email_spam.py 文件源码

def get_likelihood(term_document_matrix, label_index, smoothing=0):
    """ Compute likelihood based on training samples
    Args:
        term_document_matrix (sparse matrix)
        label_index (grouped sample indices by class)
        smoothing (integer, additive Laplace smoothing parameter)
    Returns:
        dictionary, with class as key, corresponding conditional probability P(feature|class) vector as value
    """
    likelihood = {}
    for label, index in label_index.items():
        likelihood[label] = term_document_matrix[index, :].sum(axis=0) + smoothing
        likelihood[label] = np.asarray(likelihood[label])[0]
        total_count = likelihood[label].sum()
        likelihood[label] = likelihood[label] / float(total_count)
    return likelihood