Overview

HTTPS SSH
E D A

EDA is a word-based dependency parser.

Documentation can be found at:
http://www.ar.media.kyoto-u.ac.jp/members/flannery/eda/

Building EDA:

You need to install the Boost C++ libraries to build
EDA. Additionally, you need g++ version 3.4 or greater. Assuming that
these requirements have been met, you can build EDA by running the
following commmand.

> make

Installing EDA:

After you finish building the source, you can install EDA by copying
the executables and model files to installation directories of your choice.

> cp src/eda/eda /my/install/dir
> cp src/eda/train-eda /my/install/dir
> cp data/*.dat /my/models/dir

Running EDA:
After installation, use the following command to parse the
sample sentences with the default model.

> eda -e data/sample.tree -v data/jp-0.1.0-utf8-vocab-small.dat \
-w data/jp-0.1.0-utf8-weight-small.dat

By default the results will be written to the standard output, but you
can redirect the results to a file like this.

> eda -e data/sample.tree -v data/jp-0.1.0-utf8-vocab-small.dat \
-w data/jp-0.1.0-utf8-weight-small.dat > data/sample_out.tree

Training Models:
If you annotate text in EDA's format, you can train your own model for
parsing text and use it instead of the default model. The following
command creates model files 'v.dat' and 'w.dat' from the training file
'train.tree'. Specify the model files with the 'eda' command's -v and
-w options to parse text with the new model.

> train-eda -t train.tree -a llsgd -i 10 -c 3 -v v.dat -w w.dat

Measuring Accuracy:
You can then measure the dependency accuracy of a file like the one
created above against a test set of dependencies for the same file
with the following command.

> perl eval.pl data/sample_test.tree data/sample_out.tree