K.R. Zentner

Hi Steve, Sorry for the slow response. What you want to do should be possible using a Tuple space of Discrete spaces, as you've mentioned. However, the existing policies were...

Wow, this PR looks good. The main feedback I have is that we don't actually yet have any algorithms implemented in PyTorch that can train an RNN. Unfortunately, our VPG...

@irisliucy @avnishn Is this still happening?

Thanks for reporting this. I suppose the overall effect is that the alpha learning rate is essentially multiplied by the number of tasks being trained. This should be about as...

I don't believe this is possible with the current API, but it is not difficult to implement. My recommendation is to copy [gaussian_lstm_model.py](https://github.com/rlworkgroup/garage/blob/6461a071f0155712add1b41316003e90c9c77899/src/garage/tf/models/gaussian_lstm_model.py#L16) into your project's source, and modify [line...

To add to what Avnish said, if you don't use `RaySampler` (and instead use `LocalSampler`), then garage doesn't interact with `ray` at all, and everything _should_ just work out of...

Yeah, this breakage occurred because dm-tree's build script is broken, and the platform you're on doesn't have a pre-built dm-tree package uploaded to PyPI. Here are a few workarounds you...

Hi Benedikt. Unfortunately there isn't currently any way of doing that with the `MetaEvaluator`. A pull request to implement it would be appreciated. There is currently the `num_grad_updates` in MAML,...

This change does not yet pass tests, but is 90% complete.

The core motivation here is to provide a way for recurrent and non-recurrent policies to share the same API at optimization time. However, I definitely agree that making this change...