Explain random seeds in episode 2 #521
Conversation
🆗 Pre-flight checks passed 😃This pull request has been checked and contains no modified workflow files or spoofing. Results of any additional workflows will appear here when they are done. |
carschno
left a comment
carschno
left a comment
There was a problem hiding this comment.
I've had some minor comments.
| Please note that even though you have selected a random seed, this seed is used to generate a | ||
| **different** random number every time you execute a Jupyter cell. | ||
| So, to get truly replicable deep learning pipelines you need to run the notebook from start to end in one go. | ||
| ::: |
There was a problem hiding this comment.
I have seen some confusion about this in the past, so perhaps we can make it even more explicit:
| ::: | |
| When the same code is run on another machine, the random number and hence the results can also differ. | |
| ::: |
There was a problem hiding this comment.
@carschno I think this is actually not true. Or at least not the promise of using random seeds right? The whole idea of using random seeds is that the code becomes deterministic and the results are the same on another machine. Of course you should always double check that.
I agree that often students have a different result, but that is probably due to running cells multiple times or different dependency versions.
There was a problem hiding this comment.
As discussed with @carschno we decided to leave it like was proposed in this PR.
Co-authored-by: Carsten Schnober <[email protected]>
Fixes #489