English
资源说明：HMM model for word representations, using the method of Huang + Yates (2009).
HMM model for word representations, using the method of Huang + Yates (2009).
http://www.cis.temple.edu/~yates/papers/smoothing-acl09.pdf
Implemented by Joseph Turian
============
USAGE
* Edit hyperparameters.wordrepresentations-hmm.yaml or use command
line flags. You will need integers files as the input.
See these scripts:
http://github.com/turian/common-scripts/blob/master/words-integers-mapfile.py
http://github.com/turian/common-scripts/blob/master/words-to-integers.py
For example:
words-integers-mapfile.py < tiny.txt > integersmap-tiny.txt
words-to-integers.py integersmap-tiny.txt < tiny.txt > integers-tiny.txt
To get the vocabulary for hyperparameters.wordrepresentations-hmm.yaml:
wc -l integersmap-tiny.txt
# 691
* Run ./train.py on a file with mapped integers
./train.py < integers-tiny.txt
* Decode a file with real words, not integers
./decode.py < tiny.txt
============
REQUIREMENTS
* ghmm, with Python bindings
http://ghmm.sourceforge.net/
* common Python library by Joseph Turian
http://github.com/turian/common
* numpy
For random initialization
============
WARNING
* I have not checked with Huang and Yates to see if this
implementation is correct.
* I am not sure if my Baum-Welch stopping criterion is appropriate.
* I do not know if the HMM initialization is correct.
============
TODO
* Automatically do word mapping during train.py, not just decode.py
本源码包内暂不包含可直接显示的源代码文件，请下载源码包。
