Research map / Time series / Causal and generative time series
TimeRecipe: A Time-Series Forecasting Recipe via Benchmarking Module Level Effectiveness
Zhiyuan Zhao, Juntong Ni, Shangqing Xu, Haoxin Liu, Wei Jin, B. Aditya Prakash
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| exhaustive module-level design space search instead of fixed existing models | best existing model · 102 forecasting scenarios | MSE: 5.4% error reduction; better in 92 out of 102 scenarios | Improved |
| pick the top searched configuration instead of iTransformer | iTransformer · PEMS03 horizon 12 | MSE: 0.739 -> 0.714 | Improved |
| search design space rather than best existing model, ranking | best existing model · 102 forecasting scenarios | ranking position: 13.66 ranking positions | Improved |
| use LightGBM model selection top-3 instead of existing best PatchTST | PatchTST · Social_12_S | MSE: 0.0994 -> 0.0897 | Improved |
| use LightGBM model selection top-3 instead of existing best PatchTST | PatchTST · ETTh1_24_M | MSE: 0.2988 -> 0.2963 | Improved |
| use LightGBM model selection top-3 instead of existing best DLinear | DLinear · Environment_48_S | MSE: 0.2950 -> 0.2920 | Improved |
| use LightGBM model selection top-3 instead of existing best DLinear | DLinear · Security_12_S | MSE: 75.2914 -> 74.9425 | Improved |
| use a poorly selected configuration (third pick) instead of PatchTST | PatchTST · Social_12_S | MSE: 0.0994 -> 0.1310 | Hurt |
| use instance normalization on PEMS datasets | forecasting models · PEMS | forecasting error | Hurt |
| use instance normalization when shifts are large or seasonality is low | forecasting models · multivariate forecasting | forecasting error | Improved |
| use RNN architectures when HL-ratio is low | forecasting models · multivariate forecasting | forecasting error | Improved |
| use patch embedding with MLP or RNN feed-forward on ETT | forecasting models · ETT multivariate | forecasting error | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- iTransformer — iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (ICLR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Time-MoE: Billion-Scale Time Series Foundation Models with Mixture of Experts (ICLR 2025 · similarity 0.89)
- fev-bench: A Realistic Benchmark for Time Series Forecasting (NeurIPS 2026 · similarity 0.89)
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting (NeurIPS 2025 · similarity 0.89)
- TimeBase: The Power of Minimalism in Efficient Long-term Time Series Forecasting (ICML 2025 · similarity 0.89)
- Timer-XL: Long-Context Transformers for Unified Time Series Forecasting (ICLR 2025 · similarity 0.88)
- A Closer Look at Transformers for Time Series Forecasting: Understanding Why They Work and Where They Struggle (ICML 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
