") # set engine
lm_spec
```

```
## Linear Regression Model Specification (regression)
## 
## Computational engine: lm
```

---

## Fit the data

* You can train your model using the `fit()` function

```r
fit(lm_spec,
    mpg ~ horsepower,
    data = Auto)
```

```
## parsnip model object
## 
## Fit time:  7ms 
## 
## Call:
## stats::lm(formula = mpg ~ horsepower, data = data)
## 
## Coefficients:
## (Intercept)   horsepower  
##     39.9359      -0.1578
```

---

Fit the model:

```r
library(ISLR)
lm_fit <- fit(lm_spec,
              mpg ~ horsepower,
              data = Auto)
lm_fit
```

Does this give the same results as

```r
lm(mpg ~ horsepower, data = Auto)
```

---

## Get predictions

```r
lm_fit %>%
  predict(new_data = Auto)
```

* Uses the `predict()` function
--

* ‼️ `new_data` has an underscore
--

* 😄 This automagically creates a data frame

---

## Get predictions

```r
lm_fit %>%
  predict(new_data = Auto) %>%
  bind_cols(Auto)
```

```
## # A tibble: 392 x 10
##    .pred   mpg cylinders displacement horsepower weight acceleration  year
##  * <dbl> <dbl>     <dbl>        <dbl>      <dbl>  <dbl>        <dbl> <dbl>
##  1 19.4     18         8          307        130   3504         12      70
##  2 13.9     15         8          350        165   3693         11.5    70
##  3 16.3     18         8          318        150   3436         11      70
##  4 16.3     16         8          304        150   3433         12      70
##  5 17.8     17         8          302        140   3449         10.5    70
##  6  8.68    15         8          429        198   4341         10      70
##  7  5.21    14         8          454        220   4354          9      70
##  8  6.00    14         8          440        215   4312          8.5    70
##  9  4.42    14         8          455        225   4425         10      70
## 10  9.95    15         8          390        190   3850          8.5    70
## # … with 382 more rows, and 2 more variables: origin <dbl>, name <fct>
```

---

Edit the code below to add the original data to the predicted data.

```r
mpg_pred <- lm_fit %>% 
  predict(new_data = Auto) %>% 
  ---
```

---

## Get predictions

```r
mpg_pred <- lm_fit %>%
  predict(new_data = Auto) %>%
  bind_cols(Auto)

mpg_pred
```

---

## Calculate the error

* Root mean square error

```r
mpg_pred %>%
  rmse(truth = mpg, estimate = .pred)
```

```
## # A tibble: 1 x 3
##   .metric .estimator .estimate
##   <chr>   <chr>          <dbl>
## 1 rmse    standard        4.89
```

---

## Validation set approach

```r
Auto_split <- initial_split(Auto, prop = 0.5)
Auto_split
```

```
## <Analysis/Assess/Total>
## <196/196/392>
```

* Extract the training and testing data

```r
training(Auto_split)
testing(Auto_split)
```

---

## Validation set approach

```r
Auto_train <- training(Auto_split)
```

```r
Auto_train
```

```
## # A tibble: 196 x 9
##      mpg cylinders displacement horsepower weight acceleration  year origin
##    <dbl>     <dbl>        <dbl>      <dbl>  <dbl>        <dbl> <dbl>  <dbl>
##  1    14         8          454        220   4354          9      70      1
##  2    15         8          383        170   3563         10      70      1
##  3    14         8          340        160   3609          8      70      1
##  4    14         8          455        225   3086         10      70      1
##  5    24         4          113         95   2372         15      70      3
##  6    18         6          199         97   2774         15.5    70      1
##  7    21         6          200         85   2587         16      70      1
##  8    25         4          110         87   2672         17.5    70      2
##  9    24         4          107         90   2430         14.5    70      2
## 10    25         4          104         95   2375         17.5    70      2
## # … with 186 more rows, and 1 more variable: name <fct>
```
]

---

Copy the code below, fill in the blanks to fit a model on the **training** data then calculate the **test** RMSE.

```r
set.seed(100)
Auto_split  <- ________
Auto_train  <- ________
Auto_test   <- ________
lm_fit      <- fit(lm_spec, 
                   mpg ~ horsepower, 
                   data = ________)
mpg_pred  <- ________ %>% 
  predict(new_data = ________) %>% 
  bind_cols(________)
rmse(________, truth = ________, estimate = ________)
```

---

## A faster way!

* You can use `last_fit()` and specify the split
* This will automatically train the data on the `train` data from the split
* Instead of specifying which metric to calculate (with `rmse` as before) you can just use `collect_metrics()` and it will automatically calculate the metrics on the `test` data from the split

```r
set.seed(100)

Auto_split <- initial_split(Auto, prop = 0.5)
lm_fit <- last_fit(lm_spec,
                   mpg ~ horsepower,
*                  split = Auto_split)

lm_fit %>%
* collect_metrics()
```

```
## # A tibble: 2 x 3
##   .metric .estimator .estimate
##   <chr>   <chr>          <dbl>
## 1 rmse    standard       4.87 
## 2 rsq     standard       0.625
```

---

## What about cross validation?

```r
Auto_cv <- vfold_cv(Auto, v = 5)
Auto_cv
```

```
## #  5-fold cross-validation 
## # A tibble: 5 x 2
##   splits           id   
##   <list>           <chr>
## 1 <split [313/79]> Fold1
## 2 <split [313/79]> Fold2
## 3 <split [314/78]> Fold3
## 4 <split [314/78]> Fold4
## 5 <split [314/78]> Fold5
```

---

## What about cross validation?

```r
*fit_resamples(lm_spec,
              mpg ~ horsepower,
*             resamples = Auto_cv)
```

---

## What about cross validation?

```r
fit_resamples(lm_spec,
              mpg ~ horsepower,
              resamples = Auto_cv)
```

```
## #  5-fold cross-validation 
## # A tibble: 5 x 4
##   splits           id    .metrics         .notes          
##   <list>           <chr> <list>           <list>          
## 1 <split [313/79]> Fold1 <tibble [2 × 3]> <tibble [0 × 1]>
## 2 <split [313/79]> Fold2 <tibble [2 × 3]> <tibble [0 × 1]>
## 3 <split [314/78]> Fold3 <tibble [2 × 3]> <tibble [0 × 1]>
## 4 <split [314/78]> Fold4 <tibble [2 × 3]> <tibble [0 × 1]>
## 5 <split [314/78]> Fold5 <tibble [2 × 3]> <tibble [0 × 1]>
```

---

## What about cross validation?

```r
results <- fit_resamples(lm_spec,
                         mpg ~ horsepower,
                         resamples = Auto_cv)

results %>%
  collect_metrics()
```

```
## # A tibble: 2 x 5
##   .metric .estimator  mean     n std_err
##   <chr>   <chr>      <dbl> <int>   <dbl>
## 1 rmse    standard   4.93      5  0.0779
## 2 rsq     standard   0.611     5  0.0277
```

---

Edit the code below to get the 5-fold cross validation error rate for the following model:

`\(mpg = \beta_0 + \beta_1 horsepower + \beta_2 horsepower^2+ \epsilon\)`

```r
Auto_cv <- vfold_cv(Auto, v = 5)

results <- fit_resamples(lm_spec,
                         ----,
                         resamples = ---)

results %>%
  collect_metrics()
```

* What do you think `rsq` is?

---