ModernTSF

Add a new model to ModernTSF by scaffolding it with `tool/tsf.py new-model`, filling the architecture, and verifying with `tsf smoke`. Use when the user wants to integrate a new PyTorch model, register a custom architecture, or wire a new forecaster into the benchmark pipeline.

Verify model(s) end-to-end by running their smoke configs concurrently and reporting PASS/FAIL via `tool/tsf.py smoke`. Use when the user wants to check that a model (new or existing) trains and produces the right output shape, validate a port, or run a quick CI-style gate over many models.

Understand a model in ModernTSF by reading its README card as progressive disclosure — paper venue/date/arXiv/abstract first, then source only if needed. Use when the user asks "what is model X", "tell me about X", "了解 X 模型", "how does X work", which paper is X from, or wants to compare/learn what a forecaster does before running it.

Add or run a *probabilistic* forecasting model in ModernTSF — one that outputs quantiles or a distribution (prediction intervals / uncertainty) instead of a single point, and is scored with CRPS / WQL / coverage. Use this whenever the task involves prediction intervals, quantiles, uncertainty, a probabilistic or distributional forecast, pinball / quantile loss, Gaussian NLL, or the metrics `crps` / `wql` / `coverage_80` / `width_80` — even if the user only says "predict a range" or "give me uncertainty bands". Do NOT use the plain `add-model` flow for these: it builds a point model. Probabilistic output is a separate `output_type` axis, orthogonal to `task.mode`.

Package a finished run into a TSEval Submission Report (result + agent trajectory + report) and contribute it via a GitHub PR to the TSEval leaderboard repo, via `tsf trace` / `tsf submit`. Use when the user wants to submit results to the TSEval leaderboard, capture an experiment trajectory, or build a leaderboard from submissions.

Diagnose and report a ModernTSF framework defect upstream, with user approval, as a GitHub issue or a small verified PR against Diaugeia/ModernTSF. Use after reproducing a crash in src/ or tool/, wrong output or shapes, a broken config or registry, a doc/CLI mismatch, or another repository defect rather than a problem in the user's data or code.

Guide the user through adding a new dataset to the ModernTSF project. Use when the user wants to register a new dataset, integrate a CSV file or pre-split folder, or wire up a custom data source for benchmarking.

Aggregate experiment results from work_dirs into a combined CSV and optionally plot a bubble chart. Use when the user wants to collect, summarize, filter, or visualize benchmark results for a dataset.

Extract TFB-style statistical characteristics (trend strength, seasonality strength, stationarity) from a dataset. Use when the user wants to profile, describe, or quantify the trend/seasonality/stationarity of a time-series dataset before benchmarking.

Run one-click ablation studies and hyperparameter sweeps via the [sweep] config mechanism, and visualize a trained model's forecasts vs ground truth. Use when the user wants to ablate model components, search hyperparameters, compare model variants across a grid, or plot prediction case studies.

Download GIFT-EVAL datasets from HuggingFace and run the full 53-dataset GIFT-EVAL benchmark sweep. Use when the user wants to run, reproduce, or evaluate against the GIFT-EVAL benchmark.

Preview sweep expansion for a run config — reports total run count, datasets, models, pred lengths, seeds, and per-axis sweep values without launching any training. Use when the user wants to preview how many runs or which datasets/models a config expands to before launching an experiment.

Plot a bubble chart from an already-aggregated results CSV. Use when the user wants to visualize model performance as a bubble chart and already has a CSV file (e.g. produced by the `aggregate` skill or `tool/aggregate_results.py`).

Pre-process a dataset into pre-windowed .npz files for use with the ModernTSF `pre_processed` dataset type. Use when the user wants to convert CSV data to .npz format before training, or to set up a new dataset with pre-computed windows.

Rank models per pred_len/seed for a dataset, producing MSE and MAE leaderboard CSVs. Use when the user wants leaderboards, rankings, or to compare model performance across prediction horizons.

Run a benchmark experiment using a TOML config file. Use when the user wants to run, train, or sweep experiments — single dataset/model or multi-axis sweeps — against any of the available configs/runs/ configs.

Visualize dataset samples from a TOML config in the ModernTSF project. Use when the user wants to plot, inspect, or preview time-series samples from a dataset split.

Generate a shareable Markdown benchmark report for a dataset (leaderboard + bubble chart + results table) via `tool/tsf.py report`. Use when the user wants a written report, summary, or scoreboard of experiment results for a dataset after running experiments.

Detect the machine's GPU and CUDA version and install or repair the project environment with uv, selecting the matching PyTorch backend (CPU or CUDA cuXXX). Use when setting up the project on a new machine, when torch.cuda.is_available() is False or imports fail, when switching hardware, or when the user asks how to install dependencies for their GPU.

Batch-run one or more ModernTSF experiment configs (concurrently) via `tool/tsf.py run`. Use when the user wants to launch several TOML run configs in one go, train across multiple sweeps, or run experiments on specific GPUs.