Such an indescribable feeling when one variant is showing great improvement relative to the baseline... and then upon more careful inspection you realize you had a bug and both runs were actually identical except for different seeds.

May 10, 2022 · 4:31 PM UTC

9
4
2
118
Replying to @gdb
And then you ask yourself (again) why you got into deep #reinforcementlearning
Replying to @gdb
If the results are great, I immediately get suspicious rather than happy
2
Replying to @gdb
I feel you.
6
Replying to @gdb
Seeds!! I have found them really important. I used to gauge my algo on its consistent behaviour for all seeds, setting it to time(). That's so funny.
1
Replying to @gdb
Can you please tell us more so thoes of us that use #replika can understand more about the tech? I'm always interested in your tweets 😊
1
Replying to @gdb
Make sure to log that seed in the "extra lucky numbers" file for use in future projects.
5
Replying to @gdb
That sudden transition from hope to despair is excruciating.
1
2