tidymodels/yardstick

What it solves

It provides a standardized way to estimate model performance using tidy data principles, allowing users to calculate common performance characteristics for classification and regression models without needing to manually manage complex data structures.

How it works

yardstick uses a dplyr-like syntax to compute metrics. It takes a data frame containing truth values and predicted values and returns the results as a data frame. It supports binary classification, multiclass classification (with various estimators like macro and micro averaging), and the ability to calculate metrics across multiple resamples using grouped data frames.

Who it’s for

Data scientists and machine learning practitioners using R who want a consistent, tidy interface for evaluating their models.

Highlights

• Tidy Interface: Returns metrics as data frames, making them easy to integrate into analysis pipelines.
• Multiclass Support: Offers extensive extensions for multiclass metrics with multiple calculation methods.
• Resample Evaluation: Simplifies calculating performance across multiple model resamples simultaneously.
• Integrated Visualization: Includes autoplot() methods for curve-based metrics like ROC, PR, and gain curves via ggplot2.