python - Keras 使用 LSTM 进行情感分析如何测试

Question

我正在尝试使用 Keras 对我的文本进行情感分析，使用示例 imdb_lstm.py但我不知道如何测试它。 我将模型和权重存储到文件中，如下所示:

model = model_from_json(open('my_model_architecture.json').read())
model.compile(loss='binary_crossentropy',
          optimizer='adam',
          metrics=['accuracy'])
model.load_weights('my_model_weights.h5')

results = model.evaluate(X_test, y_test, batch_size=32)

但我当然不知道 X_test 和 y_test 应该是什么样子。有人可以帮助我吗？

Answer 1

首先，将数据集拆分为测试、有效和训练并进行一些预处理:

from tensorflow import keras

print('load data')
(x_train, y_train), (x_test, y_test) = keras.datasets.imdb.load_data(num_words=10000)
word_index = keras.datasets.imdb.get_word_index()

print('preprocessing...')
x_train = keras.preprocessing.sequence.pad_sequences(x_train, maxlen=256)
x_test = keras.preprocessing.sequence.pad_sequences(x_test, maxlen=256)

x_val = x_train[:10000]
y_val = y_train[:10000]

x_train = x_train[10000:]
y_train = y_train[10000:]

如您所见，我们还加载了 word_index，因为稍后我们需要它来将句子转换为整数序列。

第二，定义您的模型:

print('build model')
model = keras.Sequential()
model.add(keras.layers.Embedding(10000, 16))
model.add(keras.layers.LSTM(100))
model.add(keras.layers.Dense(16, activation='relu'))
model.add(keras.layers.Dense(1, activation='sigmoid'))

model.compile(optimizer='adam',
              loss='binary_crossentropy',
              metrics=['accuracy'])

print('train model')
model.fit(x_train,
          y_train,
          epochs=5,
          batch_size=512,
          validation_data=(x_val, y_val),
          verbose=1)

最后，保存并加载您的模型:

print('save trained model...')
model.save('sentiment_keras.h5')
del model

print('load model...')
from keras.models import load_model
model = load_model('sentiment_keras.h5')

您可以使用test-set评估您的模型:

print('evaluation')
evaluation = model.evaluate(x_test, y_test, batch_size=512)
print('Loss:', evaluation[0], 'Accuracy:', evaluation[1])

如果您想在全新的句子上测试模型，您可以这样做:

sample = 'this is new sentence and this very bad bad sentence'
sample_label = 0
# convert input sentence to tokens based on word_index
inps = [word_index[word] for word in sample.split() if word in word_index]
# the sentence length should be the same as the input sentences
inps = keras.preprocessing.sequence.pad_sequences([inps], maxlen=256)
print('Accuracy:', model.evaluate(inps, [sample_label], batch_size=1)[1])
print('Sentiment score: {}'.format(model.predict(inps)[0][0]))