Numpy Import Convention

[RatishT]

Dear Pyro contributors,

I've taken the initiative to address and resolve several linting warnings pointed out by a pylint plugin called dslinter. The code suggestions that it gave were:

1. torch.log() Variable Clamping:
   • Enclosed the variable in the torch.log() operation with torch.clamp() to establish a minimum value. This modification ensures a safeguard against potential NaN errors, particularly when the input variable is exceptionally small.
   • Before:

     result = torch.log(input_var)

   • After:

     result = torch.log(torch.clamp(input_var, min=1e-10))

2. Numpy Import Convention:
   • I added "as np" to a numpy import, matching standard convention and consistency in code

I invite your valuable insights and feedback on these modifications to guarantee that they align seamlessly with our shared coding standards and project requirements.

[martinjankowiak]

thanks but i don't think the log change is necessary. in any case if we were to do something like that we would use torch.finfo and not some ad hoc number like 1e-10.

the np change looks fine to me

[RatishT]

thanks but i don't think the log change is necessary. in any case if we were to do something like that we would use torch.finfo and not some ad hoc number like 1e-10.

the np change looks fine to me

Thank you for your reply! I have reverted the torch.clamp in the torch.log calls
For future reference, would it be better to use variable.clamp(min=torch.finfo(variable.dtype).tiny)?