simple binary shapley values for data records against query output

ssv.py

import itertools
import math

def query_age(dataset):
    """
    A query function that returns records with age less than 26.
    """
    return [record for record in dataset if record['age'] < 26]

def query_trust(dataset):
    """
    A query function that returns records with trust equal to 100.
    """
    return [record for record in dataset if record['trust'] == 100]

def shapley_value(dataset, query_func):
    """
    Compute the Shapley value for each record in the dataset based on a given query.
    """
    n = len(dataset)
    shapley_values = [0] * n

    # Generate all possible coalitions (subsets) of the dataset
    for i in range(1, n + 1):
        for coalition in itertools.combinations(range(n), i):
            # Calculate the value of the coalition and the value of the coalition without each player
            best_of_coalition = query_func([dataset[j] for j in coalition])
            for j in coalition:
                best_without_j = query_func([dataset[k] for k in coalition if k != j]) if i > 1 else []
                marginal_contribution = 1 if dataset[j] in best_of_coalition and dataset[j] not in best_without_j else 0
                shapley_values[j] += marginal_contribution / (math.comb(n, i) * i)

    return shapley_values

# Sample dataset: 3 records with 3 fields
dataset = [
    {'age': 20, 'value': 1, 'trust': 10},
    {'age': 25, 'value': 5, 'trust': 50},
    {'age': 30, 'value': 10, 'trust': 100}
]

# Compute Shapley values for the query 'age < 26'
shapley_values_age = shapley_value(dataset, query_age)
print(f"Shapley values for query 'age < 26': {shapley_values_age}")

# Compute Shapley values for the query 'trust = 100'
shapley_values_trust = shapley_value(dataset, query_trust)
print(f"Shapley values for query 'trust = 100': {shapley_values_trust}")