build_vectors: Build fasttext vectors

In fastrtext: 'fastText' Wrapper for Text Classification and Word Representation

Description

Trains a fasttext vector/unsupervised model following method described in Enriching Word Vectors with Subword Information using the fasttext implementation.

See FastText word representation tutorial for more information on training unsupervised models using fasttext.

Usage

build_vectors(documents, model_path, modeltype = c("skipgram", "cbow"),
  bucket = 2e+06, dim = 100, epoch = 5, label = "__label__",
  loss = c("ns", "hs", "softmax", "ova", "one-vs-all"), lr = 0.05,
  lrUpdateRate = 100, maxn = 6, minCount = 5, minn = 3, neg = 5,
  t = 1e-04, thread = 12, verbose = 2, wordNgrams = 1, ws = 5)

Arguments

documents	character vector of documents used for training
model_path	Name of output file without file extension.
modeltype	Should training be done using skipgram or cbow? Defaults to skipgram.
bucket	number of buckets
dim	size of word vectors
epoch	number of epochs
label	text string, labels prefix. Default is "label"
loss	loss function ns, hs, softmax
lr	learning rate
lrUpdateRate	change the rate of updates for the learning rate
maxn	max length of char ngram
minCount	minimal number of word occurences
minn	min length of char ngram
neg	number of negatives sampled
t	sampling threshold
thread	number of threads
verbose	verbosity level
wordNgrams	max length of word ngram
ws	size of the context window

Value

path to model file, as character

Examples

## Not run: 
library(fastrtext)
text <- train_sentences
model_file <- build_vectors(text[['text']], 'my_model')
model <- load_model(model_file)

## End(Not run)

fastrtext documentation built on Oct. 30, 2019, 11:32 a.m.