google-deepmind / google-deepmind/deepmind-research
TVT reported inconsistent
- Dominant language
- Jupyter Notebook
- Stars
- 15.2k
- Forks
- 2.9k
- PR merge metrics
- No merged PRs in 30d
Description
Hi guys! Was trying out the awesome Temporal value transport code you guys kindly provided at https://github.com/deepmind/deepmind-research/tree/master/tvt
While I got matching results to the Key-to-door --> with last phase reward reaching 10 very quickly when TVT is enabled, **_I was unable to get a moving average of more than 4 or 5 for last phase reward in the Active-visual-match task_** even after many days of training epochs (with TVT enabled of course).
Are the provided **_hyperparameters_** provided by default in the repository as given -- correct for the Active-visual-match task? Or do they need to be modified?
Thank you in advance for your kind help! :)
Contributor guide
Assessment
This issue has not been assessed yet.