Open Benben0506 opened 2 years ago
I have similar problems, too.
Thanks!
I think I know what is the epoch now. You see the maximum steps parameter in the train.yml file. That is the number of mini-batch the network will be trained. You just reduce that to let's say 1000, and the training process will stop fairly quickly.
But it seems that code does not train even 1 epoch.
The number of five crop dataset images is 488510.
5000 steps which is a num of optimizer steps (self.net + self.linear_probe + self.cluster_probe) is extremely less than 488510.
update(5000 / 3) * batchsize(32) = 53000
I am curious how "5000" steps come from. There is no mention about epochs or iterations in paper.
But I did not find the max_epochs in train_config.yml. Are you saying I should add one myself?
Yes! Ex) trainer = Trainer (max_epochs = 10, ....)
Thanks, I will try.
You can manually install the package in anaconda.
You can manually install the package in anaconda. I have also tried to install each package separately, but that still doesn't work. Whether through the official website or the 'conda' command, maybe I'm not familiar with this operation. Thank you for your reply.
但仍然不起作用。不管是通過官網還是conda命令,可能我對這個操作不太熟悉。感謝你的
Hello sir, I'm spend a lot of time to set max_epochs by myself but all failed..Would you share your code where you change? thanks a lot!!!
Hi,
Could you please tell me how to control the number of epochs? I did not see this hyperparameter in tran.yml.