replace np with torch linsapce in riemann approximation #714
Conversation
|
@aobo-y has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
| import torch | ||
|
|
||
| try: | ||
| import numpy as np |
There was a problem hiding this comment.
To make this more clean can we move the import into gauss_legendre_builders. Then we won't need try-except.
There was a problem hiding this comment.
But as you can see below, I also rely on this to raise an explicit error for using gausslegendre when numpy is not available and suggesting riemann. Without it, users will only receive an contextless message like Cannot find module numpy.
I know we are doing this for a special use case and it is unlikely to have any users encounter this. So I am also open to your suggestion.
There was a problem hiding this comment.
My worry is, since this code is going to be executed, it could make problems during packaging (It might try to load numpy. We have to test, I can't say exactly). If we put it in the function, the import is not going to be executed at all when we choose riemann sum. If numpy is not available an error will raise only if the user chooses gauss legendre. If you want, you can keep try-except in the gauss_legendre_builders function.
There was a problem hiding this comment.
Sure, I will move it.
|
@aobo-y has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
Reduce the dependency on numpy in approximation methods
torch.linspacein riemann methods