simulate_with_shap.py

def simulate_with_shap(x,idx, col, val_range, explainer, pipeline_trans=None):
    """
    Function starts with an instance `idx` of dataset x and varies values of selected column `col`, 
    it needs eplainer for dataset x and optionally it can use the pipeline transformer to transform values as in the initial model

    Parameters
    ----------
    n_min : int
        minimal value to simulate
    n_max : int
        maximal value to simulate
    n_step: int
        step for the simulation values 
    x : pd.Data Frame
        dataset 
    idx : int
        selected instance
    col : str or int
        Indicate name of the column or the index
    explainer : shap.explainer
    
    pipeline_trans : pipeline 
        sklearn transformation of the dataset (optional), if absent dataset used as is
        
    """
    if isinstance(col, str) :
        i = x.columns.get_loc(col)
    else : 
        i = col
    
    rg = val_range

    r = x.iloc[np.repeat(idx, len(rg))].reset_index(drop=True)
    r.iloc[:,i] = rg
    if pipeline_trans:
        R = pipeline_trans.transform(r)
    else:
        R = r
    hypothetical_shap_values = explainer.shap_values(R)
    hypothetical_predictions = explainer.expected_value + hypothetical_shap_values.sum(axis=1)
    hypothetical_predictions = 1 / (1 + np.exp(-hypothetical_predictions))
    y_r = np.array([0 if i <= 0.5 else 1 for i in hypothetical_predictions ])

    return (r, R, hypothetical_shap_values, hypothetical_predictions, y_r)