timeseries-atlas README

The project's README from github.com/dreamers-laboratory/timeseries-atlas. As one markdown file: README.md.

Time Series Atlas

A map of the forecasting architectures that mattered, each in its own directory with a short README and, where it fits in a few hundred lines, a runnable PyTorch implementation.

We built this because the literature is confusing on purpose. Every paper since 2021 claims state of the art on the same eight datasets, the deltas are tiny, and an ICML 2026 position paper (arXiv 2512.22702) showed that normalization choices often matter more than the architecture being advertised. Reading the actual mechanisms side by side is the fastest way to form your own opinion.

The map

Directory System Year The idea in one line
01-linear-baselines DLinear / NLinear 2022 one linear layer, embarrassingly hard to beat
02-attention-era Informer, Autoformer, FEDformer, Pyraformer 2021–22 make attention cheaper: sparsity, decomposition, Fourier bases, pyramids
03-patchtst PatchTST 2023 cut the series into patches and treat each patch as a token
04-itransformer iTransformer 2024 flip the axes: each variable's whole history becomes one token
05-cyclenet CycleNet 2024 learn the weekly/daily cycle explicitly, forecast the residual with a linear head
06-mixers TimeMixer, xPatch, WPMixer 2024–25 decompose (scales, EMA, wavelets), then mix with MLPs
07-ssm-and-xlstm S-Mamba, xLSTMTime, TiRex 2024–25 recurrence returns with linear scaling
08-kan TimeKAN 2025 learnable activation functions on frequency bands
09-diffusion TimeDiff, CCDM, ARMD 2023–25 sample forecast distributions; strong calibration, slow sampling
10-foundation-models Chronos-2, TimesFM 2.5, Moirai 2.0, Toto, Sundial 2024–26 pretrained zero-shot forecasters, the current frontier
11-classical-baselines AutoETS, AutoARIMA, seasonal naive 1970s– the control group; still wins on short seasonal series

Directories 01, 03, 04, and 05 contain self-contained implementations that train on synthetic data in under a minute on a laptop. The rest are guided tours with pointers to the canonical code, because reimplementing them adds bulk without adding understanding. thuml/Time-Series-Library reimplements nearly all of them under one roof if you want to run the originals.

Run something

pip install torch numpy
python common/train.py --model dlinear
python common/train.py --model patchtst
python common/train.py --model itransformer
python common/train.py --model cyclenet

Each command trains on a synthetic multi-seasonal dataset and prints test MSE/MAE next to a seasonal-naive baseline. Drop ETTh1.csv into data/ to run on the standard benchmark instead (download).

What we actually believe

Three things survived contact with our own experiments and the 2025–26 literature:

  1. Linear baselines first. DLinear is 50 lines and beats published transformers on several standard splits (arXiv 2512.10866). If your fancy model can't beat it, stop.
  2. Benchmark wins on ETT are mostly noise. Lookback length, normalization, and channel handling flip rankings. State your protocol before your architecture.
  3. The frontier moved to pretrained models. On GIFT-Eval, Chronos-2 led with a 79.8% win rate as of October 2025, ahead of TiRex and TimesFM 2.5. Zero-shot with covariates is the setting where 2026 claims get adjudicated, on GIFT-Eval and fev-bench. Leaderboards move; re-check before you quote them.

Who made this

Dreamers Inc. We build forecasting and agent systems for clients and kept rewriting these models from papers to understand them; publishing the rewrites seemed more useful than hoarding them. Corrections welcome, issues especially.

If you are working on predictive alpha signals, or RL-based risk assessment and execution systems for live markets, write to [email protected].

Apache-2.0.

timeseries-atlas (dreamers-laboratory/timeseries-atlas)

Searching Database…

133 ·GitHub
Timeseries-atlas
这是什么?一份现代时间序列预测架构的精选图谱,从 DLinear 到 Chronos-2 这类基础模型,每个架构都有自己的目录和一份简短 README,其中一部分还附带可运行的 PyTorch 实现。
这是能导入的库,还是更像参考资料集?它是一套参考资料集:11 个带编号的目录,每个目录涵盖一族预测模型(线性基线、注意力时代的 Transformer、PatchTST、iTransformer、mixer、SSM、KAN、扩散模型、基础模型、经典基线),而不是可安装的包。
怎么运行其中的模型?每个自包含目录(01-linear-baselines、03-patchtst、04-itransformer、05-cyclenet)都在合成的多季节性数据集上训练,并打印相对季节性朴素基线的测试 MSE/MAE;把 ETTh1.csv 放进 data/,就可以改为运行标准 ETT 基准。
除了 PyTorch 还需要什么?PyTorch 是自包含实现的核心要求;这个仓库是纯 Python 项目,README 里没有列出 Docker 配置或其他依赖。
全部 11 个目录都有我能跑的代码吗?不是,只有 01、03、04 和 05 目录包含自包含、可运行的实现;其余目录是导览,指向权威的原始代码,因为重新实现它们“只会增加体量,不会增进理解”。
这份图谱是给谁看的?任何想快速看懂当前预测领域格局的人:正在选模型的研究人员和工程师,或者想并排比较各种机制、而不是轻信排行榜说法的学生。
作者为什么要做这个,而不只是读论文?他们发现时间序列文献“有意让人困惑”:2021 年以来的每一篇论文都声称在同样八个数据集上取得了最先进的结果,差距却微乎其微;而一篇 ICML 2026 立场论文表明,归一化方式的选择往往比架构更重要。把机制放在一起并排阅读,才能拨开这些噪声。
关于线性基线,最重要的结论是什么?DLinear 这个约 50 行的单层线性模型,在多个标准数据划分上仍然胜过已发表的 Transformer,所以本仓库的主张是:在相信更花哨的架构之前,先打败这个基线。
目前在实际预测基准上领先的是什么?预训练基础预测模型是前沿:截至 2025 年 10 月,Chronos-2 以 79.8% 的胜率领跑 GIFT-Eval,领先 TiRex 和 TimesFM 2.5;不过作者提醒,排行榜变化很快,应当重新核实。
有没有能运行所有这些模型完整重实现的地方?有,thuml/Time-Series-Library 在一个屋檐下重新实现了其中几乎全部模型;本仓库对只做导览、没有重新实现的模型,都链接到那里。
What are similar projects?timesfm, Time-Series-Library, Deep-Learning-Papers-Reading-Roadmap, SolarWM, Attention-Residuals, Kronos, Engram, autogluon
Which topics is it in?forecasting, time-series, foundation-models, statistics, papers, pretrained-models, paper, neural-network, academic-research, transformer, data-science, pytorch
Creator
dreamers-laboratory
Size
<0.1 MB
Stars
133
Updated
CommitSep 02
Age
1mo
Commits · 90 Days