Fix inv_transform

[eytan]

Motivation

inv_transform was crapping out when drawing a large number of samples because the base sample was exactly 1.0( see issue #161). While we were previously subtracting off a small number (1e-10), this number wasn't sufficient. I found that (1-e6) was the smallest base 10 number we could subtract off to get rid of the -inf.

Have you read the Contributing Guidelines on pull requests?

Yes.

Test Plan

import torch
from botorch.qmc.normal import NormalQMCEngine
normal_qmc_engine = NormalQMCEngine(d=2, seed=343, inv_transform=True)
samples = normal_qmc_engine.draw(10000, dtype=torch.float)
print(samples.max())

previously returned tensor(inf), but now it returns tensor(4.9010).

Related PRs

(If this PR adds or changes functionality, please take some time to update the docs at https://github.com/pytorch/botorch, and link to your PR here.)

[facebook-github-bot]

@sdaulton has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

@eytan has updated the pull request. Re-import the pull request

[facebook-github-bot]

@sdaulton has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

@eytan merged this pull request in ba370e8.