Open mchiquier opened 4 years ago
Curious what you mean, can you give an example?
If I understand you correctly, check this up.
@omry Yes! This task is to take advantage of the run output directory functionality in Hydra. We expect the base use-case for individual / small-team research is output directories grouped by experiment name, then sub-experiments (e.g. sweeps).
/outputs ----/experiment1 ------------/factor_sweep ----------------/1 ----------------/2 ----/experiment2 ------------/factor_sweep ----------------/1 ----------------/2
Gotcha. This should be doable by configuring the output directory as described in the link above.
@omry, thanks for your interest!
This feature would simply automatize modifying the hydra.yaml config from the command line. It is a simple change, but right now you have to manually modify hydra.yaml. As the link mentions, "You can include some Hydra config snippet in your own config to override it directly, or compose in different configurations provided by plugins or by your own code". Our feature does just that : code that allows the user specify what variable to group by as a variable in their own config file from the command line as they run a sweep. For example, the user could change from Matt's choice of grouping by "experiment" to grouping by "myconfig.dataset" using the command line: --group_by myconfig.dataset, and this would update the hydra.yaml file automatically.
Where are the names experiment1 and factor_sweep coming from? Are they dynamic?
Hydra is extremely flexible. I don't have a full understanding of what you are trying to do but I suspect you are doing it wrong if you need another tool to generate the config for Hydra.
Could you please point me to the tool that allows you to modify the config for Hydra non-manually?
We are not claiming that it is not extremely flexible, in fact we are using it because we believe it is. We are not adding any new functionality, just making it easier to configure on-the-fly.
Hydra generates the configuration object dynamically. There are multiple ways you can manipulate the config with Hydra, from command line overrides to composing your configuration from multiple config group options. I suggest that you go through the tutorial in Hydra to get a better understanding of how to use it.
Can you give a simple example of what you are trying to do that is not possible by using Hydra directly? Either I will tell you how to do it or I will learn about an unsupported use case.
Mia, it might be easier if you join the Hydra chat.
Will do, thanks.
Changing it so that it names it minimally according to date, but mostly includes the top 5 characteristic configs of the experiment