Adam Gleave

> @AdamGleave Porting the code to TensorFlow 2 should not take that much time, given there are already scripts that automate this process to some extent, so I think you...

> @AdamGleave Is it possible to use [`baselines`](https://github.com/openai/baselines) (rather than stable-baselines)? `baselines` apparently supports TF 2 too (see [here](https://github.com/openai/baselines/tree/tf2)). If it's not too complicated to use baselines rather stable-baselines, this...

I'm not planning on switching to `baselines`, sorry, but would merge a PR that made that change along with other ports to TensorFlow 2. Right now my priority is on...

A workaround you could try is to use the [Docker](https://hub.docker.com/repository/docker/humancompatibleai/evaluating_rewards) image which should avoid these dependency issues.

Thanks for the link! I'm confused how that implementation works: it doesn't seem to look at transition probabilities. What assumptions is it making? (A deterministic gridworld, perhaps?)

We can also use ghcr.io which is already set up

As I replied via e-mail: > Are you using the Conda environment with Ray version 0.4.1? It's pretty old code, the Ray API has likely changed a lot since I...

Hi Lukas, My understanding is that `friendlyPassOk` is _only_ an option for GTP and `analysis`. We evaluate using `match`, not GTP, so I don't think it's even a supported flag...

Hi all, Thanks for your feedback. We're well aware these games are only won under computer Go scoring rules, not human play. We should have made this clearer in the...

> I'm jumping to conclusion based around `friendlyPassOk: false` because this is exactly what happens in all your games. KataGo is (friendly) passing early. Understandable, it does look similar, but...