luxury-lakehouse/football2vec-360-training-data

Football2Vec 360 Training Data — SPADL Sequences with Freeze-Frame Context

Tokenized SPADL action sequences with StatsBomb 360 freeze-frame context for training the Football2Vec 360-Enriched model. One row per player-match, covering ~2M actions across 323 professional soccer matches from StatsBomb 360 Open Data.

Part of the (Right! Luxury!) Lakehouse soccer analytics platform.

Quick Start

from datasets import load_dataset

ds = load_dataset("luxury-lakehouse/football2vec-360-training-data")
df = ds["train"].to_pandas()
print(f"{len(df)} player-match sequences")

# Inspect one sequence
row = df.iloc[0]
print(f"Player: {row['canonical_player_id']}, Match: {row['match_id']}")
print(f"Actions: {len(row['actions'])} events")
print(f"First action: {row['actions'][0]}")
print(f"First freeze frame: {len(row['freeze_frames'][0]['players'])} visible players")
# action: {'action_type': 0, 'x': 0.52, 'y': 0.34, 'result': 1}
# freeze_frames and actions are parallel arrays — freeze_frames[i] corresponds to actions[i]

Explore interactively: HF Space demo

What Is This Dataset?

Each row represents one player's actions in one match, serialized as a struct array of SPADL-tokenized events. Each action includes the 23-type SPADL token, normalized spatial coordinates, and the StatsBomb 360 freeze-frame: the (x, y) positions of all visible opponents and teammates at the moment of that action.

The freeze-frame context is the key differentiator from the standard Football2Vec v2 training data. It enables the Deep Sets encoder to learn how players behave relative to surrounding players, capturing spatial decision-making beyond what action sequences alone reveal.

This dataset is the training corpus for Football2Vec 360-Enriched. It is exported from the platform's fct_action_values and StatsBomb 360 freeze-frame Delta tables via the export_embeddings_training_data entry point and published here for reproducibility.

Data Fields

Column	Type	Description
canonical_player_id	string	Unified player identifier (from entity resolution across data sources)
match_id	string	Match identifier
competition_id	int	Competition identifier (used as adversarial target in Stage 2 training)
season_id	int	Season identifier
position_group	string (nullable)	Player position group: GK, Def, Mid, Fwd (from dim_players)
actions	array<struct>	Ordered sequence of tokenized SPADL actions
freeze_frames	array<struct>	Parallel array of freeze-frame player positions (aligned with actions by index)

Action Struct Schema

Each element in the actions array:

Field	Type	Description
action_type	int	SPADL action type ID (0–22, 23 action types)
x	float	Normalized x coordinate [0, 1] on 105m pitch
y	float	Normalized y coordinate [0, 1] on 68m pitch
result	int	Binary outcome: 1 = success, 0 = failure

Freeze-Frame Struct Schema

Each element in the freeze_frames array contains a players field with an array of visible player positions:

Field	Type	Description
players	array<struct>	Array of player positions at this action

Each player struct:

Field	Type	Description
x	float	Normalized x coordinate [0, 1] on 120m pitch (StatsBomb)
y	float	Normalized y coordinate [0, 1] on 80m pitch (StatsBomb)
is_keeper	bool	True if this player is a goalkeeper
is_teammate	bool	True if this player is a teammate of the acting player

SPADL Action Vocabulary (23 types)

ID	Action	ID	Action	ID	Action
0	pass	8	foul	16	keeper_punch
1	cross	9	tackle	17	keeper_pick_up
2	throw_in	10	interception	18	clearance
3	freekick_crossed	11	shot	19	bad_touch
4	freekick_short	12	shot_penalty	20	non_action
5	corner_crossed	13	shot_freekick	21	dribble
6	corner_short	14	keeper_save	22	goalkick
7	take_on	15	keeper_claim

Data Sources

Source	Matches	License
StatsBomb 360 Open Data	323	CC-BY 4.0

The 323-match corpus is the complete StatsBomb 360 open-data release. Coverage includes La Liga (multiple seasons), Premier League, Champions League, Euro 2020, Women's World Cup, and Copa America matches with freeze-frame annotations.

Freshness

Metric	Value
Freshness SLA	168 hours (7 days)
Refresh trigger	Re-exported when StatsBomb releases new 360 open-data matches
Publish script	src/ingestion/export_embeddings_training_data.py (entry point: export_embeddings_training_data)

Use Cases

• Transformer + Deep Sets training: Primary training corpus for Football2Vec 360-Enriched (masked language modeling + adversarial debiasing with freeze-frame context)
• Spatial context research: Study how player spatial environment correlates with action choice and outcome
• Custom embedding models: Train your own context-aware player embedding model on standardized SPADL + 360 sequences
• Ablation studies: Compare model performance with vs. without freeze-frame context against the Football2Vec v2 training data

Limitations

• 360 matches only: Covers 323 StatsBomb 360 matches. Players with appearances only in non-360 matches are not represented.
• Uneven freeze-frame coverage: Not every action in a 360-annotated match has a freeze-frame. Actions without freeze-frame context have freeze_frames = [] and fall back to transformer-only representation.
• Open data only: Derived from publicly available StatsBomb 360 data. Coverage is uneven across leagues and seasons; some competitions have more 360 annotations than others.
• Coordinate normalization: All coordinates are normalized to [0, 1] on a 105×68m pitch (SPADL standard). Original StatsBomb coordinate system is not preserved.
• NULL position_group: Players not matched via entity resolution or lacking position metadata have position_group = NULL.

Citation

If you use this dataset, please cite the SPADL framework, the Football2Vec 360-Enriched model, and the Deep Sets architecture:

@inproceedings{decroos2019actions,
  title={Actions Speak Louder than Goals: Valuing Player Actions in Soccer},
  author={Decroos, Tom and Bransen, Lotte and Van Haaren, Jan and Davis, Jesse},
  booktitle={Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
  pages={1851--1861},
  year={2019},
  publisher={ACM}
}

@inproceedings{zaheer2017deep,
  title={Deep Sets},
  author={Zaheer, Manzil and Kottur, Satwik and Ravanbakhsh, Siamak and Poczos, Barnabas and Salakhutdinov, Ruslan and Smola, Alexander},
  booktitle={Advances in Neural Information Processing Systems},
  volume={30},
  year={2017}
}

@software{nielsen2026football2vec_360,
  title={Football2Vec 360-Enriched: Transformer + Deep Sets Player Embeddings},
  author={Nielsen, Karsten Skytt},
  year={2026},
  url={https://github.com/karsten-s-nielsen/luxury-lakehouse}
}

Companion Resources

Resource	Description
Football2Vec 360 Model	144-dim model trained on this data
360 Player Embeddings	Pre-computed 144-dim vectors per player-match
Football2Vec v2 Training Data	Event-only SPADL sequences (~87K player-matches, ~3,000 matches)
Football2Vec v2 Model	128-dim event-only model with broader coverage
SPADL/VAEP Action Values	Per-action offensive/defensive VAEP valuations

More Information

Explore interactively: HF Space demo

• Model repo: luxury-lakehouse/football2vec-360
• License: CC-BY 4.0 (StatsBomb Open Data)