tensorflow中不同数量元素的标签和预测的Precision和recall eval_metrics

Question

我在Tensorflow中作为eval_metrics注册精度和召回率时遇到了问题。我的标签和预测没有相同数量的元素，所以我不能使用已经内置的函数。我有计算精度和召回率的函数，但是我似乎不能得到precision_update_op和recall_update_op。你知道我怎么才能从标签，预测和前面提到的计算精度和召回率的函数中得到它吗？谢谢

Answer 1

这里有一个如何构建您自己的指标的简单示例。我将演示mean，您也应该能够适应上面提到的内容。

def mean_metrics(values):
   """ For mean, there are two variables that are 
 required to hold the sum and the total number of variables"""

   # total sum
   total = tf.Variable(initial_value=0., dtype=tf.float32, name='total')

   # total count
   count = tf.Variable(initial_value=0., dtype=tf.float32, name='count')

   # Update total op by updating total with the sum of the values
   update_total_op = tf.assign_add(total, tf.cast(tf.reduce_sum(values), tf.float32))

   # Update count op by updating the total size of the values
   update_count_op = tf.assign_add(count, tf.cast(tf.size(tf.squeeze(values)), tf.float32))

   # Mean
   mean = tf.div(total, count, 'value')

   # Mean update op
   update_op = tf.div(update_total_op, update_count_op, 'value')

   return mean, update_op

测试上面的代码：

tf.reset_default_graph()
values = tf.placeholder(tf.float32, shape=[None])

mean, mean_op = mean_metrics(values)

with tf.Session() as sess:
   tf.global_variables_initializer().run()
   print(sess.run([mean, mean_op], {values:[1.,2.,3.]}))
   print(sess.run([mean, mean_op], {values:[4.,5.,6.]}))

#output
#[nan, 2.0]
#[2.0, 3.5]