AGENT SKILL

datasets-diagnostics

Inspect pmdarima's built-in time-series datasets, validate target and seasonal inputs, estimate differencing, and run numeric or headless diagnostics.

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

01 前言 Front Matter

name
datasets-diagnostics
license
MIT
metadata
{ "disco-role": "operating" }
description
Inspect pmdarima's built-in time-series datasets, validate target and seasonal inputs, estimate differencing, and run numeric or headless diagnostics.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

Datasets and diagnostics

Use this route before fitting or selecting a forecasting model when the target series, observation calendar, missing-value policy, differencing need, or serial-dependence evidence is not explicit. Read the three linked references before making a material decision:

  • API reference — loaders, validators, array helpers, stationarity, seasonality, ACF/PACF, and plotting contracts.
  • Workflows — bounded offline procedures and the handoff record.
  • Troubleshooting — build/import, data, frequency, short-series, statistical, and headless-plot failures.

Run the bundled deterministic checker for a small local smoke test. It does not fetch data, fit an ARIMA model, or create plots unless --plot is explicitly requested.

Operating route

  1. Identify the source and calendar. Choose a local loader whose documented time basis matches the task. Record the loader, as_series option, Python type, shape, length, dtype, index/date semantics, and missing count. A one-dimensional array does not carry frequency metadata; choose m from domain/calendar evidence, never from the sample count or a plot. For load_msft(), inspect the seven-column DataFrame, parse/sort Date only when needed, and select one numeric target. Do not pass the full DataFrame to univariate helpers. Avoid load_gasoline() in offline work: it can use a network URL/cache and its documented weekly period is fractional.
  2. Validate the target. Normalize a univariate target with check_endog. Use force_all_finite=True before stationarity or seasonal tests, ACF/PACF, and decomposition unless the task has an explicit missing data policy. Use check_exog only for a two-dimensional exogenous matrix; route feature construction to preprocessing.
  3. Estimate differencing, then report limits. With a justified integer m > 1, call bounded ndiffs for non-seasonal d and nsdiffs for seasonal D. Record test names, alpha, bounds, warnings, and output lengths from any explicit diff calls. Compare tests only with their hypotheses and settings recorded. These are bounded recommendations, not final ARIMA order selection or proof that a model will forecast well.
  4. Decompose only compatible data. Use decompose with exactly type_="additive" or type_="multiplicative", an integer m > 1, at least two complete cycles, and a scale compatible with the chosen type. Inspect endpoint NaNs in trend and random; check reconstruction only on the finite interior. Do not feed padded components to another diagnostic without a missing-value policy.
  5. Prefer numeric dependence summaries. Use acf and pacf with explicit conservative lags and methods. PACF has stricter lag limits than ACF. Treat these as descriptive evidence, not automatic p/q order selection.
  6. Make plots optional and headless. If qualitative inspection is needed, set a non-interactive Matplotlib backend before importing pmdarima, pass show=False, and close all figures. tsdisplay requires lag_max < len(y). A missing Matplotlib installation blocks only the requested visualization path, not numeric diagnostics.
  7. Handoff, do not overreach. Send the normalized target, calendar, missing policy, d/D recommendations, transformed lengths, and diagnostic evidence to forecasting. Send target transforms, Fourier, or date features to preprocessing. Send leakage-safe holdouts and temporal cross-validation to model-selection. Persistence and update belong to the sibling persistence route. This sub-skill does not fit the primary ARIMA model, choose final orders, compare folds, or persist models.

Minimum handoff record

loader/as_series, type/shape/length/dtype, index or date evidence,
missing count and policy, calendar evidence and m, ndiffs/nsdiffs tests and
bounds, d/D and differenced lengths, decomposition type and endpoint NaNs,
ACF/PACF methods/lags/key values, plotting/backend status, warnings/errors,
package/runtime version and import/build status, unresolved assumptions

The evidence anchor is pmdarima v2.1.1 commit 4c2dfccb28f64d2c00a5e10b59c1d1a3e16576a9. Runtime files resolve imports from the installed, successfully built pmdarima distribution and do not require a source checkout.

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