Processing Event Stream Data

Event data records discrete actions during a match - passes, shots, tackles, and more. Processing this data effectively is fundamental to soccer analytics. ## Event Data Structure Event data typically includes: - Timestamp and period - Event type and subtype - Player and team identifiers - Location coordinates - Outcome (success/failure) - Additional context (body part, technique, etc.) ## Loading and Initial Processing ```python import pandas as pd import json def load_event_data(filepath): with open(filepath, 'r') as f: events = json.load(f) # Flatten nested structures events_df = pd.json_normalize(events) # Convert timestamp to seconds events_df['minute'] = events_df['minute'].astype(int) events_df['second'] = events_df['second'].astype(int) events_df['match_second'] = events_df['minute'] * 60 + events_df['second'] return events_df ``` ## Filtering and Querying Events ```python def get_team_passes(events_df, team_name): """Extract all passes for a specific team""" passes = events_df[ (events_df['type'] == 'Pass') & (events_df['team'] == team_name) ].copy() # Separate successful and unsuccessful passes['successful'] = passes['pass_outcome'].isna() return passes def get_possession_sequences(events_df): """Split events into possession sequences""" events_df['possession_change'] = ( events_df['team'] != events_df['team'].shift() ).cumsum() return events_df.groupby('possession_change') ``` ## Calculating Basic Metrics ```python def calculate_passing_metrics(passes_df): """Calculate team passing statistics""" metrics = { 'total_passes': len(passes_df), 'completed_passes': passes_df['successful'].sum(), 'pass_accuracy': passes_df['successful'].mean() * 100, 'forward_passes': (passes_df['pass_end_x'] > passes_df['x']).sum(), 'avg_pass_length': np.sqrt( (passes_df['pass_end_x'] - passes_df['x'])**2 + (passes_df['pass_end_y'] - passes_df['y'])**2 ).mean() } return metrics ``` ## Building Sequence Analysis ```python def analyze_sequences(events_df): """Analyze possession sequences""" sequences = [] for seq_id, sequence in events_df.groupby('possession_change'): sequences.append({ 'sequence_id': seq_id, 'team': sequence.iloc[0]['team'], 'duration': sequence['match_second'].max() - sequence['match_second'].min(), 'num_events': len(sequence), 'num_passes': (sequence['type'] == 'Pass').sum(), 'ended_in_shot': (sequence['type'] == 'Shot').any(), 'start_x': sequence.iloc[0]['x'], 'end_x': sequence.iloc[-1]['x'] }) return pd.DataFrame(sequences) ``` ## Handling Data Quality Issues Event data often requires cleaning: ```python def clean_event_data(events_df): # Remove events with missing critical data events_df = events_df.dropna(subset=['x', 'y', 'type']) # Standardize coordinates to 0-105 x 0-68 events_df['x'] = events_df['x'].clip(0, 105) events_df['y'] = events_df['y'].clip(0, 68) # Handle duplicate events events_df = events_df.drop_duplicates(subset=['timestamp', 'player', 'type']) # Sort by timestamp events_df = events_df.sort_values('match_second').reset_index(drop=True) return events_df ``` ## Advanced: Pass Network Construction ```python def build_pass_network(passes_df): """Create pass network between players""" successful_passes = passes_df[passes_df['successful']].copy() network = successful_passes.groupby(['player', 'pass_recipient']).agg({ 'id': 'count' }).rename(columns={'id': 'passes'}).reset_index() # Calculate edge weights total_passes = network.groupby('player')['passes'].transform('sum') network['weight'] = network['passes'] / total_passes return network ``` Proper event data processing creates the foundation for all downstream analysis, from simple statistics to complex machine learning models.