Custom performance metrics

developer tools

Create a new performance metric and integrate it with yardstick functions.

Introduction

To use code in this article, you will need to install the following packages: rlang and tidymodels.

The yardstick package already includes a large number of metrics, but there’s obviously a chance that you might have a custom metric that hasn’t been implemented yet. In that case, you can use a few of the tools yardstick exposes to create custom metrics.

Why create custom metrics? With the infrastructure yardstick provides, you get:

Standardization between your metric and other preexisting metrics
Automatic error handling for types and lengths
Automatic selection of binary / multiclass metric implementations
Support for NA handling
Support for grouped data frames
Support for use alongside other metrics in metric_set()

The implementation for metrics differ slightly depending on whether you are implementing a numeric, class, or class probability metric. Examples for numeric and classification metrics are given below. We would encourage you to look into the implementation of roc_auc() after reading this vignette if you want to work on a class probability metric.

Numeric example: MSE

Mean squared error (sometimes MSE or from here on, mse()) is a numeric metric that measures the average of the squared errors. Numeric metrics are generally the simplest to create with yardstick, as they do not have multiclass implementations. The formula for mse() is:

\[ MSE = \frac{1}{N} \sum_{i=1}^{N} (truth_i - estimate_i) ^ 2 = mean( (truth - estimate) ^ 2) \]

All metrics should have a data frame version, and a vector version. The data frame version here will be named mse(), and the vector version will be mse_vec().

Vector implementation

To start, create the vector version. Generally, all metrics have the same arguments unless the metric requires an extra parameter (such as beta in f_meas()). To create the vector function, you need to do two things:

Create an internal implementation function, mse_impl().
Use that implementation function with check_class_metric()/check_numeric_metric()/check_prob_metric(), yardstick_remove_missing(), and and yardstick_any_missing().

Below, mse_impl() contains the actual implementation of the metric, and takes truth and estimate as arguments along with any metric specific arguments. Optionally case_weights if the calculations supports it.

The yardstick function check_numeric_metric() takes truth, estimate and case_weights, and validates that they are the right type, and are the same length.

The yardstick_remove_missing() and yardstick_any_missing() yardstick functions are used to handle missing values in a consistent way, similarly to how the other metrics handle them. The code below is typically copy pasted from function to function, but certain types of metrics might want to deviate from this pattern.

You are required to supply a case_weights argument to mse_vec() for the functions to work with yardstick. If your metric in question doesn’t support case weights, you can error if they are passed, or simply ignore it.

library(tidymodels)

mse_impl <- function(truth, estimate, case_weights = NULL) {
  mean((truth - estimate) ^ 2)
}

mse_vec <- function(truth, estimate, na_rm = TRUE, case_weights = NULL, ...) {
  check_numeric_metric(truth, estimate, case_weights)

  if (na_rm) {
    result <- yardstick_remove_missing(truth, estimate, case_weights)

    truth <- result$truth
    estimate <- result$estimate
    case_weights <- result$case_weights
  } else if (yardstick_any_missing(truth, estimate, case_weights)) {
    return(NA_real_)
  }

  mse_impl(truth, estimate, case_weights = case_weights)
}

At this point, you’ve created the vector version of the mean squared error metric.

data("solubility_test")

mse_vec(
  truth = solubility_test$solubility, 
  estimate = solubility_test$prediction
)
#> [1] 0.5214438

Intelligent error handling is immediately available.

mse_vec(truth = "apple", estimate = 1)
#> Error in `mse_vec()`:
#> ! `truth` should be a numeric, not a string.

mse_vec(truth = 1, estimate = factor("xyz"))
#> Error in `mse_vec()`:
#> ! `estimate` should be a numeric, not a <factor> object.

NA values are removed if na_rm = TRUE (the default). If na_rm = FALSE and any NA values are detected, then the metric automatically returns NA.

# NA values removed
mse_vec(truth = c(NA, .5, .4), estimate = c(1, .6, .5))
#> [1] 0.01

# NA returned
mse_vec(truth = c(NA, .5, .4), estimate = c(1, .6, .5), na_rm = FALSE)
#> [1] NA

Data frame implementation

The data frame version of the metric should be fairly simple. It is a generic function with a data.frame method that calls the yardstick helper, numeric_metric_summarizer(), and passes along the mse_vec() function to it along with versions of truth and estimate that have been wrapped in rlang::enquo() and then unquoted with !! so that non-standard evaluation can be supported.

library(rlang)

mse <- function(data, ...) {
  UseMethod("mse")
}

mse <- new_numeric_metric(mse, direction = "minimize")

mse.data.frame <- function(data, truth, estimate, na_rm = TRUE, case_weights = NULL, ...) {

  numeric_metric_summarizer(
    name = "mse",
    fn = mse_vec,
    data = data,
    truth = !!enquo(truth),
    estimate = !!enquo(estimate),
    na_rm = na_rm,
    case_weights = !!enquo(case_weights)
  )
}

And that’s it. The yardstick package handles the rest.

mse(solubility_test, truth = solubility, estimate = prediction)

# Error handling
mse(solubility_test, truth = solubility, estimate = factor("xyz"))

Let’s test it out on a grouped data frame.

library(dplyr)

set.seed(1234)
size <- 100
times <- 10

# create 10 resamples
solubility_resampled <- bind_rows(
  replicate(
    n = times,
    expr = sample_n(solubility_test, size, replace = TRUE),
    simplify = FALSE
  ),
  .id = "resample"
)

solubility_resampled %>%
  group_by(resample) %>%
  mse(solubility, prediction)
#> # A tibble: 10 × 4
#>    resample .metric .estimator .estimate
#>    <chr>    <chr>   <chr>          <dbl>
#>  1 1        mse     standard       0.512
#>  2 10       mse     standard       0.454
#>  3 2        mse     standard       0.513
#>  4 3        mse     standard       0.414
#>  5 4        mse     standard       0.543
#>  6 5        mse     standard       0.456
#>  7 6        mse     standard       0.652
#>  8 7        mse     standard       0.642
#>  9 8        mse     standard       0.404
#> 10 9        mse     standard       0.479

Class example: miss rate

Miss rate is another name for the false negative rate, and is a classification metric in the same family as sens() and spec(). It follows the formula:

\[ miss\_rate = \frac{FN}{FN + TP} \]

This metric, like other classification metrics, is more easily computed when expressed as a confusion matrix. As you will see in the example, you can achieve this with a call to base::table(estimate, truth) which correctly puts the “correct” result in the columns of the confusion matrix.

Classification metrics are more complicated than numeric ones because you have to think about extensions to the multiclass case. For now, let’s start with the binary case.

Vector implementation

The vector implementation for classification metrics initially has a very similar setup as the numeric metrics. It used check_class_metric() instead of check_numeric_metric(). It has an additional argument, estimator that determines the type of estimator to use (binary or some kind of multiclass implementation or averaging). This argument is auto-selected for the user, so default it to NULL. Additionally, pass it along to check_class_metric() so that it can check the provided estimator against the classes of truth and estimate to see if they are allowed.

# Logic for `event_level`
event_col <- function(xtab, event_level) {
  if (identical(event_level, "first")) {
    colnames(xtab)[[1]]
  } else {
    colnames(xtab)[[2]]
  }
}

miss_rate_impl <- function(truth, estimate, event_level) {
  # Create 
  xtab <- table(estimate, truth)
  col <- event_col(xtab, event_level)
  col2 <- setdiff(colnames(xtab), col)

  tp <- xtab[col, col]
  fn <- xtab[col2, col]

  fn / (fn + tp)
}

miss_rate_vec <- function(truth,
                          estimate,
                          estimator = NULL,
                          na_rm = TRUE,
                          case_weights = NULL,
                          event_level = "first",
                          ...) {
  estimator <- finalize_estimator(truth, estimator)

  check_class_metric(truth, estimate, case_weights, estimator)
  
    if (na_rm) {
    result <- yardstick_remove_missing(truth, estimate, case_weights)

    truth <- result$truth
    estimate <- result$estimate
    case_weights <- result$case_weights
  } else if (yardstick_any_missing(truth, estimate, case_weights)) {
    return(NA_real_)
  }

  miss_rate_impl(truth, estimate, event_level)
}

Another change from the numeric metric is that a call to finalize_estimator() is made. This is the infrastructure that auto-selects the type of estimator to use.

data("two_class_example")
miss_rate_vec(two_class_example$truth, two_class_example$predicted)
#> [1] 0.120155

What happens if you try and pass in a multiclass result?

data("hpc_cv")
fold1 <- filter(hpc_cv, Resample == "Fold01")
miss_rate_vec(fold1$obs, fold1$pred)
#>          F          M          L 
#> 0.06214689 0.00000000 0.00000000

This isn’t great, as currently multiclass miss_rate() isn’t supported and it would have been better to throw an error if the estimator was not "binary". Currently, finalize_estimator() uses its default implementation which selected "macro" as the estimator since truth was a factor with more than 2 classes. When we implement multiclass averaging, this is what you want, but if your metric only works with a binary implementation (or has other specialized multiclass versions), you might want to guard against this.

To fix this, a generic counterpart to finalize_estimator(), called finalize_estimator_internal(), exists that helps you restrict the input types. If you provide a method to finalize_estimator_internal() where the method name is the same as your metric name, and then set the metric_class argument in finalize_estimator() to be the same thing, you can control how the auto-selection of the estimator is handled.

Don’t worry about the metric_dispatcher argument. This is handled for you and just exists as a dummy argument to dispatch off of.

It is also good practice to call validate_estimator() which handles the case where a user passed in the estimator themselves. This validates that the supplied estimator is one of the allowed types and error otherwise.

finalize_estimator_internal.miss_rate <- function(metric_dispatcher, x, estimator, call) {
  
  validate_estimator(estimator, estimator_override = "binary")
  if (!is.null(estimator)) {
    return(estimator)
  }
  
  lvls <- levels(x)
  if (length(lvls) > 2) {
    stop("A multiclass `truth` input was provided, but only `binary` is supported.")
  } 
  "binary"
}

miss_rate_vec <- function(truth,
                          estimate,
                          estimator = NULL,
                          na_rm = TRUE,
                          case_weights = NULL,
                          event_level = "first",
                          ...) {
  # calls finalize_estimator_internal() internally
  estimator <- finalize_estimator(truth, estimator, metric_class = "miss_rate")
  
  check_class_metric(truth, estimate, case_weights, estimator)
  
  if (na_rm) {
    result <- yardstick_remove_missing(truth, estimate, case_weights)

    truth <- result$truth
    estimate <- result$estimate
    case_weights <- result$case_weights
  } else if (yardstick_any_missing(truth, estimate, case_weights)) {
    return(NA_real_)
  }

  miss_rate_impl(truth, estimate, event_level)
}

# Error thrown by our custom handler
# miss_rate_vec(fold1$obs, fold1$pred)

# Error thrown by validate_estimator()
# miss_rate_vec(fold1$obs, fold1$pred, estimator = "macro")

Supporting multiclass miss rate

Like many other classification metrics such as precision() or recall(), miss rate does not have a natural multiclass extension, but one can be created using methods such as macro, weighted macro, and micro averaging. If you have not, I encourage you to read vignette("multiclass", "yardstick") for more information about how these methods work.

Generally, they require more effort to get right than the binary case, especially if you want to have a performant version. Luckily, a somewhat standard template is used in yardstick and can be used here as well.

Let’s first remove the “binary” restriction we created earlier.

rm(finalize_estimator_internal.miss_rate)

The main changes below are:

The binary implementation is moved to miss_rate_binary().
miss_rate_estimator_impl() is a helper function for switching between binary and multiclass implementations. It also applies the weighting required for multiclass estimators. It is called from miss_rate_impl() and also accepts the estimator argument using R’s function scoping rules.
miss_rate_multiclass() provides the implementation for the multiclass case. It calculates the true positive and false negative values as vectors with one value per class. For the macro case, it returns a vector of miss rate calculations, and for micro, it first sums the individual pieces and returns a single miss rate calculation. In the macro case, the vector is then weighted appropriately in miss_rate_estimator_impl() depending on whether or not it was macro or weighted macro.

miss_rate_vec <- function(truth, 
                          estimate, 
                          estimator = NULL, 
                          na_rm = TRUE, 
                          case_weights = NULL,
                          event_level = "first",
                          ...) {
  # calls finalize_estimator_internal() internally
  estimator <- finalize_estimator(truth, estimator, metric_class = "miss_rate")
  
  check_class_metric(truth, estimate, case_weights, estimator)
  
  if (na_rm) {
    result <- yardstick_remove_missing(truth, estimate, case_weights)

    truth <- result$truth
    estimate <- result$estimate
    case_weights <- result$case_weights
  } else if (yardstick_any_missing(truth, estimate, case_weights)) {
    return(NA_real_)
  }

  miss_rate_impl(truth, estimate, estimator, event_level)
}

miss_rate_impl <- function(truth, estimate, estimator, event_level) {
  xtab <- table(estimate, truth)
  # Rather than implement the actual method here, we rely on
  # an *_estimator_impl() function that can handle binary
  # and multiclass cases
  miss_rate_estimator_impl(xtab, estimator, event_level)
}

# This function switches between binary and multiclass implementations
miss_rate_estimator_impl <- function(data, estimator, event_level) {
  if(estimator == "binary") {
    miss_rate_binary(data, event_level)
  } else {
    # Encapsulates the macro, macro weighted, and micro cases
    wt <- get_weights(data, estimator)
    res <- miss_rate_multiclass(data, estimator)
    weighted.mean(res, wt)
  }
}


miss_rate_binary <- function(data, event_level) {
  col <- event_col(data, event_level)
  col2 <- setdiff(colnames(data), col)
  
  tp <- data[col, col]
  fn <- data[col2, col]
  
  fn / (fn + tp)
}

miss_rate_multiclass <- function(data, estimator) {
  
  # We need tp and fn for all classes individually
  # we can get this by taking advantage of the fact
  # that tp + fn = colSums(data)
  tp <- diag(data)
  tpfn <- colSums(data)
  fn <- tpfn - tp
  
  # If using a micro estimator, we sum the individual
  # pieces before performing the miss rate calculation
  if (estimator == "micro") {
    tp <- sum(tp)
    fn <- sum(fn)
  }
  
  # return the vector 
  tp / (tp + fn)
}

For the macro case, this separation of weighting from the core implementation might seem strange, but there is good reason for it. Some metrics are combinations of other metrics, and it is nice to be able to reuse code when calculating more complex metrics. For example, f_meas() is a combination of recall() and precision(). When calculating a macro averaged f_meas(), the weighting must be applied 1 time, at the very end of the calculation. recall_multiclass() and precision_multiclass() are defined similarly to how miss_rate_multiclass() is defined and returns the unweighted vector of calculations. This means we can directly use this in f_meas(), and then weight everything once at the end of that calculation.

Let’s try it out now:

# two class
miss_rate_vec(two_class_example$truth, two_class_example$predicted)
#> [1] 0.120155

# multiclass
miss_rate_vec(fold1$obs, fold1$pred)
#> [1] 0.5483506

Data frame implementation

Luckily, the data frame implementation is as simple as the numeric case, we just need to add an extra estimator argument and pass that through.

miss_rate <- function(data, ...) {
  UseMethod("miss_rate")
}

miss_rate <- new_class_metric(miss_rate, direction = "minimize")

miss_rate.data.frame <- function(data, 
                                 truth, 
                                 estimate, 
                                 estimator = NULL, 
                                 na_rm = TRUE, 
                                 case_weights = NULL,
                                 event_level = "first",
                                 ...) {
  class_metric_summarizer(
    name = "miss_rate",
    fn = miss_rate_vec,
    data = data,
    truth = !!enquo(truth),
    estimate = !!enquo(estimate), 
    estimator = estimator,
    na_rm = na_rm,
    case_weights = !!enquo(case_weights),
    event_level = event_level
  )
}

# Macro weighted automatically selected
fold1 %>%
  miss_rate(obs, pred)

# Switch to micro
fold1 %>%
  miss_rate(obs, pred, estimator = "micro")

# Macro weighted by resample
hpc_cv %>%
  group_by(Resample) %>%
  miss_rate(obs, pred, estimator = "macro_weighted")

# Error handling
miss_rate(hpc_cv, obs, VF)

Using custom metrics

The metric_set() function validates that all metric functions are of the same metric type by checking the class of the function. If any metrics are not of the right class, metric_set() fails. By using new_numeric_metric() and new_class_metric() in the above custom metrics, they work out of the box without any additional adjustments.

numeric_mets <- metric_set(mse, rmse)

numeric_mets(solubility_test, solubility, prediction)
#> # A tibble: 2 × 3
#>   .metric .estimator .estimate
#>   <chr>   <chr>          <dbl>
#> 1 mse     standard       0.521
#> 2 rmse    standard       0.722

Session information

#> ─ Session info ─────────────────────────────────────────────────────
#>  setting  value
#>  version  R version 4.4.0 (2024-04-24)
#>  os       macOS Sonoma 14.5
#>  system   aarch64, darwin20
#>  ui       X11
#>  language (EN)
#>  collate  en_US.UTF-8
#>  ctype    en_US.UTF-8
#>  tz       America/Los_Angeles
#>  date     2024-09-03
#>  pandoc   2.17.1.1 @ /opt/homebrew/bin/ (via rmarkdown)
#> 
#> ─ Packages ─────────────────────────────────────────────────────────
#>  package    * version    date (UTC) lib source
#>  broom      * 1.0.6      2024-05-17 [1] CRAN (R 4.4.0)
#>  dials      * 1.3.0      2024-07-30 [1] CRAN (R 4.4.0)
#>  dplyr      * 1.1.4      2023-11-17 [1] CRAN (R 4.4.0)
#>  ggplot2    * 3.5.1      2024-04-23 [1] CRAN (R 4.4.0)
#>  infer      * 1.0.7      2024-03-25 [1] CRAN (R 4.4.0)
#>  parsnip    * 1.2.1.9002 2024-08-30 [1] local
#>  purrr      * 1.0.2      2023-08-10 [1] CRAN (R 4.4.0)
#>  recipes    * 1.1.0.9000 2024-07-19 [1] local
#>  rlang      * 1.1.4      2024-06-04 [1] CRAN (R 4.4.0)
#>  rsample    * 1.2.1      2024-03-25 [1] CRAN (R 4.4.0)
#>  tibble     * 3.2.1      2023-03-20 [1] CRAN (R 4.4.0)
#>  tidymodels * 1.2.0      2024-03-25 [1] CRAN (R 4.4.0)
#>  tune       * 1.2.1      2024-04-18 [1] CRAN (R 4.4.0)
#>  workflows  * 1.1.4      2024-02-19 [1] CRAN (R 4.4.0)
#>  yardstick  * 1.3.1      2024-03-21 [1] CRAN (R 4.4.0)
#> 
#>  [1] /Library/Frameworks/R.framework/Versions/4.4-arm64/Resources/library
#> 
#> ────────────────────────────────────────────────────────────────────