如何在 TensorFlow 中评估 CIFAR-10 时获得类概率？

Question

我尝试从Convolutional Neural Network TensorFlow Tutorial 修改code 以从每个测试图像中获取每个类的单一概率。

我可以使用 tf.nn.in_top_k 的替代品吗？因为这个方法只返回一个布尔张量。但我想保留个人价值观。

我使用的是 Tensorflow 1.4 和 Python 3.5，我认为第 62-82 行和第 121-129 / 142 行可能是要修改的行。有人给我提示吗？

第 62-82 行：

def eval_once(saver, summary_writer, top_k_op, summary_op):
  """Run Eval once.
  Args:
    saver: Saver.
    summary_writer: Summary writer.
    top_k_op: Top K op.
    summary_op: Summary op.
  """
  with tf.Session() as sess:
    ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
    if ckpt and ckpt.model_checkpoint_path:
      # Restores from checkpoint
      saver.restore(sess, ckpt.model_checkpoint_path)
      # Assuming model_checkpoint_path looks something like:
      #   /my-favorite-path/cifar10_train/model.ckpt-0,
      # extract global_step from it.
      global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
    else:
      print('No checkpoint file found')
return

121-129 + 142 行

    [....]
    images, labels = cifar10.inputs(eval_data=eval_data)

    # Build a Graph that computes the logits predictions from the
    # inference model.
    logits = cifar10.inference(images)

    # Calculate predictions.
    top_k_op = tf.nn.in_top_k(logits, labels, 1)
    [....]

Answer 1

您可以从原始logits 计算类别概率：

# The vector of probabilities per each example in a batch
prediction = tf.nn.softmax(logits)

作为奖励，以下是获得准确度的方法：

correct_pred = tf.equal(tf.argmax(logits, 1), tf.argmax(labels, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))