Can't show scalar

[zhangguangzhi]

My program uses main() to call the trained function train(), which I have in my train function. Here is my program file. When I executed, there was no expected result, Can't show scalar.

[zhangguangzhi]

Chinese：我的程序是使用main()来调用训练的函数train()，我在我的train函数，以下是我的程序文件。当我执行时，没有出现预想的结果，图形可显示step，但是无法显示值。
而且只能查看到最后一个step

[jetfuel]

@zhangguangzhi It seems that the t_loss was not updated. You can try to generate some fake logs manually to check if the visualDL is working

import random
from visualdl import ROOT, LogWriter

logdir = './scratch_log'
logw = LogWriter(logdir, 2000)

# create scalars in mode train and test.
with logw.mode('train') as logger:
    scalar0 = logger.scalar("scratch/scalar")

with logw.mode('test') as logger:
    scalar1 = logger.scalar("scratch/scalar")

# add fake scalar records. 
last_record0 = 0.
last_record1 = 0.
for step in range(1, 1000):
    last_record0 += 0.1 * (random.random() - 0.3)
    last_record1 += 0.1 * (random.random() - 0.7)
    scalar0.add_record(step, last_record0)
    scalar1.add_record(step, last_record1)

[yeyupiaoling]

@zhangguangzhi
My program happen this problem just now. You can try again. close VisualDL, and then start again.

[zhangguangzhi]

@jetfuel Virsualdl is normal

[zhangguangzhi]

@jetfuel
I modified my code, trained five times, showed the first t_loss for the first time, and finally showed the fifth t_loss. Not a line.

first train

second train

[zhangguangzhi]

@jetfuel Thank you
I will take out this configuration and it will do.
logger = LogWriter("./log", sync_cycle=1) with logger.mode("t_loss") as logger: train_loss = logger.scalar("loss/val")

[jetfuel]

@zhangguangzhi Ah, I see that you create the logger within the for-loop. To use VisualDL, you only need to create the scalar log writer once. Then use that scalar log writer in your training process.

Also, it is recommended to set your sync_cycle larger if you want to record more data points. Saving to file system is expansive. The team is aware of this and is working to improve the performance.