Skip to contents

Function used to calculate some distance metrics between the predicted survival and the observed survival. fit_metrics() is used to calculate distance metrics between empirical and fitted survival for a predictions object, preferably the object $preds returned from plot_fit_on_data().

fit_metrics.Rd

Function used to calculate some distance metrics between the predicted survival and the observed survival. fit_metrics() is used to calculate distance metrics between empirical and fitted survival for a predictions object, preferably the object $preds returned from plot_fit_on_data().

Usage

fit_metrics(preds, nobs = NULL, threshold = 0.005)

Arguments

preds

The $preds object from plot_fit_on_data() applied to the model. If not this one, should be a prediction tibble with the columns time, strata (if applicable), estimate, .pred_survival, n.risk (also chain, if necessary). It's important that the quantities estimate and .pred_survival are calculated for the same time and strata. It's highly recommended to simply use the object $preds returned from the function plot_fit_on_data().

nobs

The number of observations used to fit the model. Can be ignored if threshold is set to 0. To easily calculate this value, use the function nobs() applied to the model object.

threshold

Numeric value between 0 and 1. Times with n.risk below threshold * nobs will be ignored. Default is 0.005 (0.5%). Important because the distance metrics may be too big if calculated in intervals without sufficient observations to be estimated.

Value

A tibble with the following columns:

  • strata: The stratas used to fit the model (if necessary).

  • n_strata: The number of observations in the strata.

  • chain: The chain of the Bayesian model (only if necessary).

  • metric: Which metric is being calculated.

  • value: Value for the metric.

For now, the following metrics are available and will be included:

  • MSE: Mean Squared Error (the less the better).

  • MAE: Mean Absolute Error (the less the better).

  • Hellinger Distance: Hellinger distance, sometimes called Jeffreys distance (the less the better).

  • KS Distance: Kolmogorov-Smirnov distance (the less the better).