https://github.com/hscells/sysrev-screening-prioritisation

Systematic Review Screening Prioritisation
https://github.com/hscells/sysrev-screening-prioritisation

Last synced: 6 days ago
JSON representation

Systematic Review Screening Prioritisation

• Host: GitHub
• URL: https://github.com/hscells/sysrev-screening-prioritisation
• Owner: hscells
• Created: 2017-04-19T03:02:18.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2017-05-17T05:44:21.000Z (over 7 years ago)
• Last Synced: 2024-11-07T13:57:45.110Z (about 2 months ago)
• Language: Python
• Homepage:
• Size: 215 KB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Systematic Review Screening Prioritisation

_Using learning to rank._

## Setup

 - elasticsearch version 5.3.0

 - latest java version 1.8

 - trec_eval

The experiments use the learning to rank elasticsearch plugin from:

 

https://github.com/o19s/elasticsearch-learning-to-rank

To build the plugin, ensure you have the `JAVA_HOME` environment variable set, and have at least java version 1.8u40. 

On macOS, the to set `JAVA_HOME`, use `export JAVA_HOME=$(/usr/libexec/java_home)`.

Inside the `ltr-query` submodule, run:

 

```bash

./gradlew run#installLtrQueryPlugin

```

This will generate a zip file that can be installed to elasticsearch as a module. To install into elasticsearch, run:

```bash

elasticsearch-plugin install file:///$(pwd)/build/distributions/ltr-query-0.1.1-es5.3.0.zip

```

I have included a convenience scripts called `installPlugin.sh` in this directory that will go ahead and run these 

commands for you.

### Elasticsearch configuration

Because the queries and models we are dealing with are very large, we need to ensure elasticsearch can handle them. My

elasticsearch.yml configuration file has these additional options:

```yaml

script.max_size_in_bytes: 10000000

http.max_content_length: 1gb

```

I also start elasticsearch with more heap space like so:

```bash

ES_JAVA_OPTS="-Xms6g -Xmx6g" ./bin/elasticsearch

```

## Training A Model

To train a model, this project uses RankLib. This project contains a pipeline that will perform feature extraction, 

model training, re-ranking and evaluation. To run the pipeline, use: 

 

```bash

./trainltr

```

The top of the file contains variables that may be configured to change, for instance, elasticsearch settings.

 

Additionally, the pipeline comprises the following python scripts:

 - `ltrfeatures.py`: automatically extract features and produce RankLib training data.

 - `uploadscript.py`: facilitate the uploading of RankLib models into the ltr elasticsearch plugin

 - `search.py`: compare the baseline similarity scores against the learn to rank similarity function

 

## Features

Features are constructed as subclasses of `AbstractFeature` in the `features` module. See the 

[features readme](features/README.md) to explore how to extend and modify the features.