简体繁体 English

Hyperopt超时？

[英]Hyperopt timeout?

原文 2019-07-25 16:08:05 5 1 python/ hyperparameters/ hyperopt

I'm trying without success to set a timeout when calling 'fmin' from Hyperopt package. 我试图从Hyperopt程序包中调用“ fmin”时设置超时没有成功。

I would like to stop the entire process when max_evals are reached or when time passed (from the first iteration not each trial) > timeout. 我想在达到max_evals或经过时间（从第一次迭代开始而不是每个试用版开始）>超时时停止整个过程。

Anyone had the same problem and has somehow found a solution? 任何人都有相同的问题，并以某种方式找到了解决方案？

Many thanks! 非常感谢！

1 个解决方案

Assuming each evaluation is not too long, then you can run hyperopt in a loop doing one evaluation at a time. 假设每次评估都不太长，那么您可以循环运行hyperopt一次进行一次评估。 Each time you start an evaluation, pass fmin() the previous trials. 每次您开始评估时，都要通过fmin（）进行先前的试验。 For documentation, see issue 267 . 有关文档，请参见问题267 。

I do something similar, though a problem I noticed is I am not getting the results I expect. 我做了类似的事情，尽管我注意到的一个问题是我没有得到我期望的结果。 It seems to be doing a random search because there is no correlation between the iteration and the loss function. 似乎正在执行随机搜索，因为迭代和损失函数之间没有关联。 In other words, it doesn't seem to be getting smarter. 换句话说，它似乎并没有变得越来越聪明。 Maybe that's just me. 也许就是我。