卷积网的Keras参数(Keras Parameters for Convolutional Net)

卷积网的Keras参数(Keras Parameters for Convolutional Net)

我正在研究Keras卷积神经网络的例子。 （例如，请参阅https://github.com/fchollet/keras/blob/master/examples/imdb_cnn.py 。）但是，我无法弄清楚“maxlen”参数的含义。 它会与填充有关吗？ 它不是最大数量的功能; 他们有一个max_features参数。

I'm looking at Keras' example for convolutional neural networks. (See https://github.com/fchollet/keras/blob/master/examples/imdb_cnn.py for example.) However, I cannot figure out what they mean by the "maxlen" parameter. Would it have something to do with padding? It isn't the maximum number of features; they have a max_features parameter for that.

最满意答案

maxlen参数是单词中文本样本的长度。

在Keras代码示例中，您有以下设置：

# set parameters: max_features = 5000 maxlen = 400 ... embedding_dims = 50

这意味着您有5000个单词的词汇表，每个单词都嵌入到具有50个维度的特征向量中，每个文本样本可以长达400个单词。

间接地，当文本样本短于400个单词时，这也与填充有关。 然后你必须将它们填充到400的长度。

对于用于文本分类的1D-ConvNets，请参阅本文和此博客文章：

https://arxiv.org/abs/1408.5882

http://www.wildml.com/2015/11/understanding-convolutional-neural-networks-for-nlp/

The maxlen parameter is the length of your text samples in words.

In the Keras code example you have these settings:

# set parameters: max_features = 5000 maxlen = 400 ... embedding_dims = 50

This means you have a vocabulary of 5000 words, each of these words are embedded into a feature vector with 50 dimensions and each of your text samples can be 400 words long.

Indirectly this also has a relation to padding when you have text samples that are shorter than 400 words. Then you have to pad these to a length of 400.

For 1D-ConvNets for text classification see also this paper and this blog post:

https://arxiv.org/abs/1408.5882

http://www.wildml.com/2015/11/understanding-convolutional-neural-networks-for-nlp/