This is the implementation for FSTLLM: Spatio-Temporal LLM for Few-Shot Time Series Forecasting.

Requirements

pip install -r requirements.txt

Step 1

Please run the STGNN backbone to obtain numerical prediction tokens

python train.py --config_filename=data/model/para_not.yaml

Step 2

Please fine-tune FSTLLM with

FSTLLM_fine_tune.py

Step 3

Please evaluate FSTLLM with

FSTLLM_inference.py

Step 4

Please print evaluation results with

FSTLLM_evaluation.py