AGENT SKILL

evaluation-backtesting

Evaluate GluonTS forecasts, compute aggregate/item metrics, use ev metrics, and backtest predictors without alignment mistakes.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
evaluation-backtesting
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Evaluate GluonTS forecasts, compute aggregate/item metrics, use ev metrics, and backtest predictors without alignment mistakes.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

evaluation-backtesting

Use this sub-skill when the task is to evaluate installed GluonTS predictors or already-produced forecasts, inspect aggregate and per-item metrics, or run a safe backtest on a held-out forecast horizon.

What this sub-skill covers

  • Calling make_evaluation_predictions(...) so each predictor sees only history before the trailing prediction window.
  • Pairing forecast iterators with target iterators in the same order and with compatible period indexes.
  • Computing classic aggregate and item metrics with gluonts.evaluation.Evaluator.
  • Running the one-call backtest_metrics(...) wrapper for an existing Predictor.
  • Using the newer gluonts.ev metric definitions through gluonts.model.evaluate_forecasts(...) or evaluate_model(...).
  • Diagnosing invalid forecasts, missing/invalid targets, frequency/seasonality issues, and target-length mismatches.

Use the bundled references

  • references/api-reference.md — signatures, arguments, metric keys, aggregation behavior, and gluonts.ev metric roles.
  • references/workflows.md — copyable recipes for trailing holdout evaluation, backtest wrappers, item metrics, custom metrics, and gluonts.ev evaluation.
  • references/troubleshooting.md — common failures around alignment, iterator lengths, NaNs, quantiles, seasonality, multiprocessing, and invalid targets.

Safe smoke check

After installing gluonts, run the checkout-independent helper:

python path/to/sub-skills/evaluation-backtesting/scripts/evaluate_synthetic_forecast.py --help
python path/to/sub-skills/evaluation-backtesting/scripts/evaluate_synthetic_forecast.py

The script creates tiny deterministic series, evaluates a local seasonal-naive predictor with make_evaluation_predictions and Evaluator, cross-checks backtest_metrics, optionally writes item metrics as CSV, and prints a concise success summary. It performs no network, plotting, downloads, training, or checkout-relative reads.

Routing hints

  • If the task starts from raw pandas/list/file data, use data-pipelines first.
  • If the task needs transformation chains, samplers, time features, or lags before prediction, use transforms-features first.
  • If the task needs model selection, training, forecast object inspection, or predictor persistence, use forecasting-models before evaluation.
  • If the task asks about SageMaker shell, CLI, or optional extension adapters, use deployment-extensions.

Required safety posture

Keep evaluation examples CPU-only, deterministic, and small unless the user explicitly asks for benchmark-scale work. Treat CUDA as optional acceleration only. Treat MXNet examples as legacy/unverified in this skill scope; do not claim MXNet evaluation workflows are verified unless a separate compatible MXNet environment has been installed and checked.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351