Source code for paper:
ORIS: Online Active Learning Using Reinforcement Learning-based Inclusive Sampling for Robust Streaming Analytics System
Requirements:
- Python 3.7
- PyTorch (stable version from https://pytorch.org/get-started/locally/)
- HuggingFace
- Transformers (https://huggingface.co/docs/transformers/index)
- Datasets (https://huggingface.co/docs/datasets/index)
- Scikit-learn (https://scikit-learn.org/stable/install.html)
- SciPy (https://scipy.org/install/)
- NumPy (https://numpy.org/install/)
- tqdm (https://github.com/tqdm/tqdm)
Model directory contains trained ORIS model.
Data directory contains filtered Twitter and Reddit data adapted from HuggingFace Datasets.
Original data:
- Twitter (https://huggingface.co/datasets/dair-ai/emotion)
- Reddit (https://huggingface.co/datasets/go_emotions)
Steps to replicate the experiments:
- Create a virtual environment
python -m venv venv - Activate virtual environment
source venv/bin/activate - Install and verify all packages mentioned before:
cat requirements.txt | xargs -I {} pip install {} - Make the experiments script executable:
chmod +x run_experiments.sh - Run the script to get the result:
./run_experiments.sh