This is a project from Data Science Engineering team at [24]7.ai for creating a service for deploying a combination of Spark models using Mleap technology and generic rules (in Javascript or Python) in production. The library is built on top of akka-HTTP. Currently the server supports two operations: returning the predicted class with maximum probability and top N classes and their probabilities
- JDK 8. OpenJDK 8 can be used without any issues.
- Scala: Version 2.11.12
- The binaries should be in the path
- Maven: Version 3.3.9 or above
- The binaries should be in the path
- [Optional, for development] IntelliJ IDEA Community Edition
Following are the links for docker installation.
cd Tucana
mvn clean package -DskipTests
-
cd ../Docker
-
To run the complete test suite:
./run-docker.sh "test"
The following commands would drop you into a docker bash shell with all ModelServer dependencies ready. There you can run other usual commands.
cd ../docker
./run-docker.sh
It is also possible to mount a host folder with your data into the docker container, and use those files and folders. For example, to load /home/abc
from host to docker, use the following command:
./run-docker.sh -m /home/abc
Once you are inside the container, this folder will be available at /tucana/project
inside the container.