7d9d506e77585e5600b45dc41da1a731a4b30722,nn/loss.py,,sensitivity_specificity_loss,#Any#Any#Any#,83

Before Change

                              values=[1.0] * n_voxels,
                              dense_shape=[n_voxels, n_classes])
    one_hotB = 1 - tf.sparse_tensor_to_dense(one_hot)
    SensSpec = tf.reduce_mean(
        tf.add(tf.multiply(r, tf.reduce_sum(tf.multiply(tf.square(-1 * tf.sparse_add(-1 * pred, one_hot)) \
                                                        , tf.sparse_tensor_to_dense(one_hot)),
                                            0) / tf.sparse_reduce_sum(one_hot, 0)), \
               tf.multiply((1 - r), tf.reduce_sum(tf.multiply(tf.square(-1 * tf.sparse_add(-1 * pred, one_hot)), \
                                                              one_hotB), 0) / tf.reduce_sum(one_hotB, 0))))
    return SensSpec


def l2_reg_loss(scope):

After Change

    // value of unity everywhere except for the previous "hot" locations
    one_cold = 1 - one_hot

    squared_error = tf.square(one_hot - pred)
    specificity_part = tf.reduce_sum(squared_error * one_hot, 0) / tf.reduce_sum(one_hot, 0)
    sensitivity_part = tf.reduce_sum(tf.multiply(squared_error, one_cold), 0) / tf.reduce_sum(one_cold, 0)

    return tf.reduce_sum(r * specificity_part + (1 - r) * sensitivity_part)

Instances