Global seed foundation #332
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
RichSweetWafer
commented
Sep 17, 2025
| { | ||
| if (dropout_rate >= type(1)) | ||
| { | ||
| throw runtime_error("Dropout rate should be in range [0, 1). Got: " + std::to_string(dropout_rate)); |
There was a problem hiding this comment.
Also, there probably should be a runtime error if dropout function uses rate that is too high
|
Come to think of it, maybe the global RNG should be seeded randomly by default to write less code. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Hi!
I've recently encountered a problem during training process of our model. The problem itself is not so important (division by zero during back propagation in cross_entropy_error.cpp), because I can see that
devbranch has changed a lot since the last release version. But I see that what I lacked during debugging has not been addressed yet: reproducibility. Randomization is good, but controlled randomization is even better. Frameworks like PyTorch try to provide ways to reduce the number of nondeterministic behavior. One way is to implement a global RNG, which is what was done in this PR.The user just has to write something like
before using the library.
This is a somewhat "foundational" code, some basic stuff to improve upon in the long run. Now, I don't know the codebase as well as you guys do, so I expect some notes being made and mistakes highlighted, if you will consider this idea.