TimePFN: Effective Multivariate Time Series Forecasting with Synthetic Data

Authors: Ege Onur Taga, Muhammed Emrullah Ildiz, Samet Oymak

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate Time PFN on several benchmark datasets and demonstrate that it outperforms the existing state-of-the-art models for MTS forecasting in both zero-shot and few-shot settings. Notably, fine-tuning Time PFN with as few as 500 data points nearly matches full dataset training error, and even 50 data points yield competitive results.
Researcher Affiliation Academia University of Michigan, Ann Arbor EMAIL, EMAIL, EMAIL
Pseudocode Yes Algorithm 1: LMC-Synth Input: Number of variates N, time-series length T, Weibull shape parameter β, Weibull scale parameter λ, (min, max) value of dirichlet concentration parameter (dmin, dmax), minimum number of latent functions m, maximum number of kernel composition in Kernel Synth J Output: Synthetic MTS C with N variates and length T
Open Source Code Yes Code https://github.com/egetaga/Time PFN
Open Datasets Yes We evaluated Time PFN on nine widely-used, real-world benchmark datasets for MTS forecasting. These datasets include ETTh1, ETTh2, ETTm1, ETTm2 (collectively referred to as ETT, representing Electricity Transformer Temperature), Weather, Solar Energy, ECL (Electricity Consuming Load), Exchange, and Traffic. The Solar Energy dataset was introduced by (Lai et al. 2018), while the others were introduced by (Wu et al. 2021).
Dataset Splits Yes We evaluated these models across the entire dataset and at various data budgets, including 50, 100, 500, and 1000 data points. [...] In all MTS evaluations, our primary objective is to forecast a horizon of 96 time steps using an MTS input of 96 time steps.
Hardware Specification Yes Training a single Time PFN of 8 transformer layers takes around 10 hours on L40S GPU.
Software Dependencies No No specific software dependencies with version numbers are mentioned in the main text.
Experiment Setup No We did not perform any hyperparameter tuning on Time PFN, and the same set of hyperparameters was used in all few-shot settings. Details about the model hyperparameters are provided in the appendix.