Time Filter & Animation

Filter data by a time column on the GPU and animate a sliding window at 60 fps — the data is uploaded once and never re-serialized during playback (deck.gl DataFilterExtension).

1. Give the layer a filter accessor

layer = dgl.ScatterplotLayer(
    data=points,
    get_position=["lon", "lat"],
    get_filter_value="t",          # numeric time column
)

get_filter_value auto-attaches the DataFilterExtension. Keep values float32-safe — seconds/days since your domain start, not raw epoch seconds (the GPU filter uses 32-bit floats).

2. Drive the animation

domain = dgl.compute_time_domain(points, "t")          # [t_min, t_max]
deck_map = dgl.Map(layers=[layer], time_filter=dgl.build_time_filter(
    domain,
    window=(domain[1] - domain[0]) * 0.1,  # visible slice [T-window, T]
    playing=True,
    soft_edge=2.0,                          # optional fade in/out
))

Playback runs entirely client-side; the widget reports the throttled head time back as current_time:

mo.md(f"t = {widget.value.get('current_time', 0):.1f}")

Reassign deck_map.time_filter = dgl.build_time_filter(...) from a reactive cell to play/pause, scrub (current=), or change the window.

The pure-reactive alternative

In marimo you can skip the client-side animation entirely and drive the GPU window from a slider — no playback loop, just traitlet sync:

t = mo.ui.slider(0, 100, value=50, label="t")

# reactive cell
deck_map.set_layers([
    dgl.ScatterplotLayer(
        data=points,
        get_position=["lon", "lat"],
        get_filter_value="t",
        filter_range=[t.value - 10, t.value],
    )
])

Use time_filter for smooth autonomous playback; use filter_range when a marimo control should own the clock.

See examples/17_time_filter.py for a runnable 100k-point demo.