long-forecasting
Train, test, predict, sweep, and visualize the core forecasting models: Linear, DLinear, NLinear, Informer, Transformer, and Autoformer.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- long-forecasting
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Train, test, predict, sweep, and visualize the core forecasting models: Linear, DLinear, NLinear, Informer, Transformer, and Autoformer.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/cli-reference.md4.7 KB
- 参考文档references/model-overview.md3.8 KB
- 参考文档references/troubleshooting.md2.9 KB
- 参考文档references/workflows.md4.0 KB
03 SKILL.md 正文
Long Forecasting
Use this sub-skill when the task is about the root long-term forecasting route for the core model families:
- Linear
- DLinear
- NLinear
- Informer
- Transformer
- Autoformer
It owns the common train/test/predict flow, model selection, look-back window sweeps, embedding sweeps, and DLinear weight visualization.
Start here
- CLI reference
- Model overview
- Workflows
- Troubleshooting
- Forecasting launcher
- DLinear weight plotter
- Tiny smoke check
What this sub-skill owns
- Root
run_longExp.pylong-horizon training, testing, and prediction. - Model choice among the six core forecasting families above.
- Benchmark-style presets for
scripts/EXP-LongForecasting/. - Look-back window studies from
scripts/EXP-LookBackWindow/. - Embedding studies from
scripts/EXP-Embedding/. - DLinear weight plotting from checkpoint files.
What this sub-skill does not own
- Statistical baselines such as Naive, ARIMA, SARIMA, or GBRT.
- FEDformer subrepo workflows.
- Pyraformer workflows or preprocessing helpers.
- Optional TVM internals.
Operating rules
- Treat
run_longExp.pyas the canonical root launcher. Use the bundled wrapper for safer path handling, CPU forcing, and a small amount of validation. - Keep dataset roots explicit. The common layout is a dataset directory such
as
dataset/that contains the benchmark CSVs or a custom CSV with adatecolumn plus feature columns. - Use
--individualfor Linear-family runs when following the benchmark scripts. - Prefer CUDA for Autoformer test or predict runs. The tiny smoke helper can still run a forward check in train mode.
- Use the plotting helper only on Linear-family checkpoints with seasonal and trend weights.
Routing hints
- Route baseline comparisons to
../statistical-baselines/. - Route FEDformer-only tasks to
../fedformer/. - Route Pyraformer tasks and preprocessing to
../pyraformer/. - If a request is ambiguous between this route and one of those sibling workflows, prefer the sibling route instead of growing this one.
The shared CSV layout and dataset root convention are the same as the repo-level root workflow and the FEDformer route. See the workflow reference for the exact file expectations and model-specific argument patterns.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.