public void Load()
{
ModelAttentionData tosave = new ModelAttentionData();
BinaryFormatter bf = new BinaryFormatter();
FileStream fs = new FileStream("Model.bin", FileMode.Open, FileAccess.Read);
tosave = bf.Deserialize(fs) as ModelAttentionData;
fs.Close();
fs.Dispose();
this.bd = tosave.bd;
this.clipval = tosave.clipval;
this.decoder = tosave.decoder;
this.Depth = tosave.Depth;
this.encoder = tosave.encoder;
this.hidden_size = tosave.hidden_sizes;
this.learning_rate = tosave.learning_rate;
this.word_size = tosave.letter_size;
this.max_word = 100;
this.regc = tosave.regc;
this.ReversEncoder = tosave.ReversEncoder;
this.UseDropout = tosave.UseDropout;
this.Whd = tosave.Whd;
this.Embedding = tosave.Wil;
}
public AttentionSeq2Seq(int inputSize, int hiddenSize, int depth, List <List <string> > input, List <List <string> > output, bool useDropout)
{
this.InputSequences = input;
this.OutputSequences = output;
this.Depth = depth;
// list of sizes of hidden layers
word_size = inputSize; // size of word embeddings.
this.hidden_size = hiddenSize;
solver = new Optimizer();
OneHotEncoding(input, output);
this.Whd = new WeightMatrix(hidden_size, vocab.Count + 2, true);
this.bd = new WeightMatrix(1, vocab.Count + 2, 0);
Embedding = new WeightMatrix(vocab.Count + 2, word_size, true);
encoder = new Encoder(hidden_size, word_size, depth);
ReversEncoder = new Encoder(hidden_size, word_size, depth);
decoder = new AttentionDecoder(hidden_size, word_size, depth);
}
