def test_randomly_initialized_non_trainable(self):
vocab_size = 10
emb_size = 20
trainable = False
padding_idx = 0
layer = modules.EmbeddingLayer(vocab_size, emb_size, trainable,
padding_idx)
x = torch.tensor([[1, 2, 0], [1, 2, 3]], dtype=torch.long)
x = layer(x)
self.assertListEqual(list(x.shape), [2, 3, 20])
self.assertListEqual(x[0][2].tolist(),
torch.tensor([0] * emb_size).tolist())
def test_randomly_initialized_trainable(self):
vocab_size = 10
emb_size = 20
trainable = True
padding_idx = 0
layer = modules.EmbeddingLayer(vocab_size, emb_size, trainable,
padding_idx)
x = torch.tensor([[1, 2, 0], [1, 2, 3]], dtype=torch.long)
x = layer(x)
x.sum().backward()
grad = next(layer._embedding_layer.parameters()).grad
self.assertListEqual(list(x.shape), [2, 3, 20])
self.assertListEqual(grad[0].tolist(),
torch.tensor([0] * emb_size).tolist())
def test_non_randomly_initialized_non_trainable(self):
vocab_size = 3
emb_size = 6
trainable = False
padding_idx = 0
embeddings = np.array([[1, 2, 3, 4, 5, 6], [2, 5, 1, 2, 3, 4],
[9, 1, 7, 2, 3, 7]])
layer = modules.EmbeddingLayer(vocab_size, emb_size, trainable,
padding_idx)
layer.load_embeddings(embeddings)
x = torch.tensor([[1, 2, 0], [1, 2, 1]], dtype=torch.long)
x = layer(x)
self.assertListEqual(list(x.shape), [2, 3, emb_size])
self.assertListEqual(x[0][2].tolist(),
torch.tensor(embeddings[0]).tolist())
def test_non_randomly_initialized_trainable(self):
vocab_size = 3
emb_size = 6
trainable = True
padding_idx = 0
embeddings = np.array([[1, 2, 3, 4, 5, 6], [2, 5, 1, 2, 3, 4],
[9, 1, 7, 2, 3, 7]])
layer = modules.EmbeddingLayer(vocab_size, emb_size, trainable,
padding_idx)
layer.load_embeddings(embeddings)
x = torch.tensor([[1, 2, 0], [1, 2, 1]], dtype=torch.long)
x = layer(x)
x.sum().backward()
grad = next(layer._embedding_layer.parameters()).grad
self.assertListEqual(list(x.shape), [2, 3, emb_size])
self.assertListEqual(grad[0].tolist(),
torch.tensor([0] * emb_size).tolist())