what result mean trainning is finished

[skanger]

the loss is reduce to 0 when step 170 , but it don't stop,
i pressed 'ctrl+c' breaked it, then try to eval and recognize,
but the result of recognize is so bad,

can i break the train when loss equal 0?
or it will finished , i need't do that?

[skanger]

now, it is step 2000 , every single step loss is 0, but is keep traniing
my train set has only 3 pictures, but my mac is old,
how can i know it is finished

[LxKxC]

now, it is step 2000 , every single step loss is 0, but is keep traniing
my train set has only 3 pictures, but my mac is old,
how can i know it is finished

• My captche program runs on raspberry PI 3B+.
• I use python captcha_gen_default.py to make captche，so I think it will stop in the 50000 step.
• It has been running for three days.
• You need to blow Rraspberry Pi with a small fan.If you don't, the Raspberry PI will burn down.

[LxKxC]

now, it is step 2000 , every single step loss is 0, but is keep traniing
my train set has only 3 pictures, but my mac is old,
how can i know it is finished

• I think if the training is complete, the training program will automatically save and exit, machine learning is a time-consuming process.

[skanger]

@LxKxC I got "Saving . in ./captcha_train/captcha" too. Does it could be used as train set result?
if I keep trainning , it will help the result better ?
Thanks for you reply.

[LxKxC]

@skanger I don't think that using the same picture to repeat training can improve the recognition rate.
Let me talk about my understanding:

• Picture A, training once.
• All features of the picture A have been counted in the program.
• Repeated training of the picture A, no new features found, so no help for training

[LxKxC]

@skanger I asked a friend from China who told me:
随着训练次数的上升，残差平方和会下降，但你需要控制防止过拟合
Translation into English generally means:
As the number of training increases, the sum of squared residuals will decrease, but you need to control to prevent overfitting