python - key 错误 : 'predictions' When use SimpleSeq2SeqPredictor to predict string

Question

请先在我们的 GitHub 存储库中搜索类似问题。如果您找不到类似的示例，您可以使用以下模板:

系统(请填写以下信息): - 操作系统:Ubuntu 18.04 - Python版本:3.6.7 - AllenNLP 版本:v0.8.3 - PyTorch 版本:1.1.0

问题当我尝试使用 SimpleSeq2SeqPredictor 预测字符串时，它总是显示

Traceback (most recent call last):
  File "predict.py", line 96, in <module>
    p = predictor.predict(i)
  File "venv/lib/python3.6/site-packages/allennlp/predictors/seq2seq.py", line 17, in predict
    return self.predict_json({"source" : source})
  File "/venv/lib/python3.6/site-packages/allennlp/predictors/predictor.py", line 56, in predict_json
    return self.predict_instance(instance)
  File "/venv/lib/python3.6/site-packages/allennlp/predictors/predictor.py", line 93, in predict_instance
    outputs = self._model.forward_on_instance(instance)
  File "/venv/lib/python3.6/site-packages/allennlp/models/model.py", line 124, in forward_on_instance
    return self.forward_on_instances([instance])[0]
  File "/venv/lib/python3.6/site-packages/allennlp/models/model.py", line 153, in forward_on_instances
    outputs = self.decode(self(**model_input))
  File "/venv/lib/python3.6/site-packages/allennlp/models/encoder_decoders/simple_seq2seq.py", line 247, in decode
    predicted_indices = output_dict["predictions"]
KeyError: 'predictions'

我尝试做一个翻译系统，但我是新手，大部分代码来自 https://github.com/mhagiwara/realworldnlp/blob/master/examples/mt/mt.py http://www.realworldnlpbook.com/blog/building-seq2seq-machine-translation-models-using-allennlp.html

这是我的训练代码

EN_EMBEDDING_DIM = 256
ZH_EMBEDDING_DIM = 256
HIDDEN_DIM = 256
CUDA_DEVICE = 0
prefix = 'small'

reader = Seq2SeqDatasetReader(
    source_tokenizer=WordTokenizer(),
    target_tokenizer=CharacterTokenizer(),
    source_token_indexers={'tokens': SingleIdTokenIndexer()},
    target_token_indexers={'tokens': SingleIdTokenIndexer(namespace='target_tokens')},
    lazy = True)
train_dataset = reader.read(f'./{prefix}-data/train.tsv')
validation_dataset = reader.read(f'./{prefix}-data/val.tsv')

vocab = Vocabulary.from_instances(train_dataset,
                                    min_count={'tokens': 3, 'target_tokens': 3})

en_embedding = Embedding(num_embeddings=vocab.get_vocab_size('tokens'),
                            embedding_dim=EN_EMBEDDING_DIM)
# encoder = PytorchSeq2SeqWrapper(
#     torch.nn.LSTM(EN_EMBEDDING_DIM, HIDDEN_DIM, batch_first=True))
encoder = StackedSelfAttentionEncoder(input_dim=EN_EMBEDDING_DIM, hidden_dim=HIDDEN_DIM, projection_dim=128, feedforward_hidden_dim=128, num_layers=1, num_attention_heads=8)

source_embedder = BasicTextFieldEmbedder({"tokens": en_embedding})

# attention = LinearAttention(HIDDEN_DIM, HIDDEN_DIM, activation=Activation.by_name('tanh')())
# attention = BilinearAttention(HIDDEN_DIM, HIDDEN_DIM)
attention = DotProductAttention()

max_decoding_steps = 20   # TODO: make this variable
model = SimpleSeq2Seq(vocab, source_embedder, encoder, max_decoding_steps,
                        target_embedding_dim=ZH_EMBEDDING_DIM,
                        target_namespace='target_tokens',
                        attention=attention,
                        beam_size=8,
                        use_bleu=True)
optimizer = optim.Adam(model.parameters())
iterator = BucketIterator(batch_size=32, sorting_keys=[("source_tokens", "num_tokens")])

iterator.index_with(vocab)
if torch.cuda.is_available():
    cuda_device = 0
    model = model.cuda(cuda_device)
else:
    cuda_device = -1
trainer = Trainer(model=model,
                    optimizer=optimizer,
                    iterator=iterator,
                    train_dataset=train_dataset,
                    validation_dataset=validation_dataset,
                    num_epochs=50,
                    serialization_dir=f'ck/{prefix}/',
                    cuda_device=cuda_device)

# for i in range(50):
    # print('Epoch: {}'.format(i))
trainer.train()

predictor = SimpleSeq2SeqPredictor(model, reader)

for instance in itertools.islice(validation_dataset, 10):
    print('SOURCE:', instance.fields['source_tokens'].tokens)
    print('GOLD:', instance.fields['target_tokens'].tokens)
    print('PRED:', predictor.predict_instance(instance)['predicted_tokens'])

# Here's how to save the model.
with open(f"ck/{prefix}/manually_save_model.th", 'wb') as f:
    torch.save(model.state_dict(), f)
vocab.save_to_files(f"ck/{prefix}/vocabulary")

这是我的预测代码

EN_EMBEDDING_DIM = 256
ZH_EMBEDDING_DIM = 256
HIDDEN_DIM = 256
CUDA_DEVICE = 0
prefix = 'big'

reader = Seq2SeqDatasetReader(
    source_tokenizer=WordTokenizer(),
    target_tokenizer=CharacterTokenizer(),
    source_token_indexers={'tokens': SingleIdTokenIndexer()},
    target_token_indexers={'tokens': SingleIdTokenIndexer(namespace='target_tokens')},
    lazy = True)
# train_dataset = reader.read(f'./{prefix}-data/train.tsv')
# validation_dataset = reader.read(f'./{prefix}-data/val.tsv')

# vocab = Vocabulary.from_instances(train_dataset,
#                                     min_count={'tokens': 3, 'target_tokens': 3})
vocab = Vocabulary.from_files("ck/small/vocabulary")

en_embedding = Embedding(num_embeddings=vocab.get_vocab_size('tokens'),
                            embedding_dim=EN_EMBEDDING_DIM)
# encoder = PytorchSeq2SeqWrapper(
#     torch.nn.LSTM(EN_EMBEDDING_DIM, HIDDEN_DIM, batch_first=True))
encoder = StackedSelfAttentionEncoder(input_dim=EN_EMBEDDING_DIM, hidden_dim=HIDDEN_DIM, projection_dim=128, feedforward_hidden_dim=128, num_layers=1, num_attention_heads=8)

source_embedder = BasicTextFieldEmbedder({"tokens": en_embedding})

# attention = LinearAttention(HIDDEN_DIM, HIDDEN_DIM, activation=Activation.by_name('tanh')())
# attention = BilinearAttention(HIDDEN_DIM, HIDDEN_DIM)
attention = DotProductAttention()

max_decoding_steps = 20   # TODO: make this variable
model = SimpleSeq2Seq(vocab, source_embedder, encoder, max_decoding_steps,
                        target_embedding_dim=ZH_EMBEDDING_DIM,
                        target_namespace='target_tokens',
                        attention=attention,
                        beam_size=8,
                        use_bleu=True)

# And here's how to reload the model.
with open("./ck/small/best.th", 'rb') as f:
    model.load_state_dict(torch.load(f))

predictor = Seq2SeqPredictor(model, dataset_reader=reader)
# print(predictor.predict("The dog ate the apple"))


test = [
    'Surely ,he has no power over those who believe and put their trust in their Lord ;',
    'And assuredly We have destroyed the generations before you when they did wrong ,while their apostles came unto them with the evidences ,and they were not such as to believe . In this wise We requite the sinning people .',
    'And warn your tribe ( O Muhammad SAW ) of near kindred .',
    'And to the Noble Messengers whom We have mentioned to you before ,and to the Noble Messengers We have not mentioned to you ; and Allah really did speak to Moosa .',
    'It is He who gave you hearing ,sight ,and hearts ,but only few of you give thanks .',
    'spreading in them much corruption ?',
    'That will envelop the people . This will be a painful punishment .',
    'When you received it with your tongues and spoke with your mouths what you had no knowledge of ,and you deemed it an easy matter while with Allah it was grievous .',
    'of which you are disregardful .',
    'Whoever disbelieves ,then the calamity of his disbelief is only on him ; and those who do good deeds ,are preparing for themselves .'
]




for i in test:
    p = predictor.predict(i) # <------------------- ERROR !!!!!!!
    print(p) 

我做错了什么吗？

Answer 1

解决了我忘记在加载模型后添加 model.eval()。对不起