luxury-lakehouse/xg-shot-data
收藏资源简介：
--- language: [en] license: cc-by-nc-4.0 task_categories: - tabular-classification tags: - sports-analytics - soccer - football - expected-goals - xg - shots - statsbomb - wyscout size_categories: - 100K<n<1M configs: - config_name: default data_files: - split: train path: "data/**/*.parquet" --- # xG Shot Data — StatsBomb + Wyscout **~131K professional soccer shots** from [StatsBomb Open Data](https://github.com/statsbomb/open-data) (~88K) and [Wyscout](https://figshare.com/collections/Soccer_match_event_dataset/4415000) (~43K), with geometric features, categorical context, and goal labels. Partitioned by `data_source` for selective loading. Part of the (Right! Luxury!) Lakehouse soccer analytics platform. ## Quick Start ```python from datasets import load_dataset ds = load_dataset("luxury-lakehouse/xg-shot-data") df = ds["train"].to_pandas() # Goal conversion rate by body part df.groupby("shot_body_part")["is_goal"].mean().sort_values(ascending=False) ``` > **Explore interactively:** [HF Space demo](https://huggingface.co/spaces/luxury-lakehouse/soccer-analytics-demo) ## What Is This Dataset? This dataset contains every shot from the StatsBomb and Wyscout open data collections, enriched with pre-computed geometric features (distance to goal, shot angle) and unified to a common coordinate system. It serves as the primary training input for both xG models in the platform: | Model | Repo | Architecture | |-------|------|-------------| | **xG v1** | [`xg-model-statsbomb-wyscout`](https://huggingface.co/luxury-lakehouse/xg-model-statsbomb-wyscout) | Calibrated XGBoost (13 tabular features) | | **xG v2** | [`xg-v2-model-set-encoder`](https://huggingface.co/luxury-lakehouse/xg-v2-model-set-encoder) | Deep Sets encoder + MLP with MC Dropout (tabular + freeze-frame context) | ## Data Fields | Column | Type | Description | |--------|------|-------------| | `shot_id` | `string` | Surrogate key (deterministic hash via `dbt_utils.generate_surrogate_key`) | | `match_id` | `Int64` | Match identifier | | `competition_id` | `Int64` | Competition identifier (NULL for Wyscout) | | `season_id` | `Int64` | Season identifier (NULL for Wyscout) | | `player_id` | `Int64` | Player identifier | | `team_id` | `Int64` | Team identifier | | `period` | `Int64` | Match period (1 = first half, 2 = second half, 3+ = extra time) | | `minute` | `Int64` | Match minute | | `second` | `Int64` | Second within the minute | | `location_x` | `float64` | Shot x-coordinate (StatsBomb: 0–120 yards, attacking direction) | | `location_y` | `float64` | Shot y-coordinate (StatsBomb: 0–80 yards) | | `end_location_x` | `float64` | Shot destination x-coordinate | | `end_location_y` | `float64` | Shot destination y-coordinate | | `shot_outcome` | `string` | Categorical outcome: Goal, Saved, Blocked, Off T, Wayward, Post | | `shot_body_part` | `string` | Body part used: Right Foot, Left Foot, Head, Other | | `shot_technique` | `string` | Technique: Normal, Volley, Half Volley, Lob, Overhead Kick, Backheel, Diving Header | | `shot_type` | `string` | Context: Open Play, Free Kick, Corner, Penalty, Kick Off | | `is_goal` | `bool` | Target variable — `true` if `shot_outcome = 'Goal'` | | `distance_to_goal` | `float64` | Euclidean distance from shot location to goal center (yards) | | `shot_angle` | `float64` | Angle subtended by the goal posts from the shot location (radians) | | `is_first_time` | `bool` | Shot taken first-time (no prior control touch) | | `play_pattern` | `string` | Build-up pattern: Regular Play, From Counter, From Corner, From Free Kick, From Keeper, etc. | | `statsbomb_xg` | `float64` | StatsBomb proprietary xG (NULL for Wyscout shots; useful as a benchmark label) | | `data_source` | `string` | Partition key: `statsbomb` or `wyscout` | ### Coordinate System All spatial features use the **StatsBomb coordinate system**: - Pitch dimensions: 120 yards (length) × 80 yards (width) - Origin: top-left corner of the pitch - Attacking direction: left to right (x increases toward opponent goal) - Goal center: approximately (120, 40) Wyscout coordinates (0–100% scale) are converted to StatsBomb coordinates at the dbt staging layer. ## Data Sources | Source | Shots | Matches | License | |--------|-------|---------|---------| | [StatsBomb Open Data](https://github.com/statsbomb/open-data) | ~88K | ~3,000 | CC-BY 4.0 | | [Wyscout Public Dataset](https://figshare.com/collections/Soccer_match_event_dataset/4415000) | ~43K | ~1,900 | CC-BY-NC 4.0 | | **Total** | **~131K** | | CC-BY-NC 4.0 (most restrictive applies) | Coverage includes the Premier League, La Liga, Serie A, Bundesliga, Ligue 1, Champions League, World Cup, and more. ### Partitioning Data is Hive-partitioned by `data_source`: ``` data/ data_source=statsbomb/data.parquet data_source=wyscout/data.parquet ``` Load a single source efficiently: ```python import pandas as pd from huggingface_hub import hf_hub_download # Load only StatsBomb shots path = hf_hub_download( "luxury-lakehouse/xg-shot-data", "data/data_source=statsbomb/data.parquet", repo_type="dataset", ) df_sb = pd.read_parquet(path) ``` ## Use Cases - **xG model training**: Primary input for training expected goals models (logistic regression, XGBoost, Deep Sets, or custom architectures) - **Shot analysis**: Visualize shot maps, compare conversion rates by body part, technique, or play pattern - **Benchmarking**: Compare custom xG models against the included `statsbomb_xg` column on the StatsBomb subset - **Feature engineering**: Pre-computed `distance_to_goal` and `shot_angle` ready for modeling; categorical columns ready for one-hot encoding - **Cross-source research**: Study differences in shot event classification between StatsBomb and Wyscout ## Limitations - **Open data only**: Contains only publicly available StatsBomb and Wyscout shots. Commercial datasets cover additional leagues and seasons. - **No freeze frames**: This dataset contains tabular shot features only. For player positions at the moment of each shot (used by xG v2), see the companion [xG Freeze Frame Data](https://huggingface.co/datasets/luxury-lakehouse/xg-freeze-frame-data) dataset. - **Wyscout NULL columns**: `competition_id`, `season_id`, and `statsbomb_xg` are NULL for all Wyscout shots due to the absence of a cross-source match join for those fields. - **Coordinate conversion**: Wyscout coordinates are converted from percentage-based (0–100) to StatsBomb yards (0–120, 0–80) at the dbt staging layer. Subtle conversion artifacts may exist at pitch boundaries. - **Class imbalance**: Goals are relatively rare (~9–10% of shots). Models should account for this imbalance during training. ## Citation If you use this dataset, please cite the data providers: ```bibtex @misc{statsbomb2024opendata, title={StatsBomb Open Data}, author={{StatsBomb}}, year={2024}, url={https://github.com/statsbomb/open-data}, note={CC-BY 4.0} } ``` ```bibtex @misc{pappalardo2019public, title={A public data set of spatio-temporal match events in soccer competitions}, author={Pappalardo, Luca and Cintia, Paolo and Rossi, Alessio and Massucco, Emanuele and Ferragina, Paolo and Pedreschi, Dino and Giannotti, Fosca}, journal={Scientific Data}, volume={6}, number={1}, pages={1--15}, year={2019}, publisher={Nature Publishing Group} } ``` ```bibtex @software{nielsen2026xgshotdata, title={xG Shot Data: StatsBomb + Wyscout Open Data Shot Features}, author={Nielsen, Karsten Skytt}, year={2026}, url={https://github.com/karsten-s-nielsen/luxury-lakehouse} } ``` ## Companion Resources | Resource | Type | Description | |----------|------|-------------| | [xG Model v1](https://huggingface.co/luxury-lakehouse/xg-model-statsbomb-wyscout) | Model | Calibrated XGBoost + logistic baseline (13 features) | | [xG v2 Set Encoder](https://huggingface.co/luxury-lakehouse/xg-v2-model-set-encoder) | Model | Deep Sets + MLP with MC Dropout uncertainty | | [xG Freeze Frame Data](https://huggingface.co/datasets/luxury-lakehouse/xg-freeze-frame-data) | Dataset | Player positions at shot time (15.58M rows, 323 matches) | | [SPADL/VAEP Action Values](https://huggingface.co/datasets/luxury-lakehouse/spadl-vaep-action-values) | Dataset | Per-action VAEP valuations (~9.5M actions) | | [VAEP Model](https://huggingface.co/luxury-lakehouse/vaep-model-statsbomb-wyscout) | Model | P(scores) + P(concedes) XGBClassifiers | ## Demo Try the interactive [Soccer Analytics Explorer](https://huggingface.co/spaces/luxury-lakehouse/soccer-analytics-demo) — explore shot maps with xG overlays, filter by competition, and compare custom xG against StatsBomb. > **Explore interactively:** [HF Space demo](https://huggingface.co/spaces/luxury-lakehouse/soccer-analytics-demo) ## More Information - **License**: [CC-BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) (inherited from Wyscout data) - **Publishing script**: `scripts/publish_xg_shots_hf.py` (PEP 723 standalone) - **Platform**: [Luxury Lakehouse Soccer Analytics](https://github.com/karsten-s-nielsen/luxury-lakehouse)
