Test warning on tensor construction - simutils

[alvorithm]

https://github.com/mackelab/sbi/blob/b38cc27fb33f4743f3c33e01e69ed5813146d09f/sbi/simulators/simutils.py#L150

UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requiresgrad(True), rather than torch.tensor(sourceTensor).

Present on two lines (return statements).

[janfb]

I think the warning is caused by the cast to tensor, if the parameters/observations are tensor already.

The cast to tensor is necessary because some of the simulators are in numpy, e.g., the Lotka Volterra one.

[janfb]

I suggest to refactor the simulation_wrapper:

• remove the LotkaVolterra related code
• remove the cast to tensor
• assume or require from the user, that any simulator function returns a torch tensor.

[janfb]

Similar problem occurs here: https://github.com/mackelab/sbi/blob/68ec669c2236d405bbd10c66ae6792bf4275ff37/sbi/inference/snpe/base_snpe.py#L262-L263

here we can just remove the cast to tensor because it comes from simulation_wrapper anyways.

[jan-matthis]

Reopening this issue because the UserWarnings are still present on master

[alvorithm]

Did you update pyknos/nflows?

[jan-matthis]

Sure

[alvorithm]

are you sure it worked? Yesterday I had to fight quite a bit, i.e. remove them by hand ultimately via pip uninstall. Have a look at ~/<path-to-miniconda>/envs/sbi/lib/python3.8/site-packages/nflows/transforms/standard.py, you should find there a PiecewiseAffineTransform if it's the right tagged version.

[jan-matthis]

I think so:

# pyknos is at d439bb6df487cc2004a44adc2136b8fc13e89a4c
# nflows is at 6d4f5ddcdcf077e155fb6da8d0c2c6f8c3847b35
# sbi is at fb68a06c97ab5426f3c36fa4a068d99b81c8bf93

Error:

/home/ge57bux/sbibm/src/sbi/sbi/simulators/simutils.py:141: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  return torch.tensor(parameters), torch.cat(all_x)

As a side note: It would be good if there was a simple command in sbi that would give the version/commit info of the deps. This is currently not helpful/informative:

from pyknos import version; version.__version__ # == "0.10"
from nflows import version; version.__version__  # == "0.1"
from sbi import version; version.__version__  # == "20200209-Conor_base"

[alvorithm]

How did you get the commit hashes of the dependencies? I guess with that I could solve your request.

As to the warning in simutils, that's true - I had to build that back because as_tensor conflicted with multiround SNPE.

[janfb]

We dont need the cast to Tensor anymore once the user input checks are merged, because then, everything will be Tensor.