DEV Community

Discussion on: Why You Should Not Trust the train_test_split() Function

Collapse
 
mccurcio profile image
Matt Curcio

Someone I know who worked for a Biotech company told me an interesting story.

This person claimed their company did many double-blind human research clinical trials. This company used a computer to randomly choose numbers (keys word here) for their trials.

Apparently, they realized that certain numbers were popping up again and again.
Then they realized that the computer choosing their random numbers had used the same seed for a little while.

They implemented a new random seed number for all clinical trials to be chosen every time for new work. They wrote an SOP so that the new random number seeds would be now based on the time from in seconds since 1970.
Actually, it was the last x numbers, I think. lol