forecasting-experiments
Construct and adapt TimeMixer forecasting experiment commands.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- forecasting-experiments
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Construct and adapt TimeMixer forecasting experiment commands.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/benchmark-recipes.md7.7 KB
- 参考文档references/troubleshooting.md6.6 KB
- 参考文档references/workflows.md6.4 KB
03 SKILL.md 正文
Forecasting Experiments
Use this sub-skill when the user wants to construct, adapt, audit, or explain TimeMixer commands for forecasting experiments. It covers long-term forecasting and short-term forecasting presets for ETT, electricity/ECL, Traffic, Weather, Solar, PEMS, and M4, plus custom forecasting command construction.
Route here for
- Building a safe
run.pycommand for a known forecast benchmark preset. - Adapting a benchmark command to a custom CSV or changed forecast horizon.
- Explaining training, validation, testing, checkpoints, printed metrics,
test_results/, and M4 forecast CSV outputs. - Reducing epochs, batch size, GPU usage, or worker count before a user-approved expensive run.
- Debugging forecast-specific CLI, data-dimension, GPU, checkpoint, and M4 evaluation failures.
Route elsewhere
- Raw dataset file layout, CSV columns, benchmark download placement, and deterministic data validation: use
data-preparation. - TimeMixer model internals, tensor shapes inside PDM/FMM, decomposition choices, and forward smoke checks: use
model-architecture. - Imputation, anomaly detection, or classification commands: use
universal-tasks.
Bundled operating assets
references/benchmark-recipes.mdlists source-script-derived forecasting presets and marks full benchmark execution as external-data/expensive.references/workflows.mdexplains command-building, training/test-only behavior, checkpoints, result locations, and custom adaptation.references/troubleshooting.mdcovers common forecast failures, including therun.py --helppercent-format bug and CPU fallback.scripts/build_timemixer_command.pyprints arun.pycommand or JSON plan; it never launches training.
Safe default procedure
- Identify the forecast family: long-term tabular/array data, PEMS traffic, Solar, or M4 seasonal short-term forecasting.
- Use the bundled command builder to print, inspect, and adjust the command before any execution.
- Confirm external data availability and user approval before running a full benchmark-scale training command.
- For custom data, ensure
enc_in,dec_in, andc_outmatch the selected feature mode and validated data dimensions. - Treat original benchmark shell scripts as distilled evidence only; do not invoke them from this skill.
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.