TVI Football Index

A Python library for calculating the Tactical Versatility Index (TVI), a metric that quantifies a player's ability to perform various actions across different zones of the football pitch. TVI measures how versatile a player is by analyzing their action diversity and spatial coverage.

What is TVI?

The Tactical Versatility Index measures player versatility by:

• Action Diversity: How many different types of actions a player performs
• Spatial Coverage: How many different pitch zones a player is active in
• Time Normalization: Adjusting for actual playing time

Higher TVI scores indicate more versatile players who contribute across multiple areas and action types. The contribution of each action can be optionally weighted by action rarity (quantiles), giving higher weights to the "action diversity" score for players that perform less common actions.

Installation

pip install tvi-footballindex

Quick Start

Basic Example with Sample Data

import pandas as pd
from tvi_footballindex.tvi.calculator import calculate_tvi, aggregate_tvi_by_player

# Create sample event data
events_df = pd.DataFrame({
    'player_id': [1, 1, 2, 2, 1, 2],
    'event_name': ['pass', 'dribble', 'shot', 'pass', 'tackle', 'interception'],
    'x': [30, 60, 80, 70, 20, 40],  # x-coordinate (0-100)
    'y': [50, 40, 60, 30, 50, 20],  # y-coordinate (0-100)
    'game_id': [1, 1, 1, 1, 1, 1],
    'team_id': [101, 101, 102, 102, 101, 102]
})

# Create playtime data (in minutes)
playtime_df = pd.DataFrame({
    'player_id': [1, 2],
    'play_time': [90, 75],
    'game_id': [1, 1],
    'team_id': [101, 102]
})

# Calculate TVI
tvi_results = calculate_tvi(events_df, playtime_df)
print("Game-level TVI:")
print(tvi_results[['player_id', 'action_diversity', 'TVI']].head())

# Aggregate across all games for each player
player_tvi = aggregate_tvi_by_player(tvi_results)
print("\nPlayer-level TVI:")
print(player_tvi[['player_id', 'action_diversity', 'TVI']].head())

Required Data Format

Your data needs two DataFrames:

Events DataFrame must contain:

• player_id: Unique player identifier
• event_name: Type of action (e.g., 'pass', 'shot', 'tackle')
• x, y: Coordinates on the pitch (0-100 scale recommended)
• game_id: Game identifier
• team_id: Team identifier

Playtime DataFrame must contain:

• player_id: Unique player identifier
• play_time: Minutes played in the game
• game_id: Game identifier
• team_id: Team identifier

Customization

Custom Column Names

If your data uses different column names:

tvi_results = calculate_tvi(
    events_df, 
    playtime_df,
    player_id_col='player_uuid',
    event_name_col='action_type',
    x_col='pos_x',
    y_col='pos_y'
)

Custom Pitch Zones

The default creates a 3×3 grid (9 zones). You can customize this:

# Create a 4×4 grid with 16 zones
custom_zones = [
    [1, 2, 3, 4],
    [5, 6, 7, 8], 
    [9, 10, 11, 12],
    [13, 14, 15, 16]
]

tvi_results = calculate_tvi(
    events_df, 
    playtime_df, 
    zone_map=custom_zones
)

Scaling Factor

Adjust the TVI scaling constant (default is ~2.05):

tvi_results = calculate_tvi(
    events_df, 
    playtime_df,
    C=3.0  # Higher values increase TVI scores
)

Working with F24 Data

If you have Wyscout F24 XML files:

from tvi_footballindex.parsing import f24_parser

# Parse F24 files from a folder
events_df = f24_parser.parsef24_folder("path/to/f24_folder")

# Calculate playtime (minimum 30 minutes to be included)
playtime_df = f24_parser.calculate_player_playtime(events_df, min_playtime=30)

# Extract specific action types
passes = f24_parser.get_progressive_passes(events_df)
dribbles = f24_parser.get_dribbles(events_df)
shots = f24_parser.get_shots_on_target(events_df)
# ... other action extractors available

# Combine all actions
all_actions = pd.concat([passes, dribbles, shots])

# Calculate TVI
tvi_results = calculate_tvi(all_actions, playtime_df)

Understanding the Results

The main metrics returned are:

• action_diversity: Weighted or unweighted count of unique action-zone combinations, depending on configuration
• TVI: Main versatility score (0-1, higher = more versatile)
• TVI_entropy: Alternative entropy-based score (optional)
• shannon_entropy: Raw entropy of action distribution

API Reference

Core Functions

calculate_tvi(events_df, playtime_df, **kwargs)

Calculate TVI scores for each player in each game.

Parameters:

• events_df (DataFrame): Event data with coordinates
• playtime_df (DataFrame): Playing time data
• player_id_col (str): Column name for player IDs (default: 'player_id')
• event_name_col (str): Column name for event types (default: 'event_name')
• x_col, y_col (str): Column names for coordinates (default: 'x', 'y')
• C (float): Scaling constant (default: 90/44 ≈ 2.05)
• zone_map (list): Grid defining pitch zones
• weight_by_quantiles (bool): If True, actions are weighted by rarity (inverse Q95). Default: True
• q_values (dict): Pre-computed action quantiles (per action type). If None, computed on the fly.
• q_level (float): Quantile level for rarity weighting (default: 0.95).

Returns: DataFrame with TVI scores per player per game

aggregate_tvi_by_player(tvi_df, **kwargs)

Aggregate game-level TVI into player-level statistics.

Parameters:

• tvi_df (DataFrame): Output from calculate_tvi()
• player_id_col (str): Column name for player IDs
• playtime_col (str): Column name for playtime

Returns: DataFrame with aggregated TVI per player

Examples and Use Cases

• Squad Analysis: Compare versatility across your team
• Recruitment: Identify versatile players in other teams
• Tactical Analysis: Understand how formation changes affect versatility
• Player Development: Track versatility improvement over time

Contributing

Contributions welcome! Please feel free to submit issues or pull requests.

License

MIT License - see LICENSE file for details.