def sample_values_from_classes()

in ludwig/modules/loss_modules.py [0:0]

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

def sample_values_from_classes(labels, sampler, num_classes, negative_samples,
                               unique, class_counts, distortion):
    """returns sampled_values using the chosen sampler"""
    if sampler == 'fixed_unigram':
        sampled_values = tf.random.fixed_unigram_candidate_sampler(
            true_classes=labels,
            num_true=1,
            num_sampled=negative_samples,
            unique=unique,
            range_max=num_classes,
            unigrams=class_counts,
            distortion=distortion
        )
    elif sampler == 'uniform':
        sampled_values = tf.random.uniform_candidate_sampler(
            true_classes=labels,
            num_true=1,
            num_sampled=negative_samples,
            unique=unique,
            range_max=num_classes
        )
    elif sampler == 'log_uniform':
        sampled_values = tf.random.log_uniform_candidate_sampler(
            true_classes=labels,
            num_true=1,
            num_sampled=negative_samples,
            unique=unique,
            range_max=num_classes
        )
    elif sampler == 'learned_unigram':
        sampled_values = tf.random.fixed_unigram_candidate_sampler(
            true_classes=labels,
            num_true=1,
            num_sampled=negative_samples,
            unique=unique,
            range_max=num_classes,
            unigrams=class_counts,
            distortion=distortion
        )
    else:
        raise ValueError('Unsupported sampler {}'.format(sampler))
    return sampled_values