Clean and segment a trajectory

Use this recipe when raw positions contain implausible jumps or when a long trace must be split into movement episodes. It preserves the dataframe backend and makes no assumptions beyond time, latitude, and longitude columns.

1. Filter implausible movement

from fastmob.preprocessing import filter

clean = filter(traj, method="smart_greedy", max_speed_kmh=160)

filter removes rows that imply unrealistic movement under the selected strategy. Tune the speed threshold to the transport modes represented by your data; do not use a car threshold for walking-only traces.

2. Split observation gaps

from fastmob.preprocessing import segment

segmented = segment(clean, method="observation_gap", gap_s=30 * 60)

The returned frame adds a segment identifier. Each user receives independent segments, so a gap for one person cannot split another person's trajectory.

3. Verify the result

print(segmented.select(["uid", "datetime", "segment_id"]).head())

Check that segment boundaries correspond to genuine collection gaps rather than routine sampling intervals. See the preprocessing reference for speed, stop, temporal, direction-change, and value-change strategies.