let tokens = shibboleth::tokenize("Totally! I love cupcakes!");
assert_eq!(tokens[0], "total");
assert_eq!(tokens[3], "cupcak");
Shibboleth can use training corpora provided in an sqlite file matching this schema:
CREATE TABLE documents (id PRIMARY KEY, text);
A popular resource for training purposes is Wikipedia. The script below will download and unzip such a sqlite file with just over 5 million documents. For the wiki license see here.
$ wget -O wiki.db.gz https://dl.fbaipublicfiles.com/drqa/docs.db.gz && gunzip wiki.db.gz
This example takes the wiki.db file downloaded above, runs through the first 1,000,000 documents, stems them, and builds a vocabulary of the 25,000 most common words. The output will be saved to WikiVocab25k.txt
use shibboleth;
shibboleth::build_vocab_from_db("wiki.db", "WikiVocab25k.txt", 1000000, 25000);
``` use shibboleth;
// create a new encoder object with 200 elements per word vector from a vocabulary file let mut enc = shibboleth::Encoder::new(200, "WikiVocab25k.txt");
// the prediction (sigmoid) for 'chips' occuring near 'fish' should be near 0.5 prior to training let p = enc.predict("fish", "chips"); match p { Some(val) => println!("'Fish'->'Chips' sigmoid activation before training: {}", val), None => println!("One of these words is not in your vocabulary") }
// train on two examples enc.traindoc("I like to eat fish & chips."); enc.traindoc("Steve has chips with his fish.");
// after training, the prediction should be near unity
let p = enc.predict("fish", "chips");
match p {
Some(val) => println!("'Fish'->'Chips' sigmoid activation after training: {}", val),
None => println!("One of these words is not in your vocabulary")
}
Typical Output:
'Fish'->'Chips' sigmoid activation before training: 0.5002038
'Fish'->'Chips' sigmoid activation after training: 0.999495
```