英文:
How can I set the seed that MLFlow uses to randomly generate run names?
问题
我正在为Hugging Face transformers设置随机数生成器种子，使用transformers.set_seed()。我发现该函数也会为MLFlow设置种子，导致我始终得到相同的MLFlow运行和嵌套运行名称序列，这对我来说不理想。
在设置了Hugging Face transformers的种子后，我希望能够将MLFlow的种子重置为不同的值（或随机值）。
我如何设置MLFlow用于随机生成运行名称的种子？
英文:
I am currently setting the random number generator seed for Hugging Face transformers with transformers.set_seed() . I found that function also sets the seed for MLFlow and, as a consequence, I always get the same sequence of run and nested run names from MLFlow, which is to me undesirable.
After setting the Hugging Face transformers seed, I would like to be able to reset the seed for MLFlow to a different value (or a random value).
How can I set the seed that MLFlow uses to randomly generate run names?
答案1
得分: 0
MLFLow使用Python模块random来随机生成运行名称，请查看name_utils.py中的_generate_string()的实现。因此，设置该种子的方式是只需调用random.seed()。
transformers的set_seed()实现设置了各种随机种子，包括random.seed()。
我认为为了从MLFlow获得随机运行名称，要么我不调用set_seed()，而是手动设置Numpy和Pytorch的种子，要么在调用set_seed()后自己调用random.seed()。
英文:
MLFLow uses python module random to randomly generate a run name, see the implementation of _generate_string() in name_utils.py. Therefore the way to set the seed for that would be to just call random.seed().
The implementation of the transformers set_seed() sets a variety of random seeds, including random.seed().
It seems to me that in order to have random run names from MLFlow, either I don't call set_seed(), but I manually set the seeds for Numpy and Pytorch, or I call random.seed() myself after calling set_seed().
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库，让每个人都能够通过互相帮助和分享经验来进步。
评论