Skip to content

Latest commit

 

History

History
171 lines (140 loc) · 7.77 KB

README.md

File metadata and controls

171 lines (140 loc) · 7.77 KB
Raw

GitHub Actions Docker Hub frontend tags Docker Hub frontend Docker Hub backend tags Docker Hub backend

CAM: The Comparative Argument Machine

The Comparative Argument Machine (CAM) project is developed by the Language Technology Group at the University of Hamburg. As a starting point for a bigger scientific project the current version compares two objects via a large database. The main goal of the CAM project is the comparison based on understanding of natural language and to output natural language sentences as a result.

If you want to learn more about the project or help to develop it, feel free to contact us. A live demo is available online.

Installation

Currently, there are two ways to deploy the CAM project to your own machine: With or without Docker. You will find instructions for both ways here.

Deployment with Docker

Web app

  1. Install Docker and Docker Compose
  2. Clone the CAM repository from GitHub: 
    git clone https://github.com/uhh-lt/cam.git
cd cam
  3. If you're using Docker Toolbox, you need to change the HOSTNAME constants in url-builder.service.ts to match your Docker machine IP (instead of localhost). You can check the Docker machine IP via docker-machine ip.
  4. Start Docker containers: 
    docker-compose up -d

Now CAM is up and running. You should be able to access the frontend app in your browser:
http://localhost:10101
Or directly receive search results from the backend (as JSON objects):
http://localhost:10100/cam?model=default&fs=false&objectA=dog&objectB=cat&aspect1=size&weight1=3&aspect2=food&weight2=1
(The parameters of this URL are described below.)

Elasticsearch

Preferably, Elasticsearch should also get its own Dockerfile or should be build from a Docker image with Docker Compose. To use the suggestions feature, cross-origin resource sharing must be enabled for all origins in the elasticsearch.yml config:

http.cors.enabled: true
http.cors.allow-origin: "*"

With Elasticsearch set up, the suggestions index can be created:

cd cam/src/Backend/create_suggestoins_index/
python create_es-index_from_suggestions.py

Alternatively, extract es-nodes.tar.gz to Elasticsearchs' default nodes location (/var/lib/elasticsearch/).

Deployment without Docker

  1. Clone the CAM repository from GitHub: 
    git clone https://github.com/uhh-lt/cam.git
cd cam

Backend

  1. Go to the backend folder: 
    cd src/Backend
  2. Download Python and install Pipenv.
  3. Install requirements: 
    pipenv install
pipenv run python -m nltk.downloader stopwords
pipenv run python -m nltk.downloader punkt
pipenv run python -m nltk.downloader averaged_perceptron_tagger
  4. Download the following files and place them in src/Backend/data (needed for the InferSent model):
  5. Change default hostnames and search type:
  6. Start the backend API: 
    pipenv run python main.py
    (If the Elasticsearch needs authentication, specify ES_USERNAME and ES_PASSWORD environment variables.)

Now the backend is up and running. You should be able to receive search results from the backend (as JSON objects):
http://localhost:5000/cam?model=default&fs=false&objectA=dog&objectB=cat&aspect1=size&weight1=3&aspect2=food&weight2=1
(The parameters of this URL are described below.)

Frontend

  1. Download Node.js
  2. Enter the frontend working directory: 
    cd src/Frontend/camFrontend
  3. Install Angular dependencies: 
    npm install
  4. Change default hostnames:
    On default, the backend is running on localhost. If you want to change this, maybe because you deployed the project to another server, change all HOSTNAME_ constants, e.g., HOSTNAME_DEFAULT, in url-builder.service.ts.
  5. Start the frontend app: 
    ng serve -o

The frontend app will automatically open in your web browser.

Updating

Updating with Docker

docker-compose down
docker rmi cam-frontend cam-backend
git pull
docker-compose up -d

Updating without Docker

git pull

Start the program like described above.

API

To access the API, URL parameters can be used. The structure is described underneath:

BASE_ADDRESS/cam?model=MODEL&fs=FS&objectA=OBJA&objectB=OBJB&aspect1=ASP1&weight1=WEIGHT1

The base address depends on the backend deployment URL.

  • Replace MODEL with either default, bow or infersent, to select one of the three included models. This parameter is optional.
  • Replace FS with false if you want to do the default search, or with true to enable fast search.
  • Replace OBJA and OBJB with the objects you want to compare, e.g., dog and cat. Both parameters are mandatory.
  • Replace ASP1 and WEIGHT1 with an aspect you want to include and its weight, e. g. price and 5.
  • Add as many aspects/weights as you want, but follow these rules:
    • You must enter both aspect and weight.
    • Aspects and weight parameters must be numbered consecutively. That is, if you include aspect2 and weight2, you have to include aspect1 and weight1 as well. Numbers start at 1. The order of URL parameters does not matter.
    • Aspects/weights are optional. You can skip aspect/weight parameters to compare two objects without any aspects.
    • The frontend limits weights to integers from 1 to 5. If you need equivalent results as from the frontend, use weights from 1 to 5. However, arbitrary integer values may be used. Be careful with negative values or values close to an integer overflow.

Example URL: http://localhost:5000/cam?model=default&fs=false&objectA=dog&objectB=cat&aspect1=size&weight1=3&aspect2=food&weight2=1