Fix erroneous warnings about different devices

[ThomasGesseyJonesPX]

What does this implement/fix? Explain your changes

Fix for erroneous warnings saying tensors are on different devices when they are on the same device.

This bug occurs when PyTorch device objects are used rather than their string representations. In the comparisons that check the tensors are on the right device one side is cast to a string but not the other. When device objects are used this leads to comparing an object to a string and hence them not being equal and a warning being raised, e.g.,

UserWarning: Data x has device 'cuda:0'. Moving x to the data_device 'cuda:0'. Training will proceed on device 'cuda:0'.
  theta, x = validate_theta_and_x(

This bug is fixed by casting both sides of the comparison to strings.

Does this close any currently open issues?

None

Any relevant code examples, logs, error output, etc?

Minimum working example to illustrate bug and replicate the warning shown above,

"""MWE Example of erroneous warning in device comparison."""

import torch
from sbi.inference import SNPE
from sbi.utils.user_input_checks import process_prior, process_simulator

# Set device to GPU
device = torch.device("cuda:0")

# Define prior and simulator to be on the GPU
prior = torch.distributions.MultivariateNormal(
    torch.zeros(2, dtype=float, device=device), torch.eye(2, dtype=float, device=device)
)
simulator = lambda theta: theta

prior.sample()
prior, num_parameters, prior_returns_numpy = process_prior(prior)
simulator = process_simulator(simulator, prior, prior_returns_numpy)

# Also set the NPE to be on the same GPU
inference = SNPE(prior=prior, device=device)
theta = prior.sample((100,))
x = simulator(theta)
inference = inference.append_simulations(theta, x)

# This will raise a warning saying that the device of the density estimator is different from the 
# device of the prior even though they are both on the same GPU
density_estimator = inference.train(max_num_epochs=1)

Any other comments?

None

Checklist

Put an x in the boxes that apply. You can also fill these out after creating
the PR. If you're unsure about any of them, don't hesitate to ask. We're here to
help! This is simply a reminder of what we are going to look for before merging
your code.

• I have read and understood the contribution
  guidelines
• I agree with re-licensing my contribution from AGPLv3 to Apache-2.0.
• I have commented my code, particularly in hard-to-understand areas
• I have added tests that prove my fix is effective or that my feature works
• I have reported how long the new tests run and potentially marked them
  with pytest.mark.slow.
• New and existing unit tests pass locally with my changes
• I performed linting and formatting as described in the contribution
  guidelines
• I rebased on main (or there are no conflicts with main)
• For reviewer: The continuous deployment (CD) workflow are passing.

[janfb]

Makes sense - thanks a lot for fixing this 🙏

device tests are passing locally.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 76.90%. Comparing base (5de7784) to head (6632774).
Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1225      +/-   ##
==========================================
- Coverage   85.44%   76.90%   -8.54%     
==========================================
  Files         101      101              
  Lines        7941     7941              
==========================================
- Hits         6785     6107     -678     
- Misses       1156     1834     +678     

Flag	Coverage Δ
unittests	76.90% <100.00%> (-8.54%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
sbi/utils/nn_utils.py	88.23% <100.00%> (ø)
sbi/utils/user_input_checks.py	76.31% <100.00%> (ø)

... and 25 files with indirect coverage changes