Reproducibiity bug due to global seeding
jeandut opened this issue · 1 comments
jeandut commented
When running fed_benchmark from scratch or fed_benchmark.py multiple times (with or without the --strategy) one does not retrieve exactly the same results as the seeding is done globally and not at the operation level, the order in which the operations occur affect the results..
Seeding should be done at the run level.
jeandut commented
Fixed