Delivery technique, post target, and data-driven routine clusters

restart_routines tags every corner/free-kick delivery with a delivery_technique (“inswinger”/”outswinger”, from Opta qualifiers 223/224) and a post_target (“near_post”/”far_post”/”central”, relative to which flank the restart was taken from), alongside the existing rule-based routine_type taxonomy.

cluster_routines offers a data-driven alternative to that fixed taxonomy: k-means over delivery geometry, surfacing whatever patterns a team actually repeats rather than only the hand-picked named buckets. Requires the optional ml extra (scikit-learn).

from pathlib import Path

from wa_setpieces import load_matches
from wa_setpieces.core.routines import cluster_routines, restart_routines
from wa_setpieces.viz.plots import plot_routine_clusters

try:
    _here = Path(__file__).resolve().parent
except NameError:
    _here = Path.cwd()
DATA = _here.parent / "tests" / "data" / "sample_match.json"

events = load_matches(DATA)

Delivery technique and post target, alongside the existing taxonomy:

detail = restart_routines(events, "corner")
detail[["eventId", "routine_type", "delivery_technique", "post_target"]]

	eventId	routine_type	delivery_technique	post_target
0	191	central_six_yard	inswinger	near_post
1	223	penalty_area	NaN	near_post
2	404	penalty_area	inswinger	central
3	482	central_six_yard	outswinger	central
4	610	central_six_yard	inswinger	near_post
5	613	central_six_yard	inswinger	near_post
6	723	short	NaN	near_post
7	730	central_six_yard	outswinger	near_post
8	786	central_six_yard	outswinger	far_post

Data-driven clusters, as an alternative to routine_type’s hand-picked buckets – each cluster’s label is an auto-generated summary of its own average geometry:

clustered = cluster_routines(events, "corner", n_clusters=3, random_state=0)
clustered[["eventId", "cluster", "cluster_label"]]

	eventId	cluster	cluster_label
0	191	1	long, backward, left
1	223	0	long, backward, right
2	404	0	long, backward, right
3	482	1	long, backward, left
4	610	0	long, backward, right
5	613	0	long, backward, right
6	723	2	short, backward, right
7	730	1	long, backward, left
8	786	0	long, backward, right

fig, ax = plot_routine_clusters(clustered, title="Corner delivery clusters")