First, make sure that conda and npm are installed.
- Install Anaconda from https://docs.anaconda.com/anaconda/install/index.html
- Install npm from https://docs.npmjs.com/
Whenever terminal commands are used, I use ~ to refer to the repo root, not home.
Don't just blindly copy the commands since the pathing won't work.
This is only relevant if using AWS to host the experiment. Other cloud services may have similar requirements.
Go to Security Groups under Network & Security.
Whatever security group is being used for the server, add ports 3000 and 5001 to inbound rules.
3000 is used by the React server.
5001 is used by the Python flask server.
Create and initialize the Python virtual environment using
conda create -n explearn python=3.10.8
conda activate explearn
cd ~/code/python/
pip install -r requirements.txt
conda develop .
To allow Jupyter Notebook to use the venv as its kernel, run
conda activate explearn
python -m ipykernel install --user --name=explearn
Install the react server using
cd ~/code/experiment
npm install
export NODE_OPTIONS=--openssl-legacy-provider
The config file can be found at ~/code/config.yaml.
To run the program locally, set host_url: localhost.
To run it on a server, set host_url: [[YOUR SERVER PUBLIC IP]].
The experiment requires running two separate programs. You can run a separate instance of terminal manually for each program. Alternatively, you can use screen or tmux.
Run the Python flask server using
cd ~/code/python/scripts/
python rest_api_endpoint.py
Run the React server using
cd ~/code/experiment/
npm start
Both of these servers are run in development mode, which is recommended when developing or testing the program.
Unless you plan to run the experiment at a large scale, the development mode works fine in practice, even when deployed for live data collection.
If you care about performance, you can always build the program first.
Important: In order to maintain a single source of truth for the configs, the Python flask server copies the contents of ~/code/config.yaml to
~/code/experiment/public/data/config.json. This means that if you change config.yaml, you must run the Python server before running the React server.
The front-end sends a POST request every minute and after every milestone (e.g. after a set of trials) to the Python server,
which then dumps the entire contents as a JSON file to ~/data/raw.
There has been issues of some data loss in Experiment 1, where about 4% of all participants (including pilots) were lost. This only happened during the full data collection of ~150 participants, which leads me to believe there are some traffic-related issues, though I'm not sure if this is due to buggy code, OS IO throughput, network issues, user-end issues, etc. In any case, it is advised to run the experiment in smaller batches if possible to avoid these issues.