go-easy-first

go-easy-first - Dependency Parser with Easy-First Algorithm (An Efficient Algorithm for Easy-First Non-Directional Dependency Parsing, NAACL-2010, Yoav Goldberg and Michael Elhadad) written in Go.

Build from source

% git clone https://github.com/syou6162/go-easy-first.git
% cd go-easy-first
% make deps && make bindata && make build

Usage

go-easy-first has train (training a parser phase) and eval (evaluating a trained parser phase) modes. To see the detail options, type ./go-easy-first --help.

Training a parser

To see the detail options, type ./go-easy-first train --help.

% ./go-easy-first train --train-filename path/to/train.txt --dev-filename path/to/dev.txt --max-iter 10 --model-filename model.bin
0, 0.907, 0.893
1, 0.920, 0.901
2, 0.929, 0.904
3, 0.935, 0.906
4, 0.940, 0.907
5, 0.944, 0.907
6, 0.947, 0.908
7, 0.950, 0.908
8, 0.953, 0.908
9, 0.955, 0.908

Evaluating a trained parser

To see the detail options, type ./go-easy-first eval --help.

% ./go-easy-first eval --test-filename path/to/test.txt --model-filename model.bin
| SENTENCES | SECONDS | ACCURACY |
|-----------|---------|----------|
|      1346 |    4.60 |    0.888 |

[Experimental] Obtain a single binary embedded model parameters

% ./go-easy-first train --train-filename path/to/train.txt --dev-filename path/to/dev.txt --max-iter 10 --model-filename data/model.bin
% make bindata && make build
% ./go-easy-first decode --test-filename path/to/test.txt

Roadmap

• Implement PP-Attachment features
• Beam search with max-violation perceptron
• Mini-batch update
• Embed weight parameters to a built binary file using go-bindata

Author

Yasuhisa Yoshida