The implementation of ADAPT-Z (Automatic Delta Adjustment via Persistent Tracking in Z-space)

September 25, 2025 ยท View on GitHub

This code is the implementation of ADAPT-Z (Automatic Delta Adjustment via Persistent Tracking in Z-space) algorithm, along with supplementary experiments from the paper and implementations of baselines.

This codebase is primarily based on DSOF (Fast and Slow Streams for Online Time Series Forecasting Without Information Leakage, ICLR 2025; paper: https://openreview.net/pdf?id=I0n3EyogMi, repository: https://github.com/yyalau/iclr2025_dsof)

1) Train basemodel

run train_basemodel.py to obtain base model

2) Online deployment

run adapt-z.py to conduct online prediction

Baselines and supplementary experiments

run run_dsof.py to conduct experiments using DSOF

run ADCSD.py to conduct experiments using ADCSD

run parameter_finetune to conduct experiments using OGD

run z_finetune to conduct experiments using f-OGD

run train_FAN.py/train_DishTS.py and ada_online2 fan.py/ada_online dishts.py to conduct experiments using FAN/DishTS and FAN+/DishYS+

Pseudo code

sample-wise batch-wise