Overview
The goal of PRIA is to find an ENA model with a reduced set of codes, that is highly correlated to the original model.
How To
Step 1: Generate a regular ENA Model
The first step is to generate an ENA model, one that may have more codes that is preferred.
data <- read.csv(system.file(package = "PRIA", "extdata/TADMUSCoded_6_28_19.csv"))
### Make the full set.
units <- data[, c("COND", "Scenario", "Speaker_2", "Team")]
conversation <- data[, c("Scenario", "Team")]
code_names <- c("SeekingInformation","DetectIdentify","TrackBehavior","StatusUpdate","AssessmentPrioritization","DefensiveOrders","DeterrentOrders","Recommendation")
codes <- data[, code_names]
metadata <- data[, c("COND", "ATOMTOT_GROUP")]
full.accum <- rENA::ena.accumulate.data(
units = units,
conversation = conversation,
codes = codes,
metadata = metadata,
window.size.back = 4
);
full.set <- rENA::ena.make.set(full.accum);Step 2: Create Reduced Sets
Given a set with some number of codes, we can then use PRIA. There is a simple function, pria() that takes two parameters
- the existing ENA model,
- the maximum number of codes to remove
The result will be a data.frame containing the results of comparing the models with a reduced set of codes to that of the original ENA model.
reduced_sets <- pria(full.set, 1);Step 3: View Results
The data.frame returned contains one row for each model that was generated and compared to the full, base model.
removed gof gof.lower gof.upper points points.lower points.upper nodes nodes.lower nodes.upper
1 0.9477 0.9471 0.9484 0.99535385 0.9944 0.9962 0.9927228 0.9495 0.9990
1 0.9459 0.9452 0.9466 0.95434606 0.9450 0.9622 0.9753454 0.8372 0.9965
1 0.9427 0.9419 0.9434 0.98664856 0.9839 0.9890 0.9981376 0.9869 0.9997
1 0.9410 0.9402 0.9417 0.95932074 0.9509 0.9663 0.9805566 0.8697 0.9972
1 0.9403 0.9395 0.9410 0.96818602 0.9616 0.9737 0.9786727 0.8578 0.9970
1 0.9143 0.9132 0.9153 0.85987025 0.8328 0.8829 0.9892118 0.9259 0.9985
1 0.9063 0.9052 0.9075 0.06236196 -0.0333 0.1569 -0.1095204 -0.7969 0.7014
1 0.8889 0.8876 0.8903 -0.28808287 -0.3733 -0.1981 -0.2328226 -0.8388 0.6308Step 4: Extract Results
Find Results Above Threshold
PRIA implements a special version of the greater than (>) operator, allowing for quick extraction of results that are above a defined threshold.
Note: This comparison checks all three of the lower statistics against the threshold
## When checking the previous table/set of results against 0.9, produces the following subset of the full results
# reduced_sets > 0.9
removed gof gof.lower gof.upper points points.lower points.upper nodes nodes.lower nodes.upper
1 0.9477 0.9471 0.9484 0.99535385 0.9944 0.9962 0.9927228 0.9495 0.9990
1 0.9427 0.9419 0.9434 0.98664856 0.9839 0.9890 0.9981376 0.9869 0.9997Find Top Result
PRIA also implements a version of the max() function which will, using the same criteria described for the greater than operator, return the top result, when ordered by number of removed codes and overall goodness of fit.
# max(reduced_sets)
removed gof gof.lower gof.upper points points.lower points.upper nodes nodes.lower nodes.upper
1 0.9477 0.9471 0.9484 0.99535385 0.9944 0.9962 0.9927228 0.9495 0.9990Step 5: Use the ENA sets
By default, PRIA will maintain the generated ENA sets along with the results. Although not directly present when printing or viewing the results, the sets can be accessed using $ena.set.
For example, to view the codes used to generate the individual results: reduced_sets$ena.set$rotation$codes
Or to use the ENA model from top result: top <- max(reduced_sets)$ena.set[[1]]