- Authors:
- Sébastien Mosser
- Corinne Pulgar
- Vladimir Reinharz
- Contributors:
- Benjamin Benni (Instant systems)
- Guilhem Molines (IBM)
On your local computer, we rely on Python 3.9 and pipenv
to install dependencies
pipenv install
The code uses numpy, a scientific computing library that could be complicated to setup locally. Thus, we provide a Docker version of the code that contains all the dependencies:
TODO
We assume here an up and running environment. The Makefile
provide an easy way to start the scenarios using the right environment. For example, to run the first scenario:
make scenario_1
By convention, each scenario produces a PDF file as output, stored in the output
directory
As the scenarios can consume a lot of memory (the twoi last ones), be sure that you have allocated up to 4Gb of RAM to containers in your local Docker settings.
To run a scenario, use the docker_run.sh
script, indicating which scenario you want to execute and which directory to be used as output.
We release as part of this artefact the following datasets:
dataset/raw
: the original files published by Dalpiaz in 2018;dataset/gilson
: the result of the analysis of these stories by the the team of the University of Canterburry (NZ);dataset/cases
: a JSON format of each stories processed by Visual Narrator and the approach form gilson et al.
If you want to run Visual Narrator on top of these file, we provide a docker image and a shell script to make it easier (in dataset/raw/visual_narrator.sh
).
The resulting figures for each dataset are available in the results
directory.
The Python code used to support each scenario is available in the scenarios
directory. When a scenario requires sopme additional data (e.g., ground truth, or exat information), these data are stored in the scenarios/data
directory.
The metamodel and graph-based implementation of the approach is stored in the backlog
directory.