[Bug] Algorithm tests fail in Pytorch 1.7.1

[AVSurfer123]

Steps to reproduce

1. Install this library with Python 3.7 and PyTorch 1.7.1
2. Run the test python3 -m pytest tests/algorithms

Observed Results

You get the errors

FAILED tests/algorithms/test_algorithms.py::test_pets_icem_gaussian_mlp_ensemble - TypeError: clamp(): argument 'min' (position 2) must be Number, not Tensor
FAILED tests/algorithms/test_algorithms.py::test_pets_icem_basic_ensemble_deterministic_mlp - TypeError: clamp(): argument 'min' (position 2) must be Number, not Tensor

because torch.clamp doesn't support tensors for the min and max arguments till version 1.9.0. So either the README should be updated with the newer PyTorch version or the test/code should be fixed to work with 1.7.0 still.

Expected Results

Tests pass successfully.

Relevant Code

The error occurs at

population = mbrl.util.math.powerlaw_psd_gaussian(
    self.colored_noise_exponent,
    size=(decay_population_size, x0.shape[1], x0.shape[0]),
    device=self.device,
).transpose(1, 2)
population = torch.clamp(
    population * torch.sqrt(var) + mu, self.lower_bound, self.upper_bound
)

E           TypeError: clamp(): argument 'min' (position 2) must be Number, not Tensor

mbrl/planning/trajectory_opt.py:439: TypeError

[luisenp]

Hi @AVSurfer123, thanks for reporting! Can you pull the code in #148 and test if it works for you? Thanks!