Financial disclosures

Requirements:

• Python>=3.4
• miniconda/Anaconda

How to get started

It is recommended to use miniconda (or conda) as opposed to Python only default pip as library dependecy manager and to use the local environment of this repository for reproducibility.

1. Install miniconda if conda is not yet installed on the system through Anaconda
2. Setup local environment: conda env create --prefix ./envs -f environment.yml
3. Activate the project environment by running conda activate ./envs
4. Start JupyterLab server: jupyter lab
5. To deactivate the environment, run conda deactivate

How to run the Labeler

To run the labeler locally, additionally the following needs to be done (need to restart the JupyterLab server afterwards):

6. Make sure that NodeJS is installed with: conda install -c conda-forge nodejs
7. Enable Jupyter Widgets: jupyter labextension install @jupyter-widgets/jupyterlab-manager
8. Load the spacy language model: python -m spacy download en_core_web_md

Data

Raw annual reports, sustainability reports and if available 20-F's of the Euro STOXX 50 for the years 1999-2019 can be found here.

Docker for containerization

• Install Docker Desktop Install Docker

• The docker-compose.yml in the root directory contains the container definitions and sets up networking

• The Dockerfile in the ./data directory contains the container config for the pdf mining tasks

• Start tmux: tmux or tmux new -s myname

• Connect to container: docker ps and docker exec -it pdf-mining bash

• Name the session accordingly by first send the prefix Ctrl + b and then $

• Start long running process and leave/detach the session with prefix Ctrl + b and then d

• Later, when you want to attach to the session: tmux list-sessions tmux attach-session -t 0

Cheat Sheet:

• Update environment file if a package was added: conda env export -f environment.yml --no-builds or conda env export --no-builds > environment.yml. NOTE: This will overwrite everything and adds unecessary packages, making cross-platform compatibility difficult. So better to update manually...

Additional Resources:

• Conda Cheat Sheet