This is a pre-production deployment of Warehouse. Changes made here affect the production instance of PyPI (
Help us improve Python packaging - Donate today!

TREC Dynamic Domain (DD) evaluation test harness for simulating user interaction with a search engine

Project Description


This is the official “jig” for simulating a user interacting with a TREC DD system during an interactive query session.


After installation (see below), you will want run the harness to generate a run file, you want to use the trec_dd_harness command.

The purpose of this harness is to interact with your TREC DD system by issuing queries to your system, and providing feedback (truth data) for the results produced by your system. While it does this, it keeps track of the results produced by your system in a run file. After generating a run file with this harness, you can score the run using trec_dd_scorer

The harness is run via three commands: start, step, stop. Typically, a system will invoke start, then invoke step multiple times, and then invoke stop. Every invocation must include the -c argument with a path to a valid config.yaml file, as illustrated in example/config.yaml.

For efficiency, the first time you run with a new configuration, the truth data must be loaded into your database using the load command.

(Postgres example, using psql)
CREATE USER trec_dd_user PASSWORD 'some_password';
CREATE DATABASE trec_dd OWNER trec_dd_user

(back at the Unix shell)
(set up config.yaml to point to the database and the truth data file)
trec_dd_harness -c config.yaml load

By default, when you score a system using the harness, all of the topics are applied to the system in an order selected by the harness. You can limit the topic_ids that are used by specifying the topic_ids property in the config.yaml

The harness keeps track of the topic_ids that have not yet been used in building your system’s run file. To reset this state, you must run the init command.

To progress through the topics, your system must execute this double while loop, which is exactly what is implemented in the trec_dd/system/ example:

while 1:
    topic_id <-- `start`
    if topic_id is None: break
    while 1:
        results <-- run your system
        feedback <-- `step(results)`
        if feedback is None or len(feedback) < batch_size:
            your system processes the feedback

Each of the five commands returns a JSON dictionary which your system can read using a JSON library. After a step command, the response looks like:

     "topic_id": "DD15-1"
     "confidence": 0.987,
     "on_topic": 1,
     "stream_id": "1335424206-b5476b1b8bf25b179bcf92cfda23d975",
     "subtopics": [
             "passage_text": "this is a passage of relevant text from the document 'stream_id', relevant to the 'subtopic_id' below with the 'rating' below",
             "rating": 3,
             "subtopic_id": "DD15-1.4",
             "subtopic_name": "a label for this subtopic"
 { ... }

The harness always provides feedback for every result, even if the feedback is that the system has no truth data for that result. Note that your use of the harness must call stop in the next iteration after any step in which you submit fewer than batch_size results. If you fail to do this, the harness will exit.

See trec_dd/system/ for an example of using the harness from python.

The harness outputs a runfile, whose path is set in the configuration file.

To score a runfile (see “Scoring the System”):

trec_dd_scorer -c config.yaml run_file_in.txt run_file_scored.json > pretty_table.txt 2> log.txt &

run_file_in.txt is the run file output by the harness. The scorer outputs a scored run file in run_file_scored.json, and scores to stdout.

This repository also provides a baseline system that randomizes subtopic ordering (see “Example TREC DD Systems”). In particular this baseline system shows how to hook an a system up to the jig in python. Hooking a system up to the jig via the command line is further documented below.

trec_dd_random_system -c config.yaml &> log.txt &

The scores for this baseline system using the TREC DD truth data are:

Score Metric
0.438 average_err_arithmetic
0.298 average_err_harmonic
0.125 modified_precision_at_recall
0.981 precision_at_recall
0.075 reciprocal_rank_at_recall


The recommended way to install and use the scorer is with python virtualenv, which is a standard tool on all widely used platforms. For example on Ubuntu:

apt-get install python-virtualenv
virtualenv vpy

or on CentOS:

yum install python-virtualenv
virtualenv vpy

or on MacOS X

brew install pyenv-virtualenv
pyenv-virtualenv vpy

or on Windows.

You will also need a database. We recommend postgres or mysql. You can install this on your system using standard tools. The connection information must be written into the config.yaml file referenced in the commands above. See config.yaml for an example.

Once you have a virtualenv, the following commands will install the trec_dd scorer. You should choose whether you are using mysql or postgres and specify that as a pip extras declaration in square brackets as follows:

. vpy/bin/activate
pip install trec_dd[mysql]

or to use postgres:

. vpy/bin/activate
pip install trec_dd[postgres]

That will create the shell entry points for running the two commands illustrated at the top of this file.

Simulation Harness

If you wish to evaluate a TREC DD system, you must run it against the TREC DD simulation harness. A system interacting with the simulation harness will produce a “runfile” that summarizes the simulation session. The “runfile”, for each of the system’s response, encodes information such as (1) “was the system’s response on topic?” (2) “what subtopics were contained within the system’s response?” and (3) “how relevant was the system’s response?”. Please see the specification for a “runfile” for more information.

A TREC DD system interacts with the simulation harness by invoking commands at the command line. Systems written in python may use the HarnessAmbassadorCLI to facilitate this communication. The HarnessAmbassadorCLI is also useful documentation for how one should interact with the harness via the command line.

Once you have a “runfile”, you may then score your run. Please see the section “Gathering Scores” for more information.

Example TREC DD Systems

The directory trec_dd/system holds example TREC DD systems to demonstrate interaction with the simulation harness using a TREC DD system. Right now, the only example system is

Executing the Random System


To run the example systems, you must have a truth data XML file. Make sure your database is set up as per your config.yaml, and load the truth data into the database:

trec_dd_harness -c config.yaml load

Running the System

You can run the random system in the simulation harness by calling

trec_dd_random_system -c config.yaml >log.txt 2>&1

After this command executes, you should find the resulting system runfile at the path you specified in the configuration. The runfile summarizes the responses the random system gave to the harness, as well as the harness’s thoughts on those responses. This runfile captures everything one needs to know in order to give a system a score.

Scoring the System

To score your runfile, you may use the trec_dd/scorer/ script.

trec_dd_scorer -c config.yaml run_file_in.txt run_file_scored.json > pretty_table.txt 2> log.txt &

Please see the section titled “Gathering Scores” for more information on the scoring subsystem.

Gathering Scores


You must have a runfile generated for your system if you wish to score it. You must also have access to the truth data used by the harness when generating the runfile.

Running the Scorer

There are two scoring scripts used to compute evaluation scores. bin/ is used to compute Cube Test results. To run it:

bin/ cubetest-qrels runfile cutoff

where runfile is the output runfile from the jig, cubetest-qrels is a specially-formatted version of the truth data (and available from the same place), and cutoff is the number of iterations for running the Cube Test.

trec_dd/scorer/ is used to generate other evaluation scores including u-ERR. To run it:

trec_dd_scorer -c config.yaml run_file_in.txt run_file_scored.json > pretty_table.txt 2> log.txt &

This will go through your runfile and run each configured TREC DD scorer. run_file_in.txt is the runfile produced as output by the harness. The scorer outputs an annotated version of your run in run_file_scored.json, and the scores to stdout.

If you wish to run specific scorers, rather than all of them, please see the ‘–scorer’ option on the trec_dd_scorer command. The scorers specified after the –scorer option must be the names of scorers known to the system. These are exactly the following:

  • reciprocal_rank_at_recall
  • precision_at_recall
  • modified_precision_at_recall
  • average_err_arithmetic
  • average_err_harmonic

Description of Scorers

  • The Cube Test is a search effectiveness measurement that measures the speed of gaining relevant information (could be documents or passages) in a dynamic search process. It measures the amount of relevant information a search system could gather for the entire search process with multiple runs of retrieval. A higher Cube Test score means a better DD system, which ranks relevant information (documents and/or passages) for a complex search topic as much as possible and as early as possible.
  • reciprocal_rank_at_recall calculates the reciprocal of the rank by which every subtopic for a topic is accounted for.
  • precision_at_recall calculates the precision of all results up to the point where every subtopic for a topic is accounted for.
  • average_err_arithmetic calculates the expected reciprocal rank for each subtopic, and then average the scores accross subtopics using an arithmetic average. It uses a graded relevance for computing stopping probabilities.
  • average_err_harmonic calculates the expected reciprocal rank for each subtopic, and then averages the scores accross subtopics using an arithmetic average. It uses graded relevance for computing stopping probabilities.
Release History

Release History

History Node


History Node


History Node


History Node


This version
History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


Download Files

Download Files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

File Name & Checksum SHA256 Checksum Help Version File Type Upload Date
trec_dd-0.3.6-py2.7.egg (57.3 kB) Copy SHA256 Checksum SHA256 2.7 Egg Jul 22, 2015
trec_dd-0.3.6.tar.gz (27.6 kB) Copy SHA256 Checksum SHA256 Source Jul 22, 2015

Supported By

WebFaction WebFaction Technical Writing Elastic Elastic Search Pingdom Pingdom Monitoring Dyn Dyn DNS Sentry Sentry Error Logging CloudAMQP CloudAMQP RabbitMQ Heroku Heroku PaaS Kabu Creative Kabu Creative UX & Design Fastly Fastly CDN DigiCert DigiCert EV Certificate Rackspace Rackspace Cloud Servers DreamHost DreamHost Log Hosting