Docker Endpoint for DRILL

Part of RAKI D6.1

1. Creating an endpoint for DRILL manually

Here, we provide each command to create an endpoint to use DRILL

# (1) clone the repo & unzip necessary files.
git clone https://github.com/dice-group/RAKI-Drill-Endpoint && cd RAKI-Drill-Endpoint
unzip embeddings.zip && unzip LPs.zip && unzip pre_trained_agents.zip
# (2) Clone the repository and create a python virtual enviroment via anaconda
git clone https://github.com/dice-group/DRILL_RAKI && conda create -n drill_env python=3.9 && conda activate drill_env
# (3) Install requirements
cd DRILL_RAKI && unzip KGs.zip && wget --no-check-certificate --content-disposition https://github.com/dice-group/Ontolearn/archive/refs/tags/0.5.1.zip
unzip Ontolearn-0.5.1.zip && cd Ontolearn-0.5.1 && pip install -e . && cd ..
# For the Endpoint only
cd .. && pip install flask==2.1.2
# Test the installation. No error should occur.
python -c "import ontolearn"
# (4) Execute python script to create flask based endpoint.
python DRILL_RAKI/flask_end_point.py --path_knowledge_base 'DRILL_RAKI/KGs/Biopax/biopax.owl' --path_knowledge_base_embeddings 'embeddings/ConEx_Biopax/ConEx_entity_embeddings.csv' --pretrained_drill_avg_path 'pre_trained_agents/Biopax/DrillHeuristic_averaging/DrillHeuristic_averaging.pth'
#...
# Running on http://0.0.0.0:9080/ # Copy this address

How to use the endpoint

pick one of the example learning problems and submit it to the system: (requires jq)

# (1) Open a new terminal (Ctrl+Alt+T on ubuntu) to verify the endpoint.
curl http://0.0.0.0:9080/status # => {"status":"ready"} # If you see this  all went well :)
# (2) Use an example learning problem
jq '
     .problems
       ."((pathwayStep ⊓ (∀INTERACTION-TYPE.Thing)) ⊔ (sequenceInterval ⊓ (∀ID-VERSION.Thing)))"
     | {
        "positives": .positive_examples,
        "negatives": .negative_examples
       }' LPs/Biopax/lp.json \
| curl -d@- http://0.0.0.0:9080/concept_learning

2. Creating an endpoint for DRILL via Docker

git clone https://github.com/dice-group/RAKI-Drill-Endpoint && cd RAKI-Drill-Endpoint
unzip LPs # unzip learning problems file to use it later on
sudo docker build -t drill:latest "."
# Successfully tagged drill:latest # if you see **done**, all went well
sudo docker images # to see installed image

Run the docker image (timeout in seconds).

sudo docker run \
-e KG=KGs/Biopax/biopax.owl \
-e EMBEDDINGS=embeddings/ConEx_Biopax/ConEx_entity_embeddings.csv \
-e PRE_TRAINED_AGENT=pre_trained_agents/Biopax/DrillHeuristic_averaging/DrillHeuristic_averaging.pth \
-e TIMEOUT=15 \
drill:latest
# expected to see
# Running on http://172.17.0.2:9080/

How to use the endpoint

pick one of the example learning problems and submit it to the system: (requires jq)

# (1) Open a new terminal (Ctrl+Alt+T on ubuntu) to verify the endpoint.
curl http://172.17.0.2:9080/status
{"status":"ready"} # If you see this  all went well :)
# (2) Use an example learning problem
jq '
   .problems
     ."((pathwayStep ⊓ (∀INTERACTION-TYPE.Thing)) ⊔ (sequenceInterval ⊓ (∀ID-VERSION.Thing)))"
   | {
      "positives": .positive_examples,
      "negatives": .negative_examples
     }' LPs/Biopax/lp.json \
| curl -d@- http://172.17.0.2:9080/concept_learning

response: (OWL rdf/xml)

<?xml version="1.0"?>
<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
         xmlns:xsd="http://www.w3.org/2001/XMLSchema#"
         xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"
         xmlns:owl="http://www.w3.org/2002/07/owl#"
         xml:base="https://dice-research.org/predictions/1619526593.1690164"
         xmlns="https://dice-research.org/predictions/1619526593.1690164#">

<owl:Ontology rdf:about="https://dice-research.org/predictions/1619526593.1690164">
  <owl:imports rdf:resource="file:///OntoPy/KGs/Biopax/biopax.owl"/>
</owl:Ontology>

<owl:Class rdf:about="#Pred_0">
  <owl:equivalentClass rdf:resource="http://www.biopax.org/examples/glycolysis#pathwayStep"/>
  <rdfs:label rdf:datatype="http://www.w3.org/2001/XMLSchema#string">pathwayStep</rdfs:label>
</owl:Class>


</rdf:RDF>

Congrats!

Contact

For any questions, please contact: [email protected] / [email protected]