Extras - Recipes

Introduction to tidymodels

Looking at the predictors

forested_train
#> # A tibble: 5,685 × 19
#>    forested  year elevation eastness northness roughness tree_no_tree dew_temp precip_annual temp_annual_mean temp_annual_min temp_annual_max temp_january_min vapor_min vapor_max canopy_cover   lon
#>    <fct>    <dbl>     <dbl>    <dbl>     <dbl>     <dbl> <fct>           <dbl>         <dbl>            <dbl>           <dbl>           <dbl>            <dbl>     <dbl>     <dbl>        <dbl> <dbl>
#>  1 No        2016       464       -5       -99         7 No tree          1.71           282             9.76           -4.44           16.6              2.96       191      1534            4 -121.
#>  2 Yes       2016       166       92        37         7 Tree             6             1298            10.2             0.72           14.3              6.12        60       747           33 -122.
#>  3 No        2016       644      -85       -52        24 No tree          0.67           288             8.77           -6.32           14.6              2.98       219      1396            0 -120.
#>  4 Yes       2014      1285        4        99        79 Tree             1.91          1621             5.61           -2.48            9.48             1.73        88       545           74 -123.
#>  5 Yes       2013       822       87        48        68 Tree             1.95          2200             8.62           -0.68           12.9              4.35       147       861           48 -121.
#>  6 Yes       2017         3        6       -99         5 Tree             7.93          2211            10.6             3.77           14.2              7.02        34       578           79 -124.
#>  7 Yes       2014      2041      -95        28        49 Tree            -4.22          1551             0.75           -9.47            5.17            -3.66        73       481           48 -120.
#>  8 Yes       2015      1009       -8        99        72 Tree             1.72          2396             6.59           -2.98           11.3              1.88        92       781           76 -122.
#>  9 No        2017       436      -98        19        10 No tree          1.8            234             9.8            -4.23           16.3              3.32       178      1527            0 -119.
#> 10 No        2018       775       63        76       103 No tree          0.62           432             8.51           -5.5            13.7              3.32       241      1237            7 -120.
#> # ℹ 5,675 more rows
#> # ℹ 2 more variables: lat <dbl>, land_type <fct>

Working with other models

Some models can’t handle non-numeric data

  • Linear Regression
  • K Nearest Neighbors


Some models struggle if numeric predictors aren’t scaled

  • K Nearest Neighbors
  • Anything using gradient descent

Types of needed preprocessing

  • Do qualitative predictors require a numeric encoding?

  • Should columns with a single unique value be removed?

  • Does the model struggle with missing data?

  • Does the model struggle with correlated predictors?

  • Should predictors be centered and scaled?

  • Is it helpful to transform predictors to be more symmetric?

https://www.tmwr.org/pre-proc-table.html

Two types of preprocessing

Two types of preprocessing

General definitions

  • Data preprocessing is what you do to make your model successful.
  • Feature engineering is what you do to the original predictors to make the model do the least work to perform great.

Working with dates

Datetime variables are automatically converted to an integer if given as a raw predictor. To avoid this, it can be re-encoded as:

  • Days since a reference date
  • Day of the week
  • Month
  • Year
  • Leap year
  • Indicators for holidays

Two types of transformations


Static

  • Square root, log, inverse
  • Dummies for known levels
  • Date time extractions

Trained

  • Centering & scaling
  • Imputation
  • PCA
  • Anything for unknown factor levels

Trained methods need to calculate sufficient information to be applied again.

The recipes package

  • Modular + extensible
  • Works well with pipes ,|> and %>%
  • Deferred evaluation
  • Isolates test data from training data
  • Can do things formulas can’t

How to write a recipe

forested_rec <- recipe(forested ~ ., data = forested_train) %>%
  step_dummy(all_nominal_predictors()) %>%
  step_zv(all_predictors()) %>%
  step_log(canopy_cover, offset = 0.5) %>%
  step_normalize(all_numeric_predictors())

How to write a recipe

forested_rec <- recipe(forested ~ ., data = forested_train) %>%
  step_dummy(all_nominal_predictors()) %>%
  step_zv(all_predictors()) %>%
  step_log(canopy_cover, offset = 0.5) %>%
  step_normalize(all_numeric_predictors())


Start by calling recipe() to denote the data source and variables used.

How to write a recipe

forested_rec <- recipe(forested ~ ., data = forested_train) %>%
  step_dummy(all_nominal_predictors()) %>%
  step_zv(all_predictors()) %>%
  step_log(canopy_cover, offset = 0.5) %>%
  step_normalize(all_numeric_predictors())


Specify what actions to take by adding step_*()s.

How to write a recipe

forested_rec <- recipe(forested ~ ., data = forested_train) %>%
  step_dummy(all_nominal_predictors()) %>%
  step_zv(all_predictors()) %>%
  step_log(canopy_cover, offset = 0.5) %>%   step_normalize(all_numeric_predictors())


Use {tidyselect} and recipes-specific selectors to denote affected variables.

Using a recipe

forested_rec <- recipe(forested ~ ., data = forested_train) %>%
  step_dummy(all_nominal_predictors()) %>%
  step_zv(all_predictors()) %>%
  step_log(canopy_cover, offset = 0.5) %>%   step_normalize(all_numeric_predictors())


Save the recipe we like so that we can use it in various places, e.g., with different models.


Using a recipe with workflows

Recipes are typically combined with a model in a workflow() object:


forested_wflow <- workflow() %>%
  add_recipe(forested_rec) %>%
  add_model(linear_reg())

Recipes are estimated

Every preprocessing step in a recipe that involved calculations uses the training set. For example:

  • Levels of a factor
  • Determination of zero-variance
  • Normalization
  • Feature extraction

Once a recipe is added to a workflow, this occurs when fit() is called.

Debugging a recipe

  • Typically, you will want to use a workflow to estimate and apply a recipe.
  • If you have an error and need to debug your recipe, the original recipe object (e.g. forested_rec) can be estimated manually with a function called prep(). It is analogous to fit(). See TMwR section 16.4.
  • Another function, bake(), is analogous to predict(), and gives you the processed data back.

Your turn


Take the recipe and prep() then bake() it to see what the resulting data set looks like.

Try removing steps to see how the result changes.


05:00

Printing a recipe

forested_rec
#> 
#> ── Recipe ────────────────────────────────────────────────────────────
#> 
#> ── Inputs
#> Number of variables by role
#> outcome:    1
#> predictor: 18
#> 
#> ── Operations
#> • Dummy variables from: all_nominal_predictors()
#> • Zero variance filter on: all_predictors()
#> • Log transformation on: canopy_cover
#> • Centering and scaling for: all_numeric_predictors()

Prepping a recipe

prep(forested_rec)
#> 
#> ── Recipe ────────────────────────────────────────────────────────────
#> 
#> ── Inputs
#> Number of variables by role
#> outcome:    1
#> predictor: 18
#> 
#> ── Training information
#> Training data contained 5685 data points and no incomplete rows.
#> 
#> ── Operations
#> • Dummy variables from: tree_no_tree and land_type | Trained
#> • Zero variance filter removed: <none> | Trained
#> • Log transformation on: canopy_cover | Trained
#> • Centering and scaling for: year and elevation, ... | Trained

Baking a recipe

prep(forested_rec) %>%
  bake(new_data = forested_train)
#> # A tibble: 5,685 × 20
#>      year elevation eastness northness roughness dew_temp precip_annual temp_annual_mean temp_annual_min temp_annual_max temp_january_min vapor_min vapor_max canopy_cover     lon    lat forested
#>     <dbl>     <dbl>    <dbl>     <dbl>     <dbl>    <dbl>         <dbl>            <dbl>           <dbl>           <dbl>            <dbl>     <dbl>     <dbl>        <dbl>   <dbl>  <dbl> <fct>   
#>  1  0.206   -0.450   -0.0203    -1.38    -0.874   -0.169         -0.864          0.532           -0.403           0.959           -0.0702     0.755     1.21        -0.566 -0.175  -0.988 No      
#>  2  0.206   -1.07     1.38       0.563   -0.874    1.33           0.132          0.726            1.24            0.159            1.34      -1.00     -1.00         0.494 -0.856   1.10  Yes     
#>  3  0.206   -0.0762  -1.17      -0.711   -0.506   -0.533         -0.858          0.116           -1.00            0.248           -0.0613     1.13      0.820       -1.73   0.270   0.175 No      
#>  4 -0.413    1.26     0.109      1.45     0.683   -0.0987         0.448         -1.21             0.221          -1.56            -0.621     -0.625    -1.57         0.917 -1.52    0.654 Yes     
#>  5 -0.723    0.294    1.31       0.721    0.445   -0.0847         1.02           0.0529           0.795          -0.346            0.552      0.166    -0.683        0.690 -0.419   1.32  Yes     
#>  6  0.516   -1.41     0.138     -1.38    -0.917    2.01           1.03           0.894            2.21            0.127            1.75      -1.35     -1.48         0.951 -1.84   -0.338 Yes     
#>  7 -0.413    2.83    -1.32       0.435    0.0343  -2.25           0.380         -3.26            -2.01           -3.09            -3.03      -0.826    -1.75         0.690 -0.0159  1.65  Yes     
#>  8 -0.103    0.682   -0.0636     1.45     0.532   -0.165          1.21          -0.801            0.0620         -0.915           -0.554     -0.571    -0.908        0.931 -0.636  -1.33  Yes     
#>  9  0.516   -0.508   -1.36       0.306   -0.809   -0.137         -0.911          0.549           -0.336           0.856            0.0909     0.581     1.19        -1.73   0.760  -0.209 No      
#> 10  0.826    0.196    0.960      1.12     1.20    -0.550         -0.717          0.00667         -0.741          -0.0616           0.0909     1.43      0.373       -0.296  0.155  -0.204 No      
#> # ℹ 5,675 more rows
#> # ℹ 3 more variables: tree_no_tree_No.tree <dbl>, land_type_Non.tree.vegetation <dbl>, land_type_Tree <dbl>

Tidying a recipe

Once a recipe as been estimated, there are various bits of information saved in it.

  • The tidy() function can be used to get specific results from the recipe.

Your turn

Take a prepped recipe and use the tidy() function on it.

Use the number argument to inspect different steps.


05:00

Tidying a recipe

prep(forested_rec) %>%
  tidy()
#> # A tibble: 4 × 6
#>   number operation type      trained skip  id             
#>    <int> <chr>     <chr>     <lgl>   <lgl> <chr>          
#> 1      1 step      dummy     TRUE    FALSE dummy_jlmcG    
#> 2      2 step      zv        TRUE    FALSE zv_mYCvS       
#> 3      3 step      log       TRUE    FALSE log_eme6b      
#> 4      4 step      normalize TRUE    FALSE normalize_ScVef

Tidying a recipe

prep(forested_rec) %>%
  tidy(number = 1)
#> # A tibble: 3 × 3
#>   terms        columns             id         
#>   <chr>        <chr>               <chr>      
#> 1 tree_no_tree No tree             dummy_jlmcG
#> 2 land_type    Non-tree vegetation dummy_jlmcG
#> 3 land_type    Tree                dummy_jlmcG

Using a recipe in tidymodels

The recommended way to use a recipe in tidymodels is to use it as part of a workflow().

forested_wflow <- workflow() %>%  
  add_recipe(forested_rec) %>%  
  add_model(linear_reg())

When used in this way, you don’t need to worry about prep() and bake() as it is handled for you.

More information

https://workshops.tidymodels.org