Hi, first, thank you for your work in putting this together! Good stuff.
I wanted to make a suggestion, that there could be a lot of value in adding random seeds across the board to your environments. This feature would enables users to a) focus on solving a single instance (e.g., a large/hard one where numerical optimization fails) and b) reproduce results and, by virtue of this, benchmark policies. This is alluded to in this merge, and I think the work is basically the same for the other models. I didn't see a way to do this via a PR given my own time limitations so just wanted to drop the thought here.
Hi, first, thank you for your work in putting this together! Good stuff.
I wanted to make a suggestion, that there could be a lot of value in adding random seeds across the board to your environments. This feature would enables users to a) focus on solving a single instance (e.g., a large/hard one where numerical optimization fails) and b) reproduce results and, by virtue of this, benchmark policies. This is alluded to in this merge, and I think the work is basically the same for the other models. I didn't see a way to do this via a PR given my own time limitations so just wanted to drop the thought here.
Thanks!