forested_train
#> # A tibble: 8,749 × 19
#> forested year elevation eastness 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 lat
#> <fct> <dbl> <dbl> <dbl> <dbl> <fct> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Yes 1997 66 82 10 Tree 12.2 1315 18.4 1.88 24.9 11.8 121 1888 66 -84.9 32.4
#> 2 No 1997 284 -99 58 Tree 10.3 1236 16.1 -0.26 22.3 9.92 68 1586 80 -85.0 34.1
#> 3 Yes 2022 130 86 15 Tree 11.8 1194 17.6 1.3 24.1 11.2 61 1753 96 -83.0 33.1
#> 4 Yes 2021 202 -55 3 Tree 10.7 1235 16.6 0.05 23.0 10.2 72 1682 65 -83.0 33.9
#> 5 Yes 1995 75 -89 1 Tree 13.8 1256 19.2 3.63 25.5 12.9 57 1796 88 -83.4 31.3
#> 6 No 1995 110 -53 5 Tree 12.4 1236 18.6 2.53 24.8 12.4 102 1835 51 -83.9 32.2
#> 7 Yes 2022 111 73 12 Tree 11.5 1168 17.4 1 24.0 10.9 67 1772 84 -82.2 33.6
#> 8 Yes 1997 230 96 14 Tree 9.98 1373 15.4 -1.35 21.8 9.03 46 1552 68 -85.3 34.8
#> 9 Yes 2002 160 -88 13 Tree 11.1 1219 16.9 0.07 23.6 10.2 53 1731 95 -83.2 33.6
#> 10 Yes 2020 39 9 6 Tree 13.9 1237 19.2 3.25 25.6 12.8 58 1812 86 -82.0 31.7
#> # ℹ 8,739 more rows
#> # ℹ 2 more variables: land_type <fct>, county <fct>Extras - Recipes
Introduction to Machine Learning in R with tidymodels
Looking at the predictors
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?
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 calledprep(). It is analogous tofit(). See TMwR section 16.4.
- Another function,
bake(), is analogous topredict(), 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 8749 data points and no incomplete rows.
#>
#> ── Operations
#> • Dummy variables from: tree_no_tree, land_type, county | Trained
#> • Zero variance filter removed: <none> | Trained
#> • Log transformation on: canopy_cover | Trained
#> • Centering and scaling for: year elevation, ... | TrainedBaking a recipe
prep(forested_rec) |>
bake(new_data = forested_train)
#> # A tibble: 8,749 × 177
#> year elevation eastness 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> <fct>
#> 1 -1.15 -0.521 1.13 0.0463 -0.232 0.178 0.172 -0.107 0.338 -0.00909 3.10 1.01 0.407 -1.44 -0.0184 Yes
#> 2 -1.15 1.10 -1.50 3.91 -1.39 -0.568 -1.31 -1.33 -1.28 -1.28 0.0408 -0.941 0.652 -1.51 1.36 No
#> 3 0.892 -0.0440 1.19 0.449 -0.422 -0.965 -0.305 -0.438 -0.170 -0.438 -0.363 0.140 0.884 0.331 0.556 Yes
#> 4 0.810 0.493 -0.858 -0.518 -1.11 -0.577 -1.01 -1.15 -0.878 -1.10 0.272 -0.320 0.388 0.331 1.26 Yes
#> 5 -1.31 -0.454 -1.35 -0.679 0.748 -0.379 0.701 0.893 0.683 0.702 -0.594 0.418 0.773 -0.0643 -0.992 Yes
#> 6 -1.31 -0.193 -0.829 -0.357 -0.0546 -0.568 0.331 0.264 0.281 0.366 2.00 0.670 0.0802 -0.468 -0.237 No
#> 7 0.892 -0.186 1.00 0.207 -0.631 -1.21 -0.444 -0.609 -0.264 -0.619 -0.0169 0.262 0.714 1.01 0.929 Yes
#> 8 -1.15 0.702 1.34 0.369 -1.56 0.726 -1.77 -1.95 -1.60 -1.87 -1.23 -1.16 0.445 -1.77 2.03 Yes
#> 9 -0.740 0.180 -1.34 0.288 -0.858 -0.729 -0.801 -1.14 -0.515 -1.08 -0.825 -0.00273 0.870 0.122 0.988 Yes
#> 10 0.729 -0.722 0.0719 -0.276 0.816 -0.559 0.708 0.676 0.745 0.635 -0.536 0.521 0.744 1.22 -0.653 Yes
#> # ℹ 8,739 more rows
#> # ℹ 161 more variables: tree_no_tree_No.tree <dbl>, land_type_Non.tree.vegetation <dbl>, land_type_Tree <dbl>, county_Atkinson <dbl>, county_Bacon <dbl>, county_Baker <dbl>, county_Baldwin <dbl>,
#> # county_Banks <dbl>, county_Barrow <dbl>, county_Bartow <dbl>, county_Ben.Hill <dbl>, county_Berrien <dbl>, county_Bibb <dbl>, county_Bleckley <dbl>, county_Brantley <dbl>, county_Brooks <dbl>,
#> # county_Bryan <dbl>, county_Bulloch <dbl>, county_Burke <dbl>, county_Butts <dbl>, county_Calhoun <dbl>, county_Camden <dbl>, county_Candler <dbl>, county_Carroll <dbl>, county_Catoosa <dbl>,
#> # county_Charlton <dbl>, county_Chatham <dbl>, county_Chattahoochee <dbl>, county_Chattooga <dbl>, county_Cherokee <dbl>, county_Clarke <dbl>, county_Clay <dbl>, county_Clayton <dbl>,
#> # county_Clinch <dbl>, county_Cobb <dbl>, county_Coffee <dbl>, county_Colquitt <dbl>, county_Columbia <dbl>, county_Cook <dbl>, county_Coweta <dbl>, county_Crawford <dbl>, county_Crisp <dbl>,
#> # county_Dade <dbl>, county_Dawson <dbl>, county_Decatur <dbl>, county_DeKalb <dbl>, county_Dodge <dbl>, county_Dooly <dbl>, county_Dougherty <dbl>, county_Douglas <dbl>, county_Early <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
Tidying a recipe
prep(forested_rec) |>
tidy(number = 1)
#> # A tibble: 161 × 3
#> terms columns id
#> <chr> <chr> <chr>
#> 1 tree_no_tree No tree dummy_hIEnQ
#> 2 land_type Non-tree vegetation dummy_hIEnQ
#> 3 land_type Tree dummy_hIEnQ
#> 4 county Atkinson dummy_hIEnQ
#> 5 county Bacon dummy_hIEnQ
#> 6 county Baker dummy_hIEnQ
#> 7 county Baldwin dummy_hIEnQ
#> 8 county Banks dummy_hIEnQ
#> 9 county Barrow dummy_hIEnQ
#> 10 county Bartow dummy_hIEnQ
#> # ℹ 151 more rowsUsing a recipe in tidymodels
The recommended way to use a recipe in tidymodels is to use it as part of a workflow().
When used in this way, you don’t need to worry about prep() and bake() as it is handled for you.