Kevin Greenman

Note for whoever ends up implementing this in v2: #571 contains an example of the intended change, but written in the v1 code

We have released our release candidate v2.0.0-rc.1, please see here for details: https://github.com/chemprop/chemprop/releases/tag/v2.0.0-rc.1

We have released the stable v2.0.0, please see here for details: https://github.com/chemprop/chemprop/releases/tag/v2.0.0

I've mostly been keeping track of which issues/PRs correspond to which versions by adding them each to a milestone. Having the [v2] in the title makes this easier, so you...

I don't think the "user-friendliness" question matters too much as long as we document the way to do it (IMO, `CUDA_VISIBLE_DEVICES="0,2" chemprop train [...] --n-gpu -1` and `chemprop train [...]...

As discussed in our meetings, I agree with the points above about v1 being too flexible and overcomplicating our codebase. @KnathanM I fully support your suggestion of only having a...

could you please post the full strack trace for the error?

I'm pretty sure the reason why the fix of setting `num_workers = 0` is not working for you is that this argument is not passed to `MoleculeDataLoader` during interpretation [here](https://github.com/chemprop/chemprop/blob/1785627d714c17f910af921fe518db76f4892026/chemprop/interpret.py#L59),...

The change will be reflected on PyPI once we merge the PR and after we release v1.7.0. However, the v1.7.0 PyPI release will also include this Python version restriction that...

You're correct that the [interpret function](https://github.com/chemprop/chemprop/blob/76d59ffb0096d96e92c6b7a5072ef95bdc2e883d/chemprop/interpret.py#L298) doesn't return anything (`interpret_out` should be `None` in your script). It should be printing out the results for you from [this line](https://github.com/chemprop/chemprop/blob/76d59ffb0096d96e92c6b7a5072ef95bdc2e883d/chemprop/interpret.py#L345).