Skip to contents

Extract diagnostic metrics from model fit and data

extract_diagnostics.Rd

This function extracts various diagnostic metrics from a fitted EpiNow2 model and provided data. It checks for low case counts and computes diagnostics from the fitted model, including the mean acceptance statistic, divergent transitions, maximum tree depth, and Rhat values. These diagnostics are then flagged if they exceed specific thresholds, and the results are returned as a data frame.

Usage

extract_diagnostics(fit, data, job_id, task_id, disease, geo_value, model)

Arguments

fit

The model fit object from EpiNow2

data

A data frame containing the input data used in the model fit.

job_id

A unique identifier for the job

task_id

A unique identifier for the task

disease, geo_value, model

Metadata for downstream processing.

Value

A data.frame containing the extracted diagnostic metrics. The data frame includes the following columns:

  • diagnostic: The name of the diagnostic metric.

  • value: The value of the diagnostic metric.

  • job_id: The unique identifier for the job.

  • task_id: The unique identifier for the task.

  • disease,geo_value,model: Metadata for downstream processing.

Details

The following diagnostics are calculated:

  • mean_accept_stat: The average acceptance statistic across all chains.

  • p_divergent: The proportion of divergent transitions across all samples.

  • n_divergent: The number of divergent transitions across all samples.

  • p_max_treedepth: The proportion of samples that hit the maximum tree depth.

  • p_high_rhat: The proportion of parameters with Rhat values greater than 1.05, indicating potential convergence issues.

  • n_high_rhat: The number of parameters with Rhat values greater than 1.05, indicating potential convergence issues.

  • low_case_count_flag: A flag indicating if there are low case counts in the data. See low_case_count_diagnostic() for more information on this diagnostic.

  • epinow2_diagnostic_flag: A combined flag that indicates if any diagnostic thresholds are exceeded. The diagnostic thresholds (1) mean_accept_stat < 0.1, (2) p_divergent > 0.0075, (3) p_max_treedepth > 0.05, and (4) p_high_rhat > 0.0075.