on-the-fly interactive labels

[jeremymanning]

After implementing support for streaming data, it would be nice to be able to interactively label plot points. The way I'm imagining this working is that, when visualizing a data stream, the user might press spacebar, prompting a dialogue box to ask for a label for whatever points are active at that moment (i.e. the current "event"). The user would then type an event label, which would be continually marked on the plot.

[jeremymanning]

Current status (1.0 update): not addressed. Streaming (GH #101, its stated prerequisite) is fully built — hypertools/io/streaming.py's plot_stream() fits models on the first stream_init samples then, in a while True loop (~line 260 on), pulls each new stream_chunk, projects it through the frozen head-fitted transform, and calls _redraw() which does fig.canvas.draw_idle(); plt.pause(0.001) when interactive is True. But there is no key_press_event binding anywhere in the codebase (only button_press_event/button_release_event/motion_notify_event exist, all pre-existing hover/drag behavior in matplotlib_backend.py), and no text-input/dialog mechanism at all.

How this could be built on the 1.0 architecture: plot_stream() in hypertools/io/streaming.py is the natural extension point — it already owns fig, the running accum list of consumed/projected samples, and n_seen. Concretely:

• Add fig.canvas.mpl_connect('key_press_event', on_key) right where the existing interactive flag is computed (~line 232), guarded the same way (skip for Agg/headless backends).
• on_key(event), on event.key == ' ': since matplotlib has no built-in text widget, prompt via a blocking input() in the terminal (works uniformly across backends, no new GUI dependency) or, for a nicer UX, tkinter.simpledialog.askstring when the backend is TkAgg (matplotlib's default on most platforms) — mplcursors-style optional dependency, not required.
• Record (label, n_seen, accum snapshot index) into a new fig.stream_info['labels'] list (parallel to the existing data/xform_data/n_samples keys already returned), and mark the label on the plot the same way static labels= annotations are drawn today (matplotlib_backend.annotate_plot/plotly_backend._build_point_annotations), so this reuses the existing label-rendering code rather than inventing new drawing logic.
• Plotly backend: streaming currently only renders through matplotlib (plot_stream hardcodes matplotlib's interactive loop), so full cross-backend parity would be a separate, larger effort; the initial implementation could reasonably ship matplotlib-only.

Alternatives / prior art:

• mplcursors (pypi, actively maintained, 0.7.1 current) gives clickable/hoverable annotations but no keypress-triggered text-entry dialog — would not replace the core "prompt for a label" step, only possibly the picking/highlight styling.
• Jupyter widgets (ipywidgets.Text) could replace the Tk dialog for notebook users, at the cost of a second code path.

Note: this is gated on nothing else technically (streaming already exists) — it's genuinely new interactive-UI work, not a data-pipeline gap.

Tagging as medium effort; leaving open for consideration in a future release.

[jeremymanning]

Do not implement for now