public CntkTrainDataSet Transform(DataSet dataSet)
{
var segmentator = new MeCabTextSegmentator();
var topicStore = new CntkTopicStore();
var corpus = new HashSet <string>();
int sentenceId = 0;
var n = 0;
//var count = sentences.Count;
List <CntkSentence> sentences = dataSet.Select(row =>
{
n++;
if (n % 10 == 0)
{
Console.WriteLine(n + " sentences");
}
CntkTopic topic = topicStore.GetOrRegister(row.Label);
// クレンジングをかけつつ単語収集
List <string> words = segmentator.Split(row.Sentence);
words.ForEach(w => corpus.Add(w));
return(new CntkSentence()
{
Id = sentenceId++,
Topic = topic,
Sentence = row.Sentence,
Words = words.Select(w => new CntkWord()
{
Text = w
}).ToList(),
});
}).ToList();
// 取り込む単語が全て確定しないとVector表現が決まらないので、そこだけ最後
var space = OneHotWordVectorSpace.Build(corpus);
foreach (var sentence in sentences)
{
foreach (var word in sentence.Words)
{
word.Value = space.ToVector(word.Text);
}
}
var trainData = new CntkTrainDataSet(topicStore.Indexer, space, sentences);
return(trainData);
}
/// <summary>
/// 文章を分類器にかけるための単語シーケンスに変換します。
/// </summary>
/// <param name="query"></param>
/// <param name="recipe"></param>
/// <returns></returns>
private List <string> ToWordSequence(string query)
{
List <string> wordSequence = segmantator.Split(query, true);
return(wordSequence);
}