strperc++

A C++ implementation of structured perceptron

Requirements

• googletest (Optional. This is used for unit tests.)
• gperftools (Optional. This is used for profiling)

Install

$ autoreconf -iv
$ ./configure --prefix=/path/to/install
$ make
$ make install

Data format

<FEATURE_1>[SPACE]<FEATURE_2>[SPACE] …  [SPACE]<LABEL>
<FEATURE_1>[SPACE]<FEATURE_2>[SPACE] …  [SPACE]<LABEL>
<FEATURE_1>[SPACE]<FEATURE_2>[SPACE] …  [SPACE]<LABEL>
…

Template format

Examples

CONLL2000

You can download one of famous benchmark data of text chunking as follows:

$ cd scripts
$ ./download_conll2000.sh

Then, a model is learned and an output is obtained as follows:

$ train_strpercpp -e 10 ./data/train.txt template.conll model
$ test_strpercpp -v 1 -d 0 model ./data/test.txt > output.txt

Evaluation metrics are calculated as follows:

$ ./scripts/conlleval -d "        " < output.txt
processed 47377 tokens with 23852 phrases; found: 23843 phrases; correct: 22295.
accuracy:  95.85%; precision:  93.51%; recall:  93.47%; FB1:  93.49
             ADJP: precision:  77.54%; recall:  74.89%; FB1:  76.19  423
             ADVP: precision:  83.70%; recall:  80.02%; FB1:  81.82  828
            CONJP: precision:  55.56%; recall:  55.56%; FB1:  55.56  9
             INTJ: precision: 100.00%; recall:  50.00%; FB1:  66.67  1
              LST: precision:   0.00%; recall:   0.00%; FB1:   0.00  0
               NP: precision:  93.75%; recall:  93.70%; FB1:  93.72  12415
               PP: precision:  96.87%; recall:  97.92%; FB1:  97.40  4863
              PRT: precision:  77.98%; recall:  80.19%; FB1:  79.07  109
             SBAR: precision:  88.27%; recall:  85.79%; FB1:  87.01  520
               VP: precision:  93.56%; recall:  93.90%; FB1:  93.73  4675

Results of existing models are listed in this page.

Development

For unit testing,

$ autoreconf -iv
$ ./configure --prefix=/path/to/install --with-gtest=/path/to/gtest
$ make check

Profiling result is obtained as follows:

$ autoreconf -iv
$ ./configure --prefix=/path/to/install --with-gperf=/path/to/gperf
$ make
$ make install
$ export CPUPROFILE=prof.out; time /path/to/install/bin/train_strpercpp [options]
$ pprof /path/to/install/bin/train_strpercpp prof.out

References

• Michael Collins, "Discriminative Training Methods for Hidden Markov Models: Theory and Experiments with Perceptron Algorithms.", EMNLP, 2002.
• Michael Collins and Brian Roark, "Incremental Parsing with the Perceptron Algorithm", ACL, 2004.