summary.nbc: Summarize a NBC model with metrics

In rrwen/nbc4va: Bayes Classifier for Verbal Autopsy Data

Summarize a NBC model with metrics

Description

Summarizes the results from a nbc object. The summary can be either for a particular case or for the entirety of cases.

Usage

## S3 method for class 'nbc'
summary(object, top = 5, id = NULL, csmfa.obs = NULL, ...)

Arguments

object	The result nbc object.
top	A number that produces top causes depending on id: If (id is char): provide the top causes of the case by probability If (id is NULL): provide the top causes by predicted Cause Specific Mortality Fractions (CSMF)
id	A character representing a case id in the test data.
csmfa.obs	A character vector of the true causes for calculating the CSMF accuracy.
...	Additional arguments to be passed if applicable

Details

See Methods documentation for details on calculations and metrics.

Value

out A summary object built from a nbc object with modifications/additions:

• If (id is char):

  • Additions to a nbc object:

    • $id (char): the case id chosen by the user

    • $top (numeric): the input number of top causes for id

    • $top.prob (vectorof double): the top probabilities for id

  • The following are modified from a nbc object to be id specific:
    $test, $test.ids, $test.causes, $obs.causes, $prob, $prob.causes, $pred, $pred.causes

• If (id is NULL):

  • Additions to the nbc object:

    • * indicates that the item is only available if test causes are known

    • ** indicates that the item ignores * if csmfa.obs is given

    • $top.csmf.pred (vectorof double): the top predicted CSMFs by cause

    • $top.csmf.obs* (vectorof double): the top observed CSMFs by cause

    • $metrics.all** (vectorof double): a numeric vector of overall metrics.

      • Names: TruePositives, TrueNegatives, FalsePositives, FalseNegatives, Accuracy, Sensitivity, PCCC, CSMFMaxError, CSMFaccuracy

      • TruePositives* (double): total number of true positives

      • TrueNegatives* (double): total number of true negatives

      • FalsePositives* (double): total number of false positives

      • FalseNegatives* (double): total number of false negatives

      • Sensitivity* (double): the overall sensitivity

      • PCCC* (double): the partial chance corrected concordance

      • CSMFMaxError** (double): the maximum Cause Specific Mortality Fraction Error

      • CSMFaccuracy** (double): the Cause Specific Mortaliy Fraction accuracy

    • $metrics.causes (dataframe): a perfomance table of metrics by cause.

      • Columns: Cause, Sensitivity, CSMFpredicted, CSMFobserved

      • Cause (vectorof char): The unique causes from both the obs and pred inputs

      • Sensitivity* (vectorof double): the sensitivity for a cause

      • CSMFpredicted (vectorof double): the cause specific mortality fraction for a cause given the predicted deaths

      • CSMFobserved* (vectorof double): the cause specific mortality fraction for a cause given the observed deaths

      • TruePositives (vectorof double): The total number of true positives per cause

      • TrueNegatives (vectorof double): The total number of true negatives per cause

      • FalsePositives (vectorof double): The total number of false positives per cause

      • FalseNegatives (vectorof double): The total number of false negatives per cause

      • PredictedFrequency (vectorof double): The occurence of a cause in the pred input

      • ObservedFrequency (vectorof double): The occurence of a cause in the obs input

      • Example:

        Cause	Sensitivity	Metric-n..
        HIV	0.5	#..
        Stroke	0.5	#..

See Also

Other main functions: nbc(), plot.nbc(), print.nbc_summary()

Examples

library(nbc4va)
data(nbc4vaData)

# Run naive bayes classifier on random train and test data
train <- nbc4vaData[1:50, ]
test <- nbc4vaData[51:100, ]
results <- nbc(train, test)

# Obtain a summary for the results
brief <- summary(results, top=2)  # top 2 causes by CSMF for all test data
briefID <- summary(results, id="v48")  # top 5 causes by probability for case "v48"

rrwen/nbc4va documentation built on May 11, 2022, 9:45 p.m.