def test_linear():
m = nn.Linear(2, 3)
m_ = load(dump(m))
assert isinstance(m_, nn.Linear)
assert m_.in_features == m.in_features
assert m_.out_features == m.out_features
assert (m_.bias is None) == (m.bias is None)
def test_conv1d():
m = nn.Conv1d(2, 3, 4)
m_ = load(dump(m))
assert isinstance(m_, nn.Conv1d)
assert m_.in_channels == m.in_channels
assert m_.out_channels == m.out_channels
assert m_.kernel_size == m.kernel_size
def test_module_list():
m = nn.ModuleList([nn.Linear(2, 3), nn.Embedding(10, 20)])
m_ = load(dump(m))
assert isinstance(m_, nn.ModuleList)
assert len(m_) == len(m)
assert isinstance(m_[0], nn.Linear)
assert isinstance(m_[1], nn.Embedding)
def test_embedding():
m = nn.Embedding(10, 20, padding_idx=5)
m_ = load(dump(m))
assert isinstance(m_, nn.Embedding)
assert m_.num_embeddings == m.num_embeddings
assert m_.embedding_dim == m.embedding_dim
assert m_.padding_idx == m.padding_idx
def test_lstm():
m = nn.LSTM(10,
20,
num_layers=2,
dropout=0.3,
batch_first=True,
bidirectional=True)
m_ = load(dump(m))
assert isinstance(m_, nn.LSTM)
assert m_.input_size == m.input_size
assert m_.hidden_size == m.hidden_size
assert m_.num_layers == m.num_layers
assert m_.dropout == m.dropout
assert m_.batch_first == m.batch_first
assert m_.bidirectional == m.bidirectional
def test_language_model():
m = LanguageModel(
nn.Embedding(2, 3),
nn.Embedding(4, 5),
nn.ModuleList([nn.Conv1d(2, 3, 4),
nn.Conv1d(3, 4, 5)]),
Highway(3, 4),
nn.LSTM(10, 20),
nn.Linear(4, 5),
)
m_ = load(dump(m))
assert isinstance(m_, LanguageModel)
assert isinstance(m_.word_emb, nn.Embedding)
assert isinstance(m_.char_emb, nn.Embedding)
assert isinstance(m_.char_convs, nn.ModuleList)
assert isinstance(m_.highway, Highway)
assert isinstance(m_.lstm, nn.LSTM)
assert isinstance(m_.output_layer, nn.Linear)
def test_highway():
m = Highway(10, 20)
m_ = load(dump(m))
assert isinstance(m_, Highway)
assert m_.size == m.size
assert m_.num_layers == m.num_layers