Spaces:
Running
Running
pairCount
Browse files- src/bpe.js +53 -0
- test/bpe.js +19 -1
src/bpe.js
CHANGED
|
@@ -18,3 +18,56 @@ export function preprocess(corpus) {
|
|
| 18 |
export function initialVocabulary(corpus) {
|
| 19 |
return Array.from(new Set(corpus)).sort()
|
| 20 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
export function initialVocabulary(corpus) {
|
| 19 |
return Array.from(new Set(corpus)).sort()
|
| 20 |
}
|
| 21 |
+
|
| 22 |
+
/**
|
| 23 |
+
* def pair_count(corpus_l):
|
| 24 |
+
* pairs = {}
|
| 25 |
+
*
|
| 26 |
+
* for i in range(0, len(corpus_l) - 1):
|
| 27 |
+
* if corpus_l[i + 1].startswith('β'):
|
| 28 |
+
* continue
|
| 29 |
+
*
|
| 30 |
+
* bigram = (corpus_l[i], corpus_l[i + 1])
|
| 31 |
+
* if bigram in pairs:
|
| 32 |
+
* pairs[bigram] += 1
|
| 33 |
+
* else:
|
| 34 |
+
* pairs[bigram] = 1
|
| 35 |
+
*
|
| 36 |
+
* return dict(sorted(pairs.items(), reverse=True, key=lambda pair: pair[1]))
|
| 37 |
+
*
|
| 38 |
+
* @param {string} corpus_l
|
| 39 |
+
* @returns {Record<string,number>}
|
| 40 |
+
*/
|
| 41 |
+
export function pairCount(corpus_l) {
|
| 42 |
+
const pairs = {}
|
| 43 |
+
|
| 44 |
+
for (let i = 0; i < corpus_l.length - 1; i++) {
|
| 45 |
+
const b = corpus_l[i + 1]
|
| 46 |
+
|
| 47 |
+
if (b.startsWith('β')) continue
|
| 48 |
+
|
| 49 |
+
const a = corpus_l[i]
|
| 50 |
+
|
| 51 |
+
const bigram = `${a}|${b}`
|
| 52 |
+
|
| 53 |
+
if (pairs[bigram]) {
|
| 54 |
+
pairs[bigram] = pairs[bigram] + 1
|
| 55 |
+
} else {
|
| 56 |
+
pairs[bigram] = 1
|
| 57 |
+
}
|
| 58 |
+
}
|
| 59 |
+
|
| 60 |
+
return Object.fromEntries(
|
| 61 |
+
Object.entries(pairs).sort(([, a], [, b]) => {
|
| 62 |
+
if (a > b) {
|
| 63 |
+
return -1
|
| 64 |
+
}
|
| 65 |
+
|
| 66 |
+
if (b > a) {
|
| 67 |
+
return 1
|
| 68 |
+
}
|
| 69 |
+
|
| 70 |
+
return 0
|
| 71 |
+
}),
|
| 72 |
+
)
|
| 73 |
+
}
|
test/bpe.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
import { test } from 'node:test'
|
| 2 |
import herrgard from '../corpus/herrgard.txt' with { type: 'text' }
|
| 3 |
|
| 4 |
-
import { preprocess, initialVocabulary } from '../src/bpe.js'
|
| 5 |
|
| 6 |
test('preprocess', (t) => {
|
| 7 |
const expected =
|
|
@@ -83,3 +83,21 @@ test('initialVocabulary', (t) => {
|
|
| 83 |
|
| 84 |
t.assert.deepEqual(initialVocabulary(preprocess(herrgard)), expected)
|
| 85 |
})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import { test } from 'node:test'
|
| 2 |
import herrgard from '../corpus/herrgard.txt' with { type: 'text' }
|
| 3 |
|
| 4 |
+
import { preprocess, initialVocabulary, pairCount } from '../src/bpe.js'
|
| 5 |
|
| 6 |
test('preprocess', (t) => {
|
| 7 |
const expected =
|
|
|
|
| 83 |
|
| 84 |
t.assert.deepEqual(initialVocabulary(preprocess(herrgard)), expected)
|
| 85 |
})
|
| 86 |
+
|
| 87 |
+
test('pairCount', (t) => {
|
| 88 |
+
const expected = [
|
| 89 |
+
['β|s', 4147],
|
| 90 |
+
['d|e', 4016],
|
| 91 |
+
['β|h', 3974],
|
| 92 |
+
['e|n', 3439],
|
| 93 |
+
['β|d', 2469],
|
| 94 |
+
['a|n', 2441],
|
| 95 |
+
['t|t', 2178],
|
| 96 |
+
['a|r', 1970],
|
| 97 |
+
['e|t', 1954],
|
| 98 |
+
['β|a', 1940],
|
| 99 |
+
]
|
| 100 |
+
const result = Object.entries(pairCount(preprocess(herrgard))).slice(0, 10)
|
| 101 |
+
|
| 102 |
+
t.assert.deepEqual(result, expected)
|
| 103 |
+
})
|