Towards Editing Time Series
Authors: Baoyu Jing, Shuqi Gu, Tianyu Chen, Zhiyu Yang, Dongsheng Li, Jingrui He, Kan Ren
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the efficacy of TEdit for editing specified attributes upon the existing time series data. |
| Researcher Affiliation | Collaboration | Baoyu Jing 2, Shuqi Gu 1, Tianyu Chen1, Zhiyu Yang1, Dongsheng Li3, Jingrui He2, Kan Ren 1 1Shanghai Tech University, 2University of Illinois at Urbana-Champaign, 3Microsoft Research |
| Pseudocode | Yes | Algorithm 1 Bootstrap Learning for a Batch |
| Open Source Code | Yes | The project page is at https://seqml.github.io/tse. |
| Open Datasets | Yes | We collect three datasets for TSE, including one synthetic dataset and two real-world datasets. For the Synthetic data... The Air Quality [8] dataset... The Motor Imagery [1] dataset... |
| Dataset Splits | Yes | We split the 32 combinations into train, validation, and test sets by 24:4:4. ... For pretraining dataset, there are totally 3650 time series samples, in which we randomly pick 2825 time series samples as train, 353 time series samples as validation, and 472 time series samples as the test. |
| Hardware Specification | Yes | All our experiments were conducted on a single Nvidia-A100 GPU. |
| Software Dependencies | No | The paper mentions using 'Adam optimizer [20]' but does not provide specific version numbers for software dependencies such as programming languages, deep learning frameworks (e.g., PyTorch, TensorFlow), or other relevant libraries. |
| Experiment Setup | Yes | For all experiments, we set the number of diffusion steps as T = 50, embedding size for attributes and time series as 64, and use Adam optimizer [20] to train the model. For pretraining, we set (batch size, learning rate) as (256, 1e-3); for finetuning, we set them as (64,1e-7) for Synethtic and (32,1e-7) for Air Quality and Motor Imagery. ... (R, Lp, ψ) = (3, 2, 0.5) for Synthetic, (3, 2, 0.5) for Air Quality, (3, 3, 0.5) for Motor Imagery. |