| 142 | |
| 143 | |
| 144 | class LSTMModel(model.Model): |
| 145 | |
| 146 | def __init__(self, hidden_size, seq_length, batch_size, bidirectional, |
| 147 | num_layers, return_sequences, rnn_mode, batch_first): |
| 148 | super(LSTMModel, self).__init__() |
| 149 | self.hidden_size = hidden_size |
| 150 | self.seq_length = seq_length |
| 151 | self.return_sequences = return_sequences |
| 152 | |
| 153 | self.lstm = layer.CudnnRNN(hidden_size=hidden_size, |
| 154 | num_layers=num_layers, |
| 155 | bidirectional=bidirectional, |
| 156 | return_sequences=return_sequences, |
| 157 | rnn_mode=rnn_mode, |
| 158 | batch_first=batch_first) |
| 159 | self.optimizer = opt.SGD(0.1) |
| 160 | |
| 161 | def forward(self, x): |
| 162 | y = self.lstm(x) |
| 163 | if self.return_sequences: |
| 164 | y = autograd.reshape(y, (-1, self.seq_length * self.hidden_size)) |
| 165 | return y |
| 166 | |
| 167 | |
| 168 | class TestModelMethods(unittest.TestCase): |
no outgoing calls