def client_side_multiclass_compute()

in src/ppxgboost/PPBooster.py [0:0]

• 9 lines of code
• 5 McCabe index (conditional complexity)

def client_side_multiclass_compute(predictions):
    """
    Output the actual predictions using the softmax methods.
    In particular, it computes the normalized exponential function, which converts a vector
    of K real numbers and normalizes it into a probability distribution consisting of K
    probability scores proportional to the exponentials of the input numbers, i.e. SoftMax assigns
    decimal probabilities to each class in a multi-class problem.
    Those decimal probabilities must add up to 1.0.

    :param predictions: a list of predictions, where each element is a list that contains
        the scores for each inputs - here, we allow the client to receive a list of
        prediction (consistent to xgboost's prediction) for the corresponding queries.
    :return: predicted classes: a list of the most probable classes the model predicts.
    """
    final_output = []

    # Predictions is a vector of results, the value can be any real floating point number
    # e.g. predictions for k classes -- predictions = [[x_1,... x_k], [y_1,... y_k], [z_1, ..., z_k]]
    for predict_i in predictions:

        # gets the sum of all the exponential predicted results
        # e.g. e^(x_1) + ... + e^(x_k)
        sum_score = np.sum(np.exp(predict_i))
        output = []
        for x_i in predict_i:
            # for each t in predict_i
            # e.g. e^(x_i)/(e^(x_1) + ... + e^(x_k))
            output.append(np.exp(x_i) / sum_score)

        # report the argmax as the predicted class (most probable).
        final_output.append(np.argmax(output))
    return final_output