def test_dataloader(self):
return DataLoader(StringMatchingDataset(self.test_dataset),
shuffle=True,
num_workers=NUM_WORKERS,
batch_size=BATCH_SIZE,
collate_fn=PadSequence())
for sentence in words]
test_pos_ids = [[pos_vocab.get(pos) for pos in sentence] for sentence in pos]
embedding = torch.from_numpy(embedding).float()
model = LSTMTagger(embedding, embed_dim, 100, 2, len(pos_vocab)).to(device)
optimizer = optim.Adam(
[param for param in model.parameters() if param.requires_grad == True],
lr=0.001)
criterion = nn.NLLLoss()
for epoch in range(num_epochs):
batch = Batch(word_ids, pos_ids, batch_size=batch_size)
total_step = len(batch)
i = 0
for inputs, labels in batch:
i += 1
pad_words_obj = PadSequence(inputs, [len(inputs), 100])
padded_inputs = torch.Tensor(pad_words_obj.embedding).long().to(device)
padded_inputs_lens = torch.Tensor(
pad_words_obj.lengths).long().to(device)
outputs = model(padded_inputs, padded_inputs_lens)
pad_pos_obj = PadSequence(
labels,
[len(inputs), outputs.size(2)]) # batch, num_classes, seq_len
padded_labels = torch.Tensor(pad_pos_obj.embedding).long().to(device)
loss = criterion(outputs, padded_labels)
optimizer.zero_grad()
loss.backward()