def build_array_nmt(lines, vocab, num_steps):
"""Transform text sequences of machine translation into minibatches."""
lines = [vocab[l] for l in lines]
lines = [l + [vocab['<eos>']] for l in lines]
array = d2l.tensor(
[truncate_pad(l, num_steps, vocab['<pad>']) for l in lines])
valid_len = d2l.reduce_sum(d2l.astype(array != vocab['<pad>'], d2l.int32),
1)
return array, valid_len
def predict_ch8(prefix, num_preds, net, vocab, device): #@save
"""Generate new characters following the `prefix`."""
state = net.begin_state(batch_size=1, device=device)
outputs = [vocab[prefix[0]]]
get_input = lambda: d2l.reshape(d2l.tensor([outputs[-1]], device=device),
(1, 1))
for y in prefix[1:]: # Warm-up period
_, state = net(get_input(), state)
outputs.append(vocab[y])
for _ in range(num_preds): # Predict `num_preds` steps
y, state = net(get_input(), state)
outputs.append(int(y.argmax(dim=1).reshape(1)))
return ''.join([vocab.idx_to_token[i] for i in outputs])
def predict_ch8(prefix, num_preds, net, vocab, device): #@save
"""Generate new characters following the `prefix`."""
state = net.begin_state(batch_size=1, device=device)
outputs = [vocab[prefix[0]]]
get_input = lambda: d2l.reshape(d2l.tensor([outputs[-1]], device=device),
(1, 1))
for y in prefix[1:]: # Warm-up period
_, state = net(get_input(), state)
outputs.append(vocab[y])
for _ in range(num_preds): # Predict `num_preds` steps
y, state = net(get_input(), state)
# sample from multinomial instead of argmax
# outputs.append(int(torch.multinomial(F.softmax(y, dim=1), num_samples=1).reshape(1)))
# biased α = 2, must be integer
# α = 3
# outputs.append(int(torch.multinomial(F.softmax(y**α, dim=1), num_samples=1).reshape(1)))
outputs.append(int(y.argmax(dim=1).reshape(1)))
return ''.join([vocab.idx_to_token[i] for i in outputs])