public void WordPieceTokenizeWithCutomCocabTest(string input, string[] expected)
{
var vocabText = @"[UNK]
[CLS]
[SEP]
want
##want
##ed
wa
un
runn
##ing";
var table = Bert.LoadVocabularies(vocabText);
Assert.True(table.ContainsKey("[UNK]"));
Assert.True(table.ContainsKey("want"));
Assert.True(table.ContainsKey("##want"));
ArrayEqual(expected, BertTokenizer.WordPieceTokenize(input, table));
}
public void WordPieceTokenizeTest(string input, string[] expected)
{
ArrayEqual(expected, BertTokenizer.WordPieceTokenize(input, vocabularyTable));
}
public void BasicTokenizeTest(string input, string[] expected)
{
string[] result = BertTokenizer.BasicTokenize(input);
ArrayEqual(expected, result);
}
public void FullTokenizeTest(string input, string[] expected)
{
string[] result = BertTokenizer.Fulltokenize(input, vocabularyTable);
ArrayEqual(expected, result);
}
private ContentData PreProcess(string query, string content)
{
var queryTokens = BertTokenizer.Fulltokenize(query, vocabularyTable)
.Take(MAX_QUERY_LENTH);
var contentWords = content.Split((char c) => c.IsBertWhiteSpace(), StringSplitOptions.RemoveEmptyEntries);
var contentTokenIdxToWordIdxMapping = new List <int>();
var contentTokens = new List <string>();
for (int i = 0; i < contentWords.Length; i++)
{
var wordTokens = BertTokenizer.Fulltokenize(contentWords[i], vocabularyTable);
foreach (var subToken in wordTokens)
{
contentTokenIdxToWordIdxMapping.Add(i);
contentTokens.Add(subToken);
}
}
// -3 accounts for [CLS], [SEP] and [SEP]
// int maxContentLen = MAX_SEQ_LENTH - queryTokens.Count() - 3;
// contentTokens.AddRange(new string[MAX_SEQ_LENTH - queryTokens.Count() - 3 - contentTokens.Count]);
var tokens = new List <string>(MAX_SEQ_LENTH);
var segmentIds = new List <Int32>(MAX_SEQ_LENTH);
// Map token index to original index (in feature.origTokens).
var tokenIdxToWordIdxMapping = new Dictionary <int, int>();
// Start of generating the `InputFeatures`.
tokens.Add("[CLS]");
segmentIds.Add(0);
// For query input.
foreach (string t in queryTokens)
{
tokens.Add(t);
segmentIds.Add(0);
}
// For separation.
tokens.Add("[SEP]");
segmentIds.Add(0);
// For text input.
for (int i = 0; i < contentTokens.Count; i++)
{
tokens.Add(contentTokens[i]);
segmentIds.Add(1);
tokenIdxToWordIdxMapping[tokens.Count] = contentTokenIdxToWordIdxMapping[i];
}
// For ending mark.
tokens.Add("[SEP]");
segmentIds.Add(1);
ResetArray(inputs0);
ResetArray(inputs1);
ResetArray(inputs2);
for (int i = 0; i < tokens.Count; i++)
{
// Input IDs
inputs0[i] = vocabularyTable[tokens[i]];
// Input Mask
inputs1[i] = 1;
// Segment IDs
inputs2[i] = segmentIds[i];
}
return(new ContentData()
{
contentWords = contentWords,
tokenIdxToWordIdxMapping = tokenIdxToWordIdxMapping,
originalContent = content,
});
}
