Team name: sungbinchoi
Submission id: 128954
F1 score on Task 1: 0.848
A brief intro about you: PhD. Deep learning enthusiast
Is it allowed to use models trained on task 1 or 2 train set
and then somehow finetune those models for task 3?
For example, if I’m making prediction for 100th frame,
is it allowed to utilize 99th or 101th frame’s mouse position?
Is GPU available on evaluation server?
I think GPU is necessary if we want to experiment with global observation.
My code is also running on TF 1.13.1.
But TF version 1.13.1 is specified on my environment.yml file.
I’m still confused about what went wrong.
I have a same problem.
Evaluation successful but only 1 episode with 4 steps.
Let me ask another question about MINERL_TRAINING_MAX_INSTANCES = 5 in competition_submission_starter_template. I’m not quite sure how INSTANCE is defined.
Does it mean that max 5 gym env could be run in parallel while training?
Or max 5 parallel trainer perform update function independently and somehow merge gradients periodically?
Or max 5 processes (regardless of doing what kind of job) can be deployed while training?
I think it’s stuck again.
Yes, I guess the bug is not fixed yet.
I’ll post a comment on that github issue link.
Now I got the answer.
They said tower-seed value range is intentionally limited to 100 in Round 2.
Although I defined specific tower-seed when I reset the environment,
random seed is assigned instead
when tower-seed >= 100.
It looks like env 2.1 has the same problem.