Netflow Training and generation ValueError: Pretrain_dir {} does not exist!

[netsharecmu]

Netflow Training and generation ValueError: Pretrain_dir {} does not exist!

[hallavar]

Oh thank you, yeah that should be good. I reduced the number of iteration because I've wanted to check if the training runs correctly. I will try that.

[hallavar]

Ok, so this was not the problem

I have reduced the number of vis_freq, and it seems to have solved it.
I can finish a training now.

[sniperyyc]

Hi Hallavar,

I did not run your config but something I spotted may be: You are using iteration=500 and "extra_checkpoint_freq": 20000 (500<20000) so you are not actually saving "an extra ckpt" for chunk0 to work as pre-trained models for later chunks.

If you make the extra_checkpoint_freq smaller than your iteration, it should work. Anyway, I will get back to you with more answers after I tried it and a refactored codebase hopefully soon.

Thanks,
Yucheng

[sniperyyc]

Sounds great. I noticed your comment in another issue and that may be a real problem -- I will try to get back to you on that front soon as well.

[sniperyyc]

Ok, good to know. We have not used vis_freq for a while so not sure what happened on that front. I will close the issue for now.