Changing seed number - be careful

[AdimDrewnik]

When running Robyn with different seeds be careful when changing the seed value. For example if you run seed = 123 with 10 trials and later run seed 124 with 10 trials then 9 out of 10 trials will have the same seed in the both runs. Seed number must be changed by more than the number of trials if someone wants completely new RNG in the whole run.

[laresbernardo]

You're right. Not that it is an issue but you've got a point. Adding a PR to improve this logic, using this approach:

seed <- 124
set.seed(seed)
for (i in 1:5) {
  print(seed + runif(1))
}

[AdimDrewnik]

I think this implementation will cause even worse problems. As seed needs to be an integer using runif will cause all seeds to be the same exact number in many cases.

[laresbernardo]

Hi @AdimDrewnik Seeds can be any number, no need for integer values.

[AdimDrewnik]

I have just checked. Using seed e.g. 1.1 and 1.11 gives exactly same random numbers.

[laresbernardo]

I think that's not right. Please, check:

> seed <- 1
> set.seed(seed)
> for (i in 1:5) {
+   print(seed + runif(1))
+ }
[1] 1.265509
[1] 1.372124
[1] 1.572853
[1] 1.908208
[1] 1.201682
> seed <- 1.1
> set.seed(seed)
> for (i in 1:5) {
+   print(seed + runif(1))
+ }
[1] 1.365509
[1] 1.472124
[1] 1.672853
[1] 2.008208
[1] 1.301682
> seed <- 1.11
> set.seed(seed)
> for (i in 1:5) {
+   print(seed + runif(1))
+ }
[1] 1.375509
[1] 1.482124
[1] 1.682853
[1] 2.018208
[1] 1.311682

[AdimDrewnik]

I am talking about this case

set.seed(1.1)
rnorm(5)
[1] -0.6264538 0.1836433 -0.8356286 1.5952808 0.3295078

set.seed(1.11)
rnorm(5)
[1] -0.6264538 0.1836433 -0.8356286 1.5952808 0.3295078

[laresbernardo]

Alright, but that's not the implementation I've proposed in the code.

[AdimDrewnik]

But you are generating seeds that will be used for random number generation. If you generate non integer seeds that are same integer when rounded your RNG for two trials will be the same.

[laresbernardo]

I've increased the order of magnitude to have large numeric values. Does this look good to you? Feel free to test and confirm before we merge to main branch.

[AdimDrewnik]

Please revert this change and only add a warning to the documentation. It will only cause more problems and more issues. Before I only needed to increase the seed number above the number of trials. Now I am getting a chance to run same RNG multiple times.