Picking & Events

Layers are pickable=True by default. Clicks and hovers flow back to Python through the click_info and hover_info traitlets:

deck_map = dgl.Map(layers=[layer])
widget = deck_map.as_widget()

# In a downstream cell — re-runs whenever the user clicks the map
info = widget.value.get("click_info", {})
picked_row = info.get("object")        # the full data row
coordinate = info.get("coordinate")    # [lon, lat] under the cursor
layer_id = info.get("layer_id")
index = info.get("index")              # row index within the layer

hover_info has the same shape and updates as the pointer moves over features.

How binary layers stay pickable

With binary transport, deck.gl has no row objects to hand back — a pick only carries the feature index. The widget bridges this transparently: when a pick event arrives with object: null, the Python side looks up the source row by index (cached per layer) and back-fills object, so downstream code treats JSON and binary layers identically.

Tooltips

The default hover tooltip reads a tooltip key off the picked row:

df["tooltip"] = df["name"] + ": " + df["value"].astype(str)
layer = dgl.ScatterplotLayer(data=df, get_position=["lon", "lat"])

For binary layers the tooltip strings are pre-packed at serialization time and looked up by index in JS — same column, same behavior.

Highlighting

dgl.ScatterplotLayer(..., auto_highlight=True)

Viewport readback

The current camera is always available:

vp = widget.value.get("viewport", {})
# {"longitude": ..., "latitude": ..., "zoom": ..., "pitch": ..., "bearing": ...}