Knowledge MaxxingRandom seed
A random seed is a starting value that makes a process involving randomness, such as data shuffling or weight initialization, produce the exact same sequence every time it runs.
Prerequisite: Control variable
What matters
Fixing the seed makes an experiment reproducible, letting you rerun the exact same training process later and get identical results, which is essential for fair comparisons.
Example
Two training runs using the same random seed will shuffle the data identically and initialize weights identically, isolating any other change as the true source of a performance difference.
Common misconception
A fixed random seed does not make a model's real-world behavior deterministic on new data; it only makes the training process itself reproducible.
Recall question
Why does fixing a random seed help make experiment comparisons fair?