fastTextR is an R interface to the fastText library. It can be used to word representation learning (Bojanowski et al., 2016) and supervised text classification (Joulin et al., 2016). Particularly the advantage of fastText to other software is that, it was designed for biggish data.

The following example is based on the examples provided in the fastText library, the example shows how to use fastTextR text classification.

Download Data

options(width=100L)
fn <- "dbpedia_csv.tar.gz"

if ( !file.exists(fn) ) {
    download.file("https://github.com/le-scientifique/torchDatasets/raw/master/dbpedia_csv.tar.gz",
                  fn)
    untar(fn)
}

Normalize Data

In fastText labels are typically marked with __label__1 to __label__k. Since fastText relies at the order of the trainings data it is important to ensure the order of the trainings data follows no particular pattern (which is done here with sample). The function normalize mimics the data preparation steps of the bash function normalize_text as shown in classification-example.sh.

library("fastTextR")

train <- sample(sprintf("__label__%s", readLines("dbpedia_csv/train.csv")))
head(train, 2)
## [1] "__label__5,\"Helmut Haussmann\",\" Helmut Haussmann (born 18 May 1943) is a German academic and politician. He served as minister of economy from 1988 to 1991.\""                                                                                                                                                                              
## [2] "__label__9,\"Studzianki Rawa County\",\" Studzianki [stuˈd͡ʑaŋki] is a village in the administrative district of Gmina Sadkowice within Rawa County Łódź Voivodeship in central Poland. It lies approximately 5 kilometres (3 mi) west of Sadkowice 16 km (10 mi) east of Rawa Mazowiecka and 69 km (43 mi) east of the regional capital Łódź.\""
train <- ft_normalize(train)
writeLines(train, con = "dbpedia.train")

test <- readLines("dbpedia_csv/test.csv")
labels <- trimws(gsub(",.*", "", test))
table(labels)
## labels
##    1   10   11   12   13   14    2    3    4    5    6    7    8    9 
## 5000 5000 5000 5000 5000 5000 5000 5000 5000 5000 5000 5000 5000 5000
test <- ft_normalize(test)
test <- trimws(sub(".*?,", "", test))
head(test, 2)
## [1] "\" TY KU \" , \" TY KU /taɪkuː/ is an American alcoholic beverage company that specializes in sake and other spirits . The privately-held company was founded in 2004 and is headquartered in New York City New York . While based in New York TY KU ' s beverages are made in Japan through a joint venture with two sake breweries . Since 2011 TY KU ' s growth has extended its products into all 50 states . \""                                        
## [2] "\" Odd Lot Entertainment \" , \" OddLot Entertainment founded in 2001 by longtime producers Gigi Pritzker and Deborah Del Prete ( The Wedding Planner ) is a film production and financing company based in Culver City California . OddLot produced the film version of Orson Scott Card ' s sci-fi novel Ender ' s Game . A film version of this novel had been in the works in one form or another for more than a decade by the time of its release . \""

Train Model

After the data preparation the model can be trained and is saved to the file "dbpedia.bin".

cntrl <- ft_control(word_vec_size = 10L, learning_rate = 0.1, max_len_ngram = 2L, 
                    min_count = 1L, nbuckets = 10000000L, epoch = 5L, nthreads = 4L)

model <- ft_train(file = "dbpedia.train", method = "supervised", control = cntrl)
ft_save(model, "dbpedia.bin")

Read Model

A previously trained model can be loaded via the function read.fasttext.

model <- ft_load("dbpedia.bin")

Predict / Test Model

To perform prediction the function predict can be used.

test_pred <- ft_predict(model, newdata=test, k = 1L)
str(test_pred)
## 'data.frame':    70000 obs. of  3 variables:
##  $ id   : int  1 2 3 4 5 6 7 8 9 10 ...
##  $ label: chr  "__label__1" "__label__1" "__label__1" "__label__1" ...
##  $ prob : num  1 0.72 0.999 0.998 0.993 ...
confusion_matrix <- table(truth=as.integer(labels), 
                          predicted=as.integer(gsub("\\D", "", test_pred$label)))
print(confusion_matrix)
##      predicted
## truth    1    2    3    4    5    6    7    8    9   10   11   12   13   14
##    1  4734   45   13    6   12   50   60    1    2    4    6   12    7   48
##    2    41 4912    1    1    2    0   32    3    1    0    1    0    0    6
##    3    16    2 4817   15   74    0    5    1    0    0    0   23   11   36
##    4     2    1   29 4947   15    3    0    0    0    2    0    0    1    0
##    5     7    5   70   11 4896    3    3    0    1    1    0    0    0    3
##    6    34    1    1    1    3 4936   12    5    0    0    0    2    3    2
##    7    59   31    1    1    6   17 4839   26    8    0    0    1    1   10
##    8     3    1    0    0    1    2   28 4944   16    2    2    0    0    1
##    9     1    1    0    0    2    0   10   17 4967    0    1    1    0    0
##    10    3    0    0    1    0    0    0    5    0 4952   37    1    0    1
##    11   17    1    0    0    0    1    0    2    0   32 4945    1    0    1
##    12    7    0   18    1    0    4    0    0    0    0    0 4937   21   12
##    13    7    1    8    0    0    2    3    1    0    0    0   18 4926   34
##    14   44    7   25    1    2    5    7    3    1    2    1    5   34 4863
accuracy <- sum(diag(confusion_matrix)) / sum(confusion_matrix)
print(sprintf("Accuracy: %0.4f", accuracy))
## [1] "Accuracy: 0.9802"

References

[1] P. Bojanowski, E. Grave, A. Joulin, T. Mikolov, Enriching Word Vectors with Subword Information

@article{bojanowski2016enriching,
  title={Enriching Word Vectors with Subword Information},
  author={Bojanowski, Piotr and Grave, Edouard and Joulin, Armand and Mikolov, Tomas},
  journal={arXiv preprint arXiv:1607.04606},
  year={2016}
}

[2] A. Joulin, E. Grave, P. Bojanowski, T. Mikolov, Bag of Tricks for Efficient Text Classification

@article{joulin2016bag,
  title={Bag of Tricks for Efficient Text Classification},
  author={Joulin, Armand and Grave, Edouard and Bojanowski, Piotr and Mikolov, Tomas},
  journal={arXiv preprint arXiv:1607.01759},
  year={2016}
}