Tensorflow1.5 可视化及tf.summary.merge_all()的坑

1.自动管理模式：

注意：自动管理模式下，导入已保存的模型继续训练时，会抛异常，该种场景下建议使用手动模式

summary_writer = tf.summary.FileWriter("E:/data/tensorflow-master/1.Cnn_Captcha/result/",  flush_secs=60)
summary_writer.add_graph(sess.graph)#添加graph图
tf.summary.scalar("loss", loss)
tf.summary.scalar("accuracy", accuracy)
sum_ops = tf.summary.merge_all()#自动管理
metall = sess.run(sum_ops, feed_dict={X: batch_x_test, Y: batch_y_test, keep_prob: 1.})
summary_writer.add_summary(metall, global_step=step)  # 写入文件

2.手动管理模式

summary_writer = tf.summary.FileWriter("E:/data/tensorflow-master/1.Cnn_Captcha/result/",  flush_secs=60)
summary_writer.add_graph(sess.graph)#添加graph图
loss_scalar = tf.summary.scalar("loss", loss)
accuracy_scalar = tf.summary.scalar("accuracy", accuracy)
loss_metall, accuracy_metall,  = sess.run([loss_scalar, accuracy_scalar], feed_dict={X: batch_x_test, Y: batch_y_test, keep_prob: 1.})
summary_writer.add_summary(loss_metall, global_step=step)  # 写入文件
summary_writer.add_summary(accuracy_metall, global_step=step)  # 写入文件