/// <summary>
/// 加载已有模型
/// </summary>
public static Seq2Seq Load(string modelname = "Model.lys2s")
{
//var tosave = new ModelData();
BinaryFormatter bf = new BinaryFormatter();//用二进制保存和加载，必须是相同的程序集才可以加载
FileStream fs = new FileStream(modelname, FileMode.Open, FileAccess.Read);
var mdl = bf.Deserialize(fs) as ModelData;
//var mdl = bf.Deserialize(fs) as ModelAttentionData;
fs.Close();
fs.Dispose();
var s2s = new Seq2Seq();
//s2s.bd =new WeightMatrix( mdl.bd);
//s2s.Whd =new WeightMatrix( mdl.Whd);
//s2s.Embedding = new WeightMatrix( mdl.Wil);
//s2s.decoder =new LinyeeDecoder( mdl.decoder);
//s2s.encoder =new Encoder( mdl.encoder);
//s2s.ReversEncoder =new Encoder( mdl.ReversEncoder);
s2s.bd = (mdl.bd);
s2s.Whd = (mdl.Whd);
s2s.Embedding = (mdl.Wil);
s2s.decoder = (mdl.decoder);
s2s.encoder = (mdl.encoder);
s2s.ReversEncoder = (mdl.ReversEncoder);
s2s.hidden_size = mdl.hidden_sizes;
s2s.word_size = mdl.letter_size;
s2s.Depth = mdl.Depth;
s2s.clipval = mdl.clipval;
s2s.learning_rate = mdl.learning_rate;
s2s.max_word = 300;
s2s.regc = mdl.regc;
s2s.UseDropout = mdl.UseDropout;
//s2s.wordToIndex = mdl.wordToIndex;
s2s.indexToWord = mdl.indexToWord;
Dictionary <string, WordInfoV1_1> w2i = new Dictionary <string, WordInfoV1_1>();
foreach (var kv in s2s.indexToWord)
{
w2i.Add(kv.Value.w, kv.Value);
}
s2s.wordToIndex = w2i;
s2s.vocab = new List <string>();
s2s.vocab.AddRange(w2i.Keys);
s2s.newType = "retrain";
//s2s.Preprocess();
//s2s.Save("Model.lys2s");
return(s2s);
}
/// <summary>
/// 保存模型
/// </summary>
public static void Save(this Seq2Seq s2s, bool hadinback = true, string ModelName = null)
{
if (inSave && hadinback)
{
return;
}
lock (lockSave)
{
inSave = true;
var mname = ModelName;
if (string.IsNullOrEmpty(mname))
{
mname = s2s.ModelName;
}
ModelData tosave = new ModelData();
//tosave.wordToIndex = s2s.wordToIndex;
tosave.indexToWord = s2s.indexToWord;
tosave.bd = s2s.bd;
tosave.clipval = s2s.clipval;
tosave.decoder = s2s.decoder;
tosave.Depth = s2s.Depth;
tosave.encoder = s2s.encoder;
tosave.hidden_sizes = s2s.hidden_size;
tosave.learning_rate = s2s.learning_rate;
tosave.letter_size = s2s.word_size;
tosave.max_chars_gen = s2s.max_word;
tosave.regc = s2s.regc;
tosave.ReversEncoder = s2s.ReversEncoder;
tosave.UseDropout = s2s.UseDropout;
tosave.Whd = s2s.Whd;
tosave.Wil = s2s.Embedding;
BinaryFormatter bf = new BinaryFormatter();
FileStream fs = new FileStream(mname, FileMode.OpenOrCreate, FileAccess.Write);
bf.Serialize(fs, tosave);
fs.Close();
fs.Dispose();
inSave = false;
}
}
internal static void Preprocess(this Seq2Seq s2s)
{
/// <summary>
/// 输入语句
/// </summary>
List <List <string> > input = new List <List <string> >();
/// <summary>
/// 输出语句
/// </summary>
List <List <string> > output = new List <List <string> >();
var HumanTextRaw = File.ReadAllLines("human_text.txt");
var RobotTextRaw = File.ReadAllLines("robot_text.txt");
for (int i = 0; i < HumanTextRaw.Length; i++)
{
HumanTextRaw[i] = RemoveAccentMark(HumanTextRaw[i]);
RobotTextRaw[i] = RemoveAccentMark(RobotTextRaw[i]);
}
var HumanText = new List <string>();
var RobotText = new List <string>();
for (int i = 0; i < 1000; i++)
{
HumanText.Add(HumanTextRaw[i]);
RobotText.Add(RobotTextRaw[i]);
}
for (int i = 0; i < HumanText.Count; i++)
{
input.Add(HumanText[i].ToLower().Trim().Split(' ').ToList());
output.Add(RobotText[i].ToLower().Trim().Split(' ').ToList());
}
s2s.OneHotEncoding(input, output);
}