Feat set seed #929
Feat set seed #929
Conversation
Draft fuinction that worked on my initial testing using OpSo 0.9.1, but still needs testing on 0.10.1
opensoundscape/seed.py
Outdated
There was a problem hiding this comment.
I think this can go in opensoundscape/utils.py rather than being a separate module
There was a problem hiding this comment.
Done, but Let the unit test separate because I didn't want to mess-up test_utils.py.
tests/test_seed.py
Outdated
|
|
||
| import pytest | ||
|
|
||
| pytestmark = pytest.mark.parametrize("input", [1, 11, 13, 42, 59, 666, 1234]) |
There was a problem hiding this comment.
This is a global version of the @pytest.mark.parametrize decorator to use the list as inputs for all functions in the script. See https://docs.pytest.org/en/7.3.x/how-to/parametrize.html for more explanation.
opensoundscape/seed.py
Outdated
| import random | ||
|
|
||
|
|
||
| def seed(seed, verbose=True): |
There was a problem hiding this comment.
needs a docstring, with Args and Return fields; see other functions for examples
opensoundscape/utils.py
Outdated
| @@ -329,3 +330,25 @@ def generate_opacity_colormaps( | |||
| colormaps.append(cmap) | |||
|
|
|||
| return colormaps | |||
|
|
|||
|
|
|||
| import numpy as np | |||
There was a problem hiding this comment.
imports should always be at the top of a file. VS Code will help detect duplicate imports
There was a problem hiding this comment.
Oh, sorry, missed that. Just pushed the correction.
| model_resnet3 = cnn_architectures.resnet18(num_classes=10, weights=None) | ||
| lw3 = model_resnet3.layer1[0].conv1.weight | ||
|
|
||
| assert torch.all(lw1 == lw2) & torch.any(lw1 != lw3) |
There was a problem hiding this comment.
don't understand the logic here, can you add a comment? is the any asserting that none of the weights should be the same? That doesn't seem like the correct logic, rather we want to assert that the entire array isn't equal to the other?
There was a problem hiding this comment.
In summary, also assert that at least one element is different if different seeds. Added as a comment in tests/test_utils.py.
Feature to set random state for reproducible training. Solving #576