Skip to content

Latest commit

 

History

History
96 lines (38 loc) · 3.79 KB

README.md

File metadata and controls

96 lines (38 loc) · 3.79 KB

LEV4REC-deployment

This repository contains the Docker configuration files for the deployment of LEV4REC. The source code and development support is available at https://github.com/MDEGroup/LEV4REC-Tool/ You should conduct the following steps to run this setup:

  1. (Re)Build the docker images using:

    docker-compose build

image

  1. Run the LEV4REC platform using: Please note that the last command can take more than 5 minutes, depending on the platform you are using. It needs to download all the dependencies.

    docker-compose up

image

  1. Access the LEV4Rec web app by using the following address in the web browser: http://localhost:8891/lev4rec/. The LEV4REC user guide is available at https://github.com/MDEGroup/LEV4REC-Tool/blob/main/use_case_artifacts/Documentation/DSL_wiki.md

image

Docker-compose

In case you want to make sure to start from a fresh installation, please execute the following command:

docker system prune -a --volumes

Use cases

We use LEV4REC to design, tune, and deploy two existing recommender systems:

  1. a k-nearest neighbor-based algorithm (named KNN hereafter) that aims to address the scalability problem in personalized recommendations (a predefined form is available at http://localhost:8891/lev4rec/knn) and
  2. AURORA, a feed-forward neural network trained with a curated labeled dataset (a predefined form is available at http://localhost:8891/lev4rec/ml).

We make available the output of KNN in the output_sample. A detailed guide on how to run the generated artifacts is available in each supported presentation layer, i.e., evaluation by Python script, docker container with Jupyter notebook, and a web services by flask.

Replicating the use cases

KNN Algorithm

Using the RS Configuration Form, the user can specify the KNN algorithm as follows:

image info

Afterward, the user can fine-tune the specification by adding additional parameters. The whole KNN settings should be similar to the following:

image info

ML-based approach

The procedure can be followed to AURORA, a classification approach based on a feed-forward neural network.

First, the user can select the proper type of network from the RS Configuration form:

image info

Once all the components have been selected, LEV4REC generates automatically the DSL string that the user can customize:

image info

Acceleo Templates

Once the user has enhanced the system's specification using the web editor, the corresponding implementation can be generated by using the 'Generate' button available in the form. We employ a dedicated Acceleo template generate.mtl file that has been fed directly with the DSL string contained in the form.