Rerunning an Experiment
Rerunning an experiment means running the same procedure again and comparing the output.
With a fixed seed, the controlled random choices repeat.
Without a fixed seed, the random choices can change.
Compare fixed and unfixed runs
Fixed seed versus no fixed seed
Ready to run.
The fixed rows match. The unfixed rows usually do not.
What can still vary?
A seed controls the generator's starting state. It does not make every possible environment identical. Larger experiments may also depend on:
- library versions;
- hardware;
- parallel execution;
- data files;
- preprocessing choices.
For this Python subject, the main habit is simpler: record the seed and configuration values before comparing outputs.
With a fixed seed and the same code, what should happen to the generated random sequence?
Select one choice, then check.
Hint
The same seed returns a new generator to the same initial state, within the same controlled environment.
Solution
The generated sequence should repeat when the seed, code, call order, data, and relevant environment are the same. A seed alone does not control every source of variation in a larger experiment.