Alvis Docker

Dockerizing Alvis and its components

Use the docker image

A docker image for the alvisnlp engine in command line is present into Docker Hub

/!\ Prerequisites are docker installed (>= 1.13.1) and 4Go of free disk space

Here are some examples of command you may use to run the docker image, some use an alvis plan. An example is also provided to deal with external parameters to the alvis plan.

1. display the alvis help

docker run mandiayba/alvisengine:1.0.0 
       alvisnlp -help

2. list the modules supported by alvis

docker run mandiayba/alvisengine 
       alvisnlp -supportedModules

3. display the doc of the module named SimpleProjector

docker run mandiayba/alvisengine 
       alvisnlp -moduleDoc GeniaTagger

4. run an alvis workflow that trains a ML model for binary relation extraction from a text corpus

docker run -i --rm -v $PWD/workdir:/opt/alvisnlp/data  -a stderr mandiayba/alvisengine:1.0.0 
       alvisnlp /opt/alvisnlp/data/plans/train.plan

5. run an alvis workflow that classifies binary relations from a text corpus

docker run -i --rm -v $PWD/workdir:/opt/alvisnlp/data  -a stderr mandiayba/alvisengine:1.0.0 
           alvisnlp /opt/alvisnlp/data/plans/predict.plan

6. run the train workflow with the main parameters passed to the workflow

docker run -i --rm -v $PWD/workdir:/opt/alvisnlp/data -a stderr mandiayba/alvisengine:1.0.0 
           alvisnlp
           -param train textDir /opt/alvisnlp/data/corpus/train 
           -param dev textDir /opt/alvisnlp/data/corpus/dev 
           -param test textDir /opt/alvisnlp/data/corpus/test 
           -param TEESTrain model /opt/alvisnlp/data/models
           /opt/alvisnlp/data/plans/train.plan

The train and classify workflows (called plans into alvis) are based on GeniaTagger and TEES tools integrated to AlvisNLP. The corpus used is Bacteria Biotope 2016. The binary relation here is named "Lives_in" ant it expresses the fact that some bacteries live in some habitats.

The docker image creation

• automatically clone and install alvisnlp with Dockerfile

• then, install the following external programs in an interactive way

  • BioLG 1.1.12
  • CCGParser 1.00
  • CCGPosTagger 1.00
  • EnjuParser 2.4.2
  • EnjuParser2 2.4.2
  • GeniaTagger 3.0.1^*
  • StanfordNER 2014-06-16^*
  • Species 1.0
  • TEESClassify 2.2.1
  • TEESTrain 2.2.1
  • TreeTagger 3.2^*
  • WapitiLabel 1.5.0
  • WapitiTrain 1.5.0
  • YateaExtractor 0.5^*

^* Not the latest version, we might want to test with the latest version.