trainSimilarityBasedReasoning2: Train Similarity Based Probability Model with anonymized...
In malsch/occupationCoding: Supervised Learning for Occupation Coding

View source: R/trainSimilarityBasedReasoning2.R

trainSimilarityBasedReasoning2

R Documentation

Train Similarity Based Probability Model with anonymized training data

Description

The output of this function is the same as from trainSimilarityBasedReasoning, but as input trainSimilarityBasedReasoning2 only requires aggregated (and thus anonymized) training data. We provide such training data for coding of German occupations into the German classification of Occupations (KldB 2010) as part of this package (see surveyCountsSubstringSimilarity and surveyCountsWordwiseSimilarity). Parameter settings for this function should be the same as those used to anonymize the training data. The examples below detail recommended application.

Usage

trainSimilarityBasedReasoning2(
  anonymized_data,
  num.allowed.codes = 1291,
  coding_index_w_codes,
  coding_index_without_codes = NULL,
  preprocessing = list(stopwords = NULL, stemming = NULL, strPreprocessing = TRUE,
    removePunct = FALSE),
  dist.type = c("wordwise", "substring", "fulltext"),
  dist.control = list(method = "osa", weight = c(d = 1, i = 1, s = 1, t = 1)),
  threshold = c(max = 3, use = 1),
  simulation.control = list(n.draws = 250, check.normality = FALSE),
  tmp_folder = NULL
)

Arguments

`anonymized_data`	`surveyCountsSubstringSimilarity` or `surveyCountsWordwiseSimilarity`
`num.allowed.codes`	the number of allowed codes in the target classification. There are 1286 categories in the KldB 2010 plus 5 special codes in both anonymized training data sets, so the default value is 1291.
`coding_index_w_codes`	a data.table with columns bezMale a character vector, contains masculine job titles from the coding index. bezFemale a character vector, contains feminine job titles from the coding index. Code a character vector with associated classification codes.
`coding_index_without_codes`	(not used, but automatically determined) Any words from `anonymized_data$dictString` that are not found within `coding_index_w_codes` belong into this character vector.
`preprocessing`	a list with elements stopwords a character vector, use `tm::stopwords("de")` for German stopwords. Only used if `dist.type = "wordwise"`. stemming `NULL` for no stemming and `"de"` for stemming using the German porter stemmer. Do not use unless the job titles in `coding_index_w_codes` were stemmed. strPreprocessing `TRUE` if `stringPreprocessing` shall be used. removePunct `TRUE` if `removePunctuation` shall be used.
`dist.type`	How to calculate similarity between entries from both coding_indices and verbal answers from the survey? Three options are currently supported. Since we use the `stringdist`-function excessively, one could easily extend the functionality of this procedure to other distance metrics. dist.type = "fulltext" Uses the `stringdist`-function directly after preprocessing to calculate distances. (the simplest approach but least useful.) dist.type = "substring" An entry from the coding index and a verbal answer are similar if the entry from the coding index is a substring of the verbal answer. dist.type = "wordwise" After preprocessing, split the verbal answer into words. Then calculate for each word separately the the similarity with entries from the coding index, using `stringdist`. Not the complete verbal answer but only the words (0 or more) that have highest similarity are then used to determine similarity with entries from the coding index.
`dist.control`	If `dist.type = "fulltext"` or `dist.type = "wordwise"` the entries from this list will be passed to `stringdist`. Currently only two possible entries are supported (method = "osa", weight = c(d = 1, i = 1, s = 1, t = 1) is recommended), but one could easily extend the functionality.
`threshold`	A numeric vector with two elements. If `dist.type = "fulltext"` or `dist.type = "wordwise"`, the threshold determines up to which distance a verbal answer and an entry from the coding index are similar. The second number actually gets used. The first number is only used to speed up similarity calculations. It should be identical or larger than the second number.
`simulation.control`	a list with two components, n.draws Number of draws from the posterior distribution to determine posterior predictive probabilities. The larger, the more precise the results will be. check.normality We would like that the hyperprior distribution is normal. Set check.normality to TRUE to do some diagnostics about this.
`tmp_folder`	(not used)

Value

a list with components

prediction.datasets$modelProb: Contains all entries from the coding index. dist = "official" if the entry stems from coding_index_w_codes and dist = selfcreated if the entry stems from coding_index_without_codes. string.prob is used for weighting purposes (model averaging) if a new verbal answer is similar to multiple strings. unobserved.mean.theta gives a probability (usually very low) for any category that was not observed in the training data together with this string.
prediction.datasets$categoryProb: mean.theta is the probability for code given that an incoming verbal answer is similar to string. Only available if this code was at least a single time observed with this string (Use unobserved.mean.theta otherwise).
num.allowed.codes: Number of categories in the classification.
preprocessing: The input parameter stored to replicate preprocessing with incoming data.
dist.type: The input parameter stored to replicate distance calculations with incoming data.
dist.control: The input parameter stored to replicate distance calculations with incoming data.
threshold: The input parameter stored to replicate distance calculations with incoming data.
simulation.control: The input parameters controlling the Monte Carlo simulation.

Examples

# set up test data
data(occupations)
allowed.codes <- c("71402", "71403", "63302", "83112", "83124", "83131", "83132", "83193", "83194", "-0004", "-0030")
allowed.codes.titles <- c("Office clerks and secretaries (without specialisation)-skilled tasks", "Office clerks and secretaries (without specialisation)-complex tasks", "Gastronomy occupations (without specialisation)-skilled tasks",
 "Occupations in child care and child-rearing-skilled tasks", "Occupations in social work and social pedagogics-highly complex tasks", "Pedagogic specialists in social care work and special needs education-unskilled/semiskilled tasks", "Pedagogic specialists in social care work and special needs education-skilled tasks", "Supervisors in education and social work, and of pedagogic specialists in social care work", "Managers in education and social work, and of pedagogic specialists in social care work",
 "Not precise enough for coding", "Student assistants")
proc.occupations <- removeFaultyAndUncodableAnswers_And_PrepareForAnalysis(occupations, colNames = c("orig_answer", "orig_code"), allowed.codes, allowed.codes.titles)

# set up dictionary (see help file for how to obtain the dictionary)
path_to_file <- "./Gesamtberufsliste_der_BA.xlsx" # change path
try({coding_index_w_codes <- prepare_German_coding_index_Gesamtberufsliste_der_BA(path_to_file, count.categories = FALSE)}, silent = TRUE)
# or, if the file does not exist at the given path, just use coding_index_excerpt
if (!exists("coding_index_w_codes")) coding_index_w_codes <- coding_index_excerpt

data(surveyCountsSubstringSimilarity)
simBasedModelSubstring <- trainSimilarityBasedReasoning2(anonymized_data = surveyCountsSubstringSimilarity,
                                                         num.allowed.codes = 1291,
                                                         coding_index_w_codes = coding_index_w_codes,
                                                         preprocessing = list(stopwords = NULL, stemming = NULL, strPreprocessing = TRUE, removePunct = FALSE),
                                                         dist.type = "substring",
                                                         dist.control = NA,
                                                         threshold = NA,
                                                         simulation.control = list(n.draws = 250, check.normality = FALSE)
                                                         )

res <- predictSimilarityBasedReasoning(simBasedModelSubstring, proc.occupations)

# Look at most probable answer from each id
res[, .SD[which.max(pred.prob), list(ans, true.code = code, pred.code, acc = code == pred.code)], by = id]
res[, .SD[which.max(pred.prob), list(ans, true.code = code, pred.code, acc = code == pred.code)], by = id][, mean(acc)] # calculate aggrement rate

# Look at a single person and order predictions by their probability. According to the algorithm the code 81112 has the highest probability, but the code 71402 (which was selected by a coder) has second-highest probability
res[id == 11][order(pred.prob, decreasing = TRUE)]

data(surveyCountsWordwiseSimilarity)
simBasedModelWordwise <- trainSimilarityBasedReasoning2(anonymized_data = surveyCountsWordwiseSimilarity,
                                                         num.allowed.codes = 1291,
                                                         coding_index_w_codes = coding_index_w_codes,
                                                         preprocessing = list(stopwords = NULL, stemming = NULL, strPreprocessing = TRUE, removePunct = FALSE),
                                                         dist.type = "wordwise",
                                                         dist.control = list(method = "osa", weight = c(d = 1, i = 1, s = 1, t = 1)),
                                                         threshold = c(max = NA, use = 1),
                                                         simulation.control = list(n.draws = 250, check.normality = FALSE)
)

res <- predictSimilarityBasedReasoning(simBasedModelWordwise, proc.occupations)

# Look at most probable answer from each id
res[, .SD[which.max(pred.prob), list(ans, true.code = code, pred.code, acc = code == pred.code)], by = id]
res[, .SD[which.max(pred.prob), list(ans, true.code = code, pred.code, acc = code == pred.code)], by = id][, mean(acc)] # calculate aggrement rate

# Look at a single person and order predictions by their probability. Other than previously, this algorithm predicts 71402, the correct code.
res[id == 11][order(pred.prob, decreasing = TRUE)]

malsch/occupationCoding documentation built on March 14, 2024, 8:09 a.m.