In ML4LHS/runway: Visualizing Prediction Model Performance

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

knitr::opts_chunk$set(warning = FALSE, message = FALSE)

runway

The goal of runway is to generate statistics and plots to calculate discrimination, calibration, and decision curves for prediction models.

Why is it called runway?

Because you can use it to visually compare models.

Sometimes your models look quite different.

Other times your models look the same...

Installation

You can install runway from GitHub with:

remotes::install_github('ML4LHS/runway')

Load the package

First, load the package.

library(runway)

Sample datasets

Runway comes with two sample datasets.

data(single_model_dataset)
head(single_model_dataset)

data(multi_model_dataset)
head(multi_model_dataset)

Evaluating a single model

Threshold-performance plot (single model)

threshperf_plot(single_model_dataset,
                outcome = 'outcomes',
                positive = '1',
                prediction = 'predictions')

Calibration plot with 10 bins (single model)

Note: 10 bins is the default.

cal_plot(single_model_dataset,
         outcome = 'outcomes',
         positive = '1',
         prediction = 'predictions')

Calibration plot with 5 bins (single model)

cal_plot(single_model_dataset,
         outcome = 'outcomes',
         positive = '1',
         prediction = 'predictions',
         n_bins = 5)

Calibration plot with loess curve (single model)

cal_plot(single_model_dataset,
         outcome = 'outcomes',
         positive = '1',
         prediction = 'predictions',
         n_bins = 0,
         show_loess = TRUE)

Comparing multiple models

Threshold-performance plot (multiple models)

threshperf_plot_multi(multi_model_dataset,
                      outcome = 'outcomes',
                      positive = '1',
                      prediction = 'predictions',
                      model = 'model_name')

Calibration plot with 10 bins (multiple models)

Note: 10 bins is the default.

cal_plot_multi(multi_model_dataset,
         outcome = 'outcomes',
         positive = '1',
         prediction = 'predictions',
         model = 'model_name')

Calibration plot with 5 bins (multiple models)

cal_plot_multi(multi_model_dataset,
         outcome = 'outcomes',
         positive = '1',
         prediction = 'predictions',
         model = 'model_name',
         n_bins = 5)

Calibration plot with loess curve (multiple models)

Unlike single calibration plots, the choice of binned calibration and loess calibration are mutually exclusive. To show less curves, you must set show_loess to TRUE and n_bins to 0.

cal_plot_multi(multi_model_dataset,
         outcome = 'outcomes',
         positive = '1',
         prediction = 'predictions',
         model = 'model_name',
         n_bins = 0,
         show_loess = TRUE)

ROC curve w/CI

roc_plot(single_model_dataset, 
         outcome = 'outcomes', 
         positive = '1',
         prediction = 'predictions',
         ci = TRUE, 
         plot_title = 'Single ROC curve w/CI ribbon')

Multiple ROC curves w/CI ribbons

roc_plot_multi(multi_model_dataset, 
         outcome = 'outcomes', 
         positive = '1',
         prediction = 'predictions', 
         model = 'model_name',
         ci = TRUE,
         plot_title = 'Multiple model ROC curves w/CI ribbons')

ML4LHS/runway documentation built on Sept. 9, 2023, 8:27 p.m.