把Python变量记录到TensorFlow的summary中

一般做法如下：

...
# 1、计算准确率
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
 
# 2、添加scalar，汇总
tf.summary.scalar('accuracy', accuracy)
merged = tf.summary.merge_all()
 
...
# 3、运行，写入log
summary, acc = sess.run([merged, accuracy], feed_dict=feed_dict(False))  
test_writer.add_summary(summary, i)  

上面的做法有个问题，在计算准确率时传入的是整个数据集，如果数据集很大，内存会不够用。一般需要分多个batch，每次输入一个batch，计算预测正确的样本数，然后求和除以总样本数：

steps_per_epoch = data.num_examples // FLAGS.batch_size
num_examples = steps_per_epoch * FLAGS.batch_size
true_count = 0
for i in range(steps_per_epoch):
    images_feed, labels_feed = data.next_batch(FLAGS.batch_size)
    true_count += sess.run(correct, feed_dict={images_pl: images_feed, labels_pl:labels_feed, training_pl: False})
validate_accuracy = true_count / num_examples

但是这样得到的结果validate_accuracy是一个变量，如何把普通的Python变量加到summary中？

官方APIhttps://tensorflow.google.cn/api_docs/python/tf/summary/FileWriter#add_summary中描述如下：

You can pass the result of evaluating any summary op, using tf.Session.run or tf.Tensor.eval, to this function. Alternatively, you can pass a tf.Summary protocol buffer that you populate with your own data. The latter is commonly done to report evaluation results in event files.

实现如下：

summary = tf.Summary()
summary.value.add(tag='val_acc', simple_value=validate_accuracy)
summary_writer.add_summary(summary, i)

最后运行结果：

————————————————
版权声明：本文为CSDN博主「qiumokucao」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/qiumokucao/article/details/81489460