Evaluate the accuracy of your UDPipe model on holdout data
Get precision, recall and F1 measures on finding words / sentences / upos / xpos / features annotation as well as UAS and LAS dependency scores on holdout data in conllu format.
udpipe_accuracy(
object,
file_conllu,
tokenizer = c("default", "none"),
tagger = c("default", "none"),
parser = c("default", "none")
)object |
an object of class |
file_conllu |
the full path to a file on disk containing holdout data in conllu format |
tokenizer |
a character string of length 1, which is either 'default' or 'none' |
tagger |
a character string of length 1, which is either 'default' or 'none' |
parser |
a character string of length 1, which is either 'default' or 'none' |
a list with 3 elements
accuracy: A character vector with accuracy metrics.
error: A character string with possible errors when calculating the accuracy metrics
model <- udpipe_download_model(language = "dutch-lassysmall")
if(!model$download_failed){
ud_dutch <- udpipe_load_model(model$file_model)
file_conllu <- system.file(package = "udpipe", "dummydata", "traindata.conllu")
metrics <- udpipe_accuracy(ud_dutch, file_conllu)
metrics$accuracy
metrics <- udpipe_accuracy(ud_dutch, file_conllu,
tokenizer = "none", tagger = "default", parser = "default")
metrics$accuracy
metrics <- udpipe_accuracy(ud_dutch, file_conllu,
tokenizer = "none", tagger = "none", parser = "default")
metrics$accuracy
metrics <- udpipe_accuracy(ud_dutch, file_conllu,
tokenizer = "default", tagger = "none", parser = "none")
metrics$accuracy
}
## cleanup for CRAN only - you probably want to keep your model if you have downloaded it
if(file.exists(model$file_model)) file.remove(model$file_model)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.