---

name: approximate-streaming-pan description: "Use STUMPY anytime approximation, streaming updates, online segmentation, and pan matrix profile APIs safely." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

approximate-streaming-pan

Use this sub-skill when the task is about STUMPY workflows that update over time or trade exactness for speed:

• anytime/approximate matrix profiles with scrump, scraamp, prescrump, or prescraamp;
• incremental matrix profiles for streaming data with stumpi or aampi;
• online semantic-segmentation state with floss;
• pan matrix profile / window-size exploration with stimp or aamp_stimp.

Route away

• Exact 1-D matrix-profile setup, self-joins, AB-joins, dtype/window validation, and profile-column fundamentals belong in ../matrix-profile-basics/SKILL.md.
• Dask/Ray stimped/aamp_stimped and CUDA gpu_stimp/gpu_aamp_stimp setup belong in ../distributed-gpu-acceleration/SKILL.md.
• Interpreting discovered motifs, discords, regime changes, snippets, or segmentation decisions belongs in ../motifs-anomalies-segmentation/SKILL.md.

Operating flow

1. Confirm the user needs approximate, streaming, online segmentation, or window-size-selection behavior rather than an exact one-shot profile.
2. Choose normalized (scrump, stumpi, stimp, floss with normalize=True) versus non-normalized p-norm (scraamp, aampi, aamp_stimp, floss(..., normalize=False, p=...)).
3. For anytime objects, create the object, then call .update() repeatedly until the budget or stability criterion is met; read .P_ and .I_ after updates.
4. For streaming objects, call .update(t) once per scalar observation in arrival order; do not pass a batch array into a single update.
5. For pan matrix profiles, call .update() once per processed window-size row; use .M_ to map rows to window sizes, .P_ for raw profiles, and .PAN_/.pan(...) for transformed views.

Bundled references and smoke script

• references/api-reference.md gives signatures, object attributes, and parameter meanings.
• references/workflows.md gives task-oriented recipes and acceptance guidance for approximate outputs.
• references/troubleshooting.md covers update ordering, egress, convergence, window bounds, pan-state interpretation, and FLOSS state traps.
• scripts/streaming_pan_smoke.py runs tiny no-network demos with --workflow scrump|stumpi|stimp.

Construction evidence is summarized in the root provenance and integration records; runtime use should rely on this sub-skill's bundled references and scripts.