This readme is out of date. Stay tuned for an updated version...

Running the PSL harness requires python, java and maven 3.

Currently, the installation process is:

1. in psl/ run, install_psl.sh: this will use maven to install the PSL binaries in the maven repository (default ~/.m2)

2. Set up config files:

• psl/psl-rss/src/main/resources/log4j.properties sets up PSL's logging.

• psl/psl-rss/src/main/resources/psl.properties contains various settings for PSL. Most important are:

  • rss.dbpath file system path where H2 database file will be stored

  • rss.auxdatapath file system path where gazatteer, and precomputed predicates are stored (currently psl/psl-rss/aux_data)

  • rss.jsonserver.port local port to send messages from python to java

  • rss.jsonserver.processor name of PSL program to run on messages (currently RSSLocationProcessor)

  • rss.rsslocationprocessor.outputdir directory to dump json text files of PSL-predicted locations. Useful for offline evaluation, but it's best to comment this out for production so no files will be output.

0. To set where PSL and log4j will look for configuration files, use the system properties -Dlog4j.configuration=file://[log4j config file path] and -Dpsl.configuration=file://[psl config file path]

1. (start up java zmq application)

Note this uses the old python-to-java harness. An analogous setup is possible with the new native Java application.

0. To evaluate this system, we can run on the January GSR news articles. Do this by first uncommenting the option rss.rsslocationprocessor.outputdir and set it to ./results/.

1. Start the psl harness with these options to listen to local ports:

   python psl_harness.py --pub [anything] --sub tcp://127.0.0.1:1234 --local_port 9999

2. And then in a separate shell, start the test feed script

   python test_publisher.py --pub tcp://127.0.0.1:1234 --json_file ../psl-rss/aux_data/januaryGSRGeocode.json

This should process all the January planned protest events.

3. Evaluate against the old system with

   java -cp ./target/classes:cat classpath.out edu.umd.cs.linqs.embers.ResultsEvaluator