glove-py is an implementation of the GloVe algorithm for learning word vectors from a corpus.

Installation

using pip:

pip install glove-py

install with github code :

git clone git@github.com:f20500909/glove_py.git
python3 setup.py install

using in shell

Run

with python:

from glove import *

model = Glove(params)
model.train(input_file)
model.to_txt()
words = model.most_similary("one", 10)
print(words)

List of available params and their default value:

help info......
  -input_file               input_file [small_text]
  -vocab_size               vocabulary size to train  [0]
  -max_size                 maximum size to keep in courpus [10000000]
  -min_count                minimum times  [1]
  -window                   the window size to get common occurrence [10]
  -embed_size               embedded size  [80]
  -epoch                    train epoch [1]
  -threads                  train thread [40]
  -memory_limit             memory used to tain (GB) [1]
  -lr                       learning rate [0.05]

with shell: ./scripts/demo.sh

我去掉了很过我看上去用处不大的功能(只是我认为的用处打大的功能,有些细节并不理解也移除了),如提取共现矩阵的symmetric功能

忽略了大小写,统一改大写为小写

TODO:

修复有时会发生的错误: read err...: Connection reset by peer

在合并共现词时，高频和低频矩阵好像写在了一个文件里面，感觉有点问题.....

在编译pybind_11时不能让对应文件被include 必须用cmake加载

Reference

• GloVe
• GloVe-cpp
• Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. GloVe: Global Vectors for Word Representation.