visualize

name	visualize
description	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.

When to use / what to ask

Plots raw dataset samples (input + forecast window) from a config. For forecast-vs-truth case plots from a trained model, use the experiments skill instead.

Ask for:

1. Which dataset config (e.g. configs/datasets/etth1.toml) — any dataset-only or full run TOML works
2. Which split: train, val, or test (default train)
3. How many samples (--num-samples, default 3) or a specific --index

Command

uv run python tool/visual_data.py \
  --config configs/datasets/etth1.toml --split train --num-samples 3

Saves to work_dirs/plots/<dataset>_<split>.png by default (input series solid, forecast window after the dashed line). Other knobs: --channels 0,1,2 (default all), --seed S, --save PATH, --show.

Reference

Full flags: docs/en/visualize-data.md. Dataset configs live in configs/datasets/.