python - ValueError : Layer weight shape (43, 100) 与提供的权重形状不兼容 (412457, 400)

Question

我为该项目准备了一个小数据集。它给出了

ValueError: Layer weight shape (43, 100) not compatible with provided weight shape (412457, 400)

错误。我认为标记器有问题。

train_test_split 的 X 和 Y

X = []
sentences = list(titles["title"])
for sen in sentences:
    X.append(preprocess_text(sen))

y = titles['Unnamed: 1']



X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, random_state=42)

这里是分词器

tokenizer = Tokenizer(num_words=5000)
tokenizer.fit_on_texts(X_train)

X_train = tokenizer.texts_to_sequences(X_train)
X_test = tokenizer.texts_to_sequences(X_test)

vocab_size = len(tokenizer.word_index) + 1 #vocab_size 43

maxlen = 100

X_train = pad_sequences(X_train, padding='post', maxlen=maxlen)
X_test = pad_sequences(X_test, padding='post', maxlen=maxlen)

所以，我的预训练 word2vec 模型具有 (412457, 400) 形状。

from numpy import array
from numpy import asarray
from numpy import zeros

from gensim.models import KeyedVectors
embeddings_dictionary = KeyedVectors.load_word2vec_format('drive/My Drive/trmodel', binary=True)

我使用预训练的 word2vec 模型而不是 GloVe。 (vocab_size:43、100，权重来自 embeddings_dictionary.vectors)

from keras.layers.recurrent import LSTM

model = Sequential()
embedding_layer = Embedding(vocab_size, 100, weights=[embeddings_dictionary.vectors], input_length=maxlen , trainable=False)
model.add(embedding_layer)
model.add(LSTM(128))

model.add(Dense(1, activation='sigmoid'))
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['acc'])

ValueError: Layer weight shape (43, 100) not compatible with provided weight shape (412457, 400)

Answer 1

如果您想使用预训练权重，则必须将适当的大小参数传递给嵌入层，以便它可以将预训练权重矩阵分配给嵌入层的权重矩阵。

因此您必须进行以下操作:

embedding_layer = Embedding(412457, 400, weights=[embeddings_dictionary.vectors], input_length=maxlen , trainable=False)

在训练之前，您必须更改填充以使其符合Embedding层:

maxlen = 400

X_train = pad_sequences(X_train, padding='post', maxlen=maxlen)
X_test = pad_sequences(X_test, padding='post', maxlen=maxlen)