CausalTAD

This repository contains the code for the CausalTAD paper.

Environment setup

• Python 3.9+

Core scientific stack

• numpy
• pandas
• scipy
• scikit-learn

Causal discovery

• causallearn

LLM and ML training

• openai
• torch
• transformers
• datasets
• peft
• wandb
• tqdm

Other utilities and baselines

• pyod
• deepod
• modelscope
• gensim
• inflect
• feature_engine
• ucimlrepo

One-command environment setup (requirements.txt)

Create a fresh environment and install all dependencies in one go:

conda create -n causalTAD python=3.10 -y && conda activate causalTAD && pip install -r requirements.txt

Usage overview

1. Run python CausalTAD/graph_gen/main.py to generate the graph.
2. Based on the algorithm type used in the previous step, select the corresponding code under CausalTAD/use_graph to perform graph-based ranking and weight generation.
3. In CausalTAD/AnoLLM, use the results from the previous step to inject causal knowledge into the language model for training and inference.