repository·main·Indexed 19 days ago
https://github.com/tidymodels/yardstickAn R package for estimating model performance using tidy data principles. It provides metrics for binary and multiclass classification, regression, and other evaluation tasks, including support for macro and micro averaging, resample-based calculations, and visualization of performance curves via autoplot().
Curve-based methods such as roc_curve(), pr_curve(), and gain_curve() support ggplot2::autoplot(). This allows for easy visualization of performance curves, which can be faceted by resample or class when using grouped data.
library(ggplot2)
hpc_cv |>
group_by(Resample) |>
roc_curve(obs, VF:L) |>
autoplot()You can install the stable version of yardstick from CRAN or install the development version using pak.
install.packages("yardstick")
# Development version:
# install.packages("pak")
pak::pak("tidymodels/yardstick")If your data is grouped by resamples (e.g., from cross-validation), you can use dplyr::group_by() in conjunction with yardstick metric functions to calculate metrics for each resample simultaneously.
hpc_cv |>
group_by(Resample) |>
roc_auc(obs, VF:L)For binary classification, you can use metrics() to compute multiple performance characteristics at once, or call specific metric functions like roc_auc() directly. The results are returned as tibbles following tidy data principles.
library(yardstick)
library(dplyr)
# Compute multiple metrics
metrics(two_class_example, truth, predicted)
# Compute a specific metric (e.g., ROC AUC)
two_class_example |>
roc_auc(truth, Class1)All classification metrics in yardstick have multiclass extensions. You can control the aggregation method using the estimator argument. Common options include:
macro: Macro-averaged metrics.micro: Micro-averaged metrics.# Macro averaged multiclass precision
precision(hpc_cv, obs, pred)
# Micro averaged multiclass precision
precision(hpc_cv, obs, pred, estimator = "micro")