ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detect ed shape was (19179,) + inhomogeneous part.

[ZXXSG]

when i run the code, I meet the following problem. I hope that you can help me. Thanks. (pytorch_gpu) G:\RNN-Capsule>python main.py --rnn_type LSTM --name_model 'LSTM-Capsule' --data_dir ./data-rt-acl --bidirectional Tr ue Namespace(batch_size=64, bidirectional=True, cell_dropout=0.5, data_dir='./data-rt-acl', embed_dropout=0.3, final_dropout=0.5, hidd en_dim=256, iter_num=8192, learning_rate=0.001, lr_word_vector=0.0001, max_length=64, n_label=2, n_layer=2, name_model="'LSTM-Capsu le'", optim_type='Adam', per_checkpoint=32, rnn_type='LSTM', seed=1705216, voc_size=32768, weight_decay=0, word_dim=300, wordvec_na me='glove.840B.300d.txt') Creating vocabulary... Loading word vectors... 16525 words found in vocab 2654 words not found in vocab Traceback (most recent call last): File "main.py", line 85, in vocab, embed, vocab_dict = datamanager.build_vocab('%s/%s' % (FLAGS.data_dir, FLAGS.wordvec_name), data['train']) File "G:\RNN-Capsule\datamanager.py", line 59, in build_vocab embed = np.array(embed, dtype=np.float32) ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (19179,) + inhomogeneous part.

[123kristin]

I have encountered the same problem as you. May I ask how you solved it?

[akakak47]

you need to modify the build_vocab function in datamanager.py to the following code

def build_vocab(self, path, data):
    print("Creating vocabulary...")
    vocab = {}
    for pair in data:
        for token in pair['sentence']:
            if token in vocab:
                vocab[token] += 1
            else:
                vocab[token] = 1
    vocab_list = sorted(vocab, key=vocab.get, reverse=True)
    if len(vocab_list) > self.FLAGS.voc_size:
        vocab_list = vocab_list[:self.FLAGS.voc_size]
    vocab_list.append('<unk>')

    print("Loading word vectors...")
    vectors = {}
    word_dim = None  # 用于存储词向量的维度
    with open('%s' % path) as f:
        for line in f:
            s = line.strip()
            word = s[:s.find(' ')]
            vector = s[s.find(' ')+1:].split()
            if word_dim is None:
                word_dim = len(vector)
            vectors[word] = np.array(vector, dtype=np.float32)

    embed = []
    num_not_found, num_found = 0, 0
    for word in vocab_list:
        if word in vectors:
            vector = vectors[word]
            num_found += 1
        else:
            num_not_found += 1
            vector = np.zeros((word_dim,), dtype=np.float32)
        embed.append(vector)

    print('%s words found in vocab' % num_found)
    print('%s words not found in vocab' % num_not_found)
    embed = np.array(embed, dtype=np.float32)
    return vocab_list, embed, vocab