osirrc / ielab-docker

OSIRRC Docker Image for Elasticsearch

ielab-docker

Harry Scells and Guido Zuccon

This is the docker image for Elasticsearch conforming to the OSIRRC jig for the Open-Source IR Replicability Challenge (OSIRRC) at SIGIR 2019.

Currently supported:

• test collections: robust04, core17, core18, (cw12b should work but is untested)
• hooks: init, index, search

Quick Start

The following jig command can be used to index TREC disks 4/5 for robust04:

python run.py prepare \
  --repo osirrc/ielab-docker \
  --collections robust04=/path/to/disk45=trectext

The following jig command can be used to perform a retrieval run on the collection with the robust04 test collection.

python run.py search \
  --repo osirrc/ielab-docker \
  --output out/ielab \
  --qrels qrels/qrels.robust04.txt \
  --topic topics/topics.robust04.txt \
  --collection robust04

Retrieval Methods

The container currently supports the default Elasticsearch implementattrecion of BM25.

Expected Results

robust04

MAP	BM25
TREC 2004 Robust Track Topics	0.1826

prepare

python3 run.py prepare --repo osirrc/ielab-docker --collections robust04=/path/to/disk45=trectext

search

python3 run.py search --repo osirrc/ielab-docker --qrels qrels/qrels.robust04.txt --topic topics/topics.robust04.txt --collection robust04 --output output/ielab

core17

MAP	BM25
TREC 2017 Common Core Track Topics	0.0831

prepare

python3 run.py prepare --repo osirrc/ielab-docker --collections core17=/path/to/NYTCorpus=nyt

search

python3 run.py search --repo osirrc/ielab-docker --qrels qrels/qrels.core17.txt --topic topics/topics.core17.txt --collection core17 --output output/ielab

core18

MAP	BM25
TREC 2018 Common Core Track Topics	0.1899

prepare

python3 run.py prepare --repo osirrc/ielab-docker --collections core18=/path/to/WashingtonPost.v2=wp

search

python3 run.py search --repo osirrc/ielab-docker --qrels qrels/qrels.core18.txt --topic topics/topics.core18.txt --collection core18 --output output/ielab

cw12b

Need to run experiments

prepare

python3 run.py prepare --repo osirrc/ielab-docker --collections cw12b=/path/to/cw12b=warc

search

python3 run.py search --repo osirrc/ielab-docker --qrels qrels/qrels.web-n.txt --topic topics/topics.web-n.txt --collection cw12b --output output/ielab

Implementation

Dockerfile

The Dockerfile uses a multi-stage build system. First, all development dependencies are installed, code is compiled, and Elasticsearch is downloaded. Next, a runtime image is built and the artifacts from the previous stage are copied (minimising the size of the image and speeding up development time).

init

Since most of the image is configured in the Dockerfile, the init script does not do much (only some permission changes are made).

index

The index Python script reads a JSON object containing the instructions for how and what to index as specified in the jig. This script invokes index.sh.

Since Elasticsearch is built to be run as a service and not as a standalone application, there are some considerations that need to be made. Firstly, all scripts that involve Elasticsearch must execute the script eswait.sh which will start Elasticsearch as a daemon and wait for it to complete starting up. This script is run for both the index and search phases.

Elasticsearch uses a "schema-less" model for indexing documents. This means that in order to index a document, one must first convert it into a format that Elasticsearch can consume (json). The cparser package implements document parsing routines to transform test collection files into bulk index Elasticsearch actions. It can be installed as a stand-alone command-line application irrespective of this repository using the Go toolchain: go get -u github.com/osirrc/ielab-docker/cparser.

search

The search Python script reads a JSON object containing the instructions for how and what to search as specified in the jig. This script invokes search.sh.

Elasticsearch only supports a specific query language, so standard topic file formats cannot directly be used. The tsearcher package implements topic file parsing routines to transform a topic file into a query suitable for Elasticsearch to execute, perform the search, and write a run file. It can be installed as a stand-alone command-line application irrespective of this repository using the Go toolchain: go get -u github.com/osirrc/ielab-docker/tsearcher.