Browse Teach Curate

Random Seeds as a Recorded Input of a Run

The sub-domain brief names seeds directly, and on rented parallel compute the problem worsens: many workers each need distinct, recorded seeds or they duplicate each other's draws. It is testable well beyond recall -- students can analyse a seeded run that is still irreproducible, and evaluate when varying the seed is the honest thing to report.

jeremy Video

Producing Reproducible Random Numbers with np.random.seed in NumPy

Pseudorandom number generators are deterministic algorithms whose output sequence is fully determined by an initial internal state called the seed; fixing that seed therefore makes an otherwise "rand…