Added type assertion for better code clarity

[GautamV234]

I am working with Prof. @nipunbatra and I am currently working on this PR to solve #3026.

[fritzo]

Can you make lint and ensure your code is well formatted? I believe you'll need to change the assertion to

assert isinstance(X, torch.Tensor)

[fehiepsi]

I thought that it was recommended to use torch.is_tensor(x) but looking at its docs again, it recommends to use is_instance(...) pattern.

[GautamV234]

Can you make lint and ensure your code is well formatted? I believe you'll need to change the assertion to

assert isinstance(X, torch.Tensor)

Sure I will do that and resubmit

[GautamV234]

I have modified all the files (added assertion checks) having this problem in the pyro/contrib/gp/models folder

Below is an example that illustrates the assertion error.

import pyro
import pyro.contrib.gp as gp
import numpy as np
import torch 
X = np.random.randn(200, 2)
Y = np.logical_xor(X[:, 0] > 0, X[:, 1] > 0).astype(np.float64) 
kernel = gp.kernels.RBF(input_dim=2)
likelihood = gp.likelihoods.Binary()
model = gp.models.VariationalGP(X, Y, kernel, likelihood=likelihood, whiten=True)

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
Input In [6], in <cell line: 11>()
      9 kernel = gp.kernels.RBF(input_dim=2)
     10 likelihood = gp.likelihoods.Binary()
---> 11 model = gp.models.VariationalGP(X, Y, kernel, likelihood=likelihood, whiten=True)

File C:\work\Pyro\pyro\pyro\contrib\gp\models\vgp.py:74, in VariationalGP.__init__(self, X, y, kernel, likelihood, mean_function, latent_shape, whiten, jitter)     63 def __init__(
     64     self,
     65     X,
   (...)
     72     jitter=1e-6,
     73 ):
---> 74     assert isinstance(
     75         X, torch.Tensor
     76     ), "X needs to be a torch Tensor instead of a {}".format(type(X))
     77     assert isinstance(
     78         y, torch.Tensor
     79     ), "y needs to be a torch Tensor instead of a {}".format(type(y))
     80     super().__init__(X, y, kernel, mean_function, jitter)

AssertionError: X needs to be a torch Tensor instead of a <class 'numpy.ndarray'>

Please let me know if some tests need to be added pertaining to the updated code.

[fehiepsi]

Thanks for adding the checks! I think no need to add tests. Pyro is based on PyTorch, not numpy, so we would expect all inputs are torch tensors.

Could you fix the lint issue? You can run make lint, make format to catch and fix the issues.

[fritzo]

Run make format and push

[GautamV234]

Run make format and push

Thanks for the help!

[fehiepsi]

Thanks, @GautamV234 !