fasttext

fasttext with hierarchical softmax, implemented by tensorflow.

The corpus can be find here.

reference: Bag of Tricks for Efficient Text Classification.

Requirements

• python 3.4 or newer
• tensorflow 0.12.0rc1

About the code

The huffman tree should be constructed before training model.

paths_length.npy    the length of huffman coding of every label.

cooking.train       train file , contained 12404 examples.

cooking.valid       test file , contained 3000 examples.

labels.dict         the dict of labels.

words.dict          the dict of all words.