timemixer
Use TimeMixer to inspect the model API, build forecasting or non-forecast commands, validate dataset layouts, and debug shape or task-specific workflow issues.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- timemixer
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use TimeMixer to inspect the model API, build forecasting or non-forecast commands, validate dataset layouts, and debug shape or task-specific workflow issues.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/cli-reference.md4.9 KB
- 参考文档references/repo-provenance.md1.4 KB
- 参考文档references/repo-routing-metadata.json495 B
- 参考文档references/troubleshooting.md4.4 KB
03 SKILL.md 正文
TimeMixer
Use this repo skill for the TimeMixer time-series repository: model inspection, forecasting command construction, dataset layout validation, and the imputation/anomaly/classification branches.
Start here
-
Install the runtime dependencies from
requirements.txtin a Python environment that can import the source checkout.- The repository notes a Python 3.8 sktime compatibility pin (
sktime==0.29.1) if you are reproducing the historical runtime exactly. - The repository is source-first rather than a packaged distribution, so the bundled helpers add the checkout to
sys.pathinstead of depending on a wheel install.
- The repository notes a Python 3.8 sktime compatibility pin (
-
Keep helper invocation and source execution roots distinct. The helper files live under the skill directory, while
run.pyand the source packages live at the checkout root. From any shell, use explicit absolute paths:CHECKOUT_ROOT=/path/to/TimeMixer-checkout SKILL_ROOT="$CHECKOUT_ROOT/skills/disco/timemixer" cd "$CHECKOUT_ROOT" python "$SKILL_ROOT/scripts/check_timemixer_environment.py" --repo-root "$CHECKOUT_ROOT"Generated
run.pycommands must likewise be executed with cwdCHECKOUT_ROOT;--repo-root .is only correct when the current directory is the source checkout, never when it is the skill directory. -
Then route the task to the smallest matching sub-skill.
Route map
| If the user asks for... | Use |
|---|---|
| Model construction, forward shapes, decomposition choices, downsampling, or classification shape debugging | sub-skills/model-architecture/SKILL.md |
| Forecasting benchmark commands, custom forecast adaptation, checkpoints, or M4/PEMS/ECL/Weather/Solar recipes | sub-skills/forecasting-experiments/SKILL.md |
| Dataset file layouts, CSV/NPY/TS validation, time-feature conventions, or benchmark data placement | sub-skills/data-preparation/SKILL.md |
| Imputation, anomaly detection, or classification workflows and task-specific command construction | sub-skills/universal-tasks/SKILL.md |
Quick reference files
- Read
references/cli-reference.mdwhen you need therun.pyargument groups or task-to-flag mapping. - Read
references/troubleshooting.mdfor install/import failures, therun.py --helppercent-format bug, or other cross-cutting setup problems. - Read
references/repo-provenance.mdwhen you need to check whether this skill matches a different TimeMixer checkout or before refreshing the skill. - Read
references/repo-routing-metadata.jsononly as managed-router metadata; it is not a user guide.
Safe checks
- Use
scripts/check_timemixer_environment.pyto confirm that the checkout can be imported from a clean source-root path. - Use
sub-skills/model-architecture/scripts/smoke_timemixer_forward.pywhen you need a tiny CPU-only forward shape smoke. - Use the forecasting or universal-task command builders before launching any long experiment.
What this skill covers
TimeMixer.Modelarchitecture and forward behavior.- Long-term and short-term forecast command construction.
- Custom dataset and benchmark data validation.
- Imputation, anomaly detection, and classification task branches.
- Common
run.py, dataset, and tensor-shape failure modes.
What it does not do
- It does not reproduce benchmark-scale training runs.
- It does not download datasets for you.
- It does not depend on the original source checkout being opened through the skill files themselves; the bundled references and scripts summarize the needed behavior.
Notes for future agents
run.py --helpis unreliable because one help string contains an unescaped percent sign.- Prefer the bundled command builders for safe command generation.
- Keep the root skill as a router; deeper API, workflow, and troubleshooting detail lives in the sub-skills.
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.