SciTech-BigDataAIML-NLP-TensorFlow-KerasNLP-预处理: tensorflow.keras.preprocess.{text, sequence}

KerasNLP API reference

The easiest way to get started processing text in TensorFlow is to use KerasNLP, a natural language processing library that provides modular components with state-of-the-art preset weights and architectures. You can use KerasNLP components out-of-the-box or customize them as needed. KerasNLP emphasizes in-graph computation for all workflows, so you can expect easy productionization using the TensorFlow ecosystem.

To install KerasNLP, see Installation.

TensorFlow 的“新版”API如下：

https://tensorflow.google.cn/text/api_overview?hl=en

https://tensorflow.google.cn/text/api_docs/python/text/Splitter?hl=en

如下的“旧版”API已经为“新版”取代；

tensorflow.keras.preprocess.{text, sequence}

from tensorflow.keras.preprocess import text.Tokenizer as txtTok

from tensorflow.keras.preprocess import sequence.pad_sequences as seqPad

如上示例：

TensorFlow 的 keras.preprocess 集合包，

有 text 与 sequence 等多种类型的分包，适用于预处理对应类型的数据。

但是NLP的原理是通用的；大体上：

text -Tokenizer -> token sequence

sequence -padding/truncating-> RaggedTensor 或 RegularizedMatrixs